The Influence of Built Environment on Travel Carbon Emissions in Old Communities: A Case Study of Chengdu

Abstract

Old communities are the foundational units for low-carbon transformation in the background of high-quality urban development and dual carbon goals. However, existing research prioritizes building energy-efficient technologies and macro-level spatial indicators, with limited attention to how community-scale built environments specifically influence residents’ behaviors. This study takes five old communities in Chengdu as its subject and quantitatively measure residents’ perceptions of their built environment. Using multiple regression and subgroup regression analyses, it systematically identifies key built environment factors in old communities that influence low-carbon travel behavior. The results show that: (1) Diversity, accessibility, street connectivity, and aesthetics consistently demonstrated significant negative effects across demographic groups; (2) As people age, the carbon emissions from their travel to tend to decrease. The impact intensity of street connectivity on low-carbon travel varies significantly among different age groups; (3) Compared with woman, men overall have higher travel carbon emissions. All findings indicate that complete spatial functions, clear road networks, and accessible facilities promote low-carbon travel. This offers key insights for upgrading built environments in old communities.

1. Introduction

The relationship between climate and carbon emissions is the core of current global environmental issues [1,2,3]. The increase in carbon emissions is the main driver of the imbalance in the global climate system [4,5]. Among various sources of carbon emissions, transportation accounts for a significant proportion [6,7,8,9,10,11]. Numerous studies have established that residents’ choice of travel mode is related to the built environment of their residential areas [10,12,13,14,15]. Therefore, optimizing the built environment of residential areas is considered effective in changing residents’ travel modes and habits, and has great potential to reduce carbon emissions [16,17].

The built environment, as a key physical and spatial medium shaping residents’ daily travel behavior, has been the focus of significant paradigmatic shifts over the past half-century. Early studies (1970s–1990s) were primarily grounded in neoclassical economics, emphasizing the role of policy instruments in influencing travel costs, for example, curbing automobile growth by fuel taxation [18] and fuel price fluctuations [19]. Newman and Kenworthy’s groundbreaking work established, for the first time on a global scale, an inverse relationship between urban density and transportation energy consumption, providing key empirical support for compact city development [20]. However, their macro-level policy perspective offers limited guidance for micro-scale spatial optimization within communities—particularly in old communities with established built environments and limited payment ability.

From the 1990s to the 2010s, with the rise of New Urbanism, research shifted toward the built environment. A core breakthrough was the evolution from a singular focus on “density” to more comprehensive theoretical frameworks. Initially, Cervero and Kockelman proposed the classic “3D” framework—Density, Diversity, and Design [21]. Ewing and Cervero’s meta-analysis further extended the framework to “5D” with Destination Accessibility and Distance to Transit [22]. Handy et al. emphasized the role of street connectivity and pedestrian environments—key aspects of design—in promoting non-motorized travel [23]. Krizek found that a high level of “5D” integration, particularly in transit-oriented developments, significantly reduces car dependence [24]. On the basis of previous studies, Ewing and Cervero further expanded the elements affecting residents’ travel characteristics and proposed the “7D” framework, adding “Demand Management” and “Demographics” [22]. Yet it must be noted that the empirical basis for these frameworks largely derives from Western urban forms and newly developed communities. Their generalizability and the relative importance of their components require localized validation when applied to old communities in Chinese high-density cities—contexts characterized by unique socio-spatial structures, aging infrastructure, and often high functional diversity.

A growing body of literature post-2010 has significantly refined our understanding of the built environment-travel nexus, converging on three critical frontiers. First, the mediating role of subjective perceptionsis increasingly recognized as pivotal, often outweighing objective metrics in influencing travel choices [25,26]. Second, research has progressively linked environmental attributes directly to travel carbon emissions, shifting the focus from mode share to climate impact [27,28]. Third, substantial emphasis is placed on social heterogeneity, analyzing how effects diverge across age, gender, and socioeconomic status to inform equitable planning [29,30]. However, there is a lack of research that integrates all three pivotal elements within a unified analytical framework. Existing studies often investigate these dimensions—perception, carbon outcome, and social equity—in isolation or pairwise combinations.

This study is designed to bridge this multidimensional gap: (1) targeting the unique context of Chinese high-density old communities; (2) employing individual travel carbon emissions as the direct outcome variable; and (3) conducting systematic subgroup analysis to unpack effect heterogeneity. This study discusses two main research questions:

(1)

In old communities in China, which key elements of the built environment are significantly effective in reducing residents’ travel carbon emissions? What are their degrees and directions of impact?

(2)

Do the effects of the same built environment variables vary across groups, and what are the implications for designing old communities?

The structure of this article is as follows. Section 2 introduces the methods and data. In Section 3, we analyze the data. Section 4 conducts regression analysis and presents the results. Section 5 discusses the results. Finally, Section 6 summarizes the conclusions.

2. Methods and Materials

2.1. Study Design

This study analyzes the mechanism through which key built environment characteristics influence residents’ low-carbon travel behavior. The research was divided into 3 steps: (1) Questionnaire Design: The questionnaire was reconstructed for local adaptation based on the internationally widely used NEWS-CFA (Neighborhood Environment Walkability Scale—Confirmatory Factor Analysis version), covering four modules: socioeconomic characteristics, built environment perception, low-carbon awareness, and travel behavior [31]. (2) Data Processing and Questionnaire Adjustment: Questionnaire collection (questionnaires distributed n = 550, questionnaires returned n = 507), reliability, validity, and exploratory factor analysis were first conducted to ensure the adaptability of the scale in the context of old communities in China. (3) Multiple Linear Regression: After correlation tests, a multiple linear regression model was constructed to systematically evaluate the strength and direction of built environment elements on low-carbon travel. Considering the differences in gender and age, a grouped regression analysis was further conducted to reveal the age and gender heterogeneity mechanism of low-carbon travel behavior in old communities.

2.2. Study Area and Data Collection

Old communities are early-built (mostly pre-2000) urban residential complexes that form functionally and spatially continuous built-up areas [32,33]. They suffer from poor maintenance, a lack of functional facilities, and inadequate management.

Chengdu is the capital city of Sichuan Province, China. Like most Chinese cities, its development has expanded outward gradually based on the old city. This has led to the long-term retention of old communities in the core district. Besides, Chengdu is a plain city with flat terrain and a well-developed public transportation system, which provides residents with more diverse travel choices. The study selected five old communities in Chengdu’s core district, which are required to take the lead in functional reconstruction and quality improvement. The construction dates of these five communities span from the 1950s to the 1990s (

Table 1. Sample-related Information.

Name	Year	District	Households	Site Area	Number of Floors	Building Type
The staff dormitory of Gonghe Village	1972	Wuhou	Approximately 500	Approximately 0.133 km2	6 stories	Staff Dormitory
Railway New Village	1952	Jinniu	Approximately 900	Approximately 0.282 km2	3 stories	Staff Dormitory
Shuangting Home	1967	Qingyang	Approximately 750	Approximately 0.167 km2	6 stories	Resettlement Housing
No. 6, Lane 1, North Dongfeng Road	1982	Chenghua	Approximately 850	Approximately 0.200 km2	6 stories	Resettlement Housing
No. 28 Daci Temple Road	1998	Jinjiang	Approximately 200	Approximately 0.067 km2	6 stories	Work-unit Allocated Housing

, Figure 1).

The questionnaire consisted of four modules: socioeconomic characteristics, environmental perception, low-carbon cognition, and carbon emission calculation.

Socioeconomic characteristics are widely regarded as one of the most important factors affecting residents’ travel behavior. To quantitatively analyze the relationship between these socioeconomic factors and travel behavior, variables such as gender, age, education level, personal and family annual income, possession of a driver’s license, and family-owned means of transportation were specifically included in the questionnaire design.

Built environment perception is measured by a questionnaire, for it is difficult to quantify using traditional big data measurement methods. Considering the length of the questionnaire and the difficulties in completing the questionnaire, while hoping to retain key observations on walking and cycling environments, this study finally selected a partial coefficient version of the NEWS-CFA to measure six elements: residential density, land-use mix—diversity (diversity), land-use mix—access (accessibility), street connectivity, infrastructure and safety for walking, and aesthetics.

Low-carbon awareness is a further relevant variable. It includes low-carbon attitude, low-carbon travel intention and policy advocacy.

Carbon emission calculation used the Emission Factor Method to estimate residents’ travel carbon emissions (Formula (1)). This method is the mainstream approach in transportation carbon emission accounting. Based on the carbon emission coefficient (kg CO₂/km) of different transportation modes, combined with the travel distance and frequency self-reported by respondents, the total weekly travel carbon emissions per person per kilometer were estimated. The formula is as follows:

$$\mathrm{C}={\displaystyle \frac{{\sum }_{\mathrm{i}=1}^6\left({\mathrm{V}}_{\mathrm{i}}\cdot {\mathrm{T}}_{\mathrm{i}}\cdot \mathrm{E}{\mathrm{F}}_{\mathrm{i}}\right)}{{\sum }_{\mathrm{i}=1}^6{\mathrm{V}}_{\mathrm{i}}\cdot {\mathrm{T}}_{\mathrm{i}}}}$$

(1)

where, C = total carbon emissions; i = travel behavior code (6 types in total: 1 = walking, 2 = bicycling, 3 = electric bike, 4 = bus, 5 = subway, 6 = car); ${\mathrm{V}}_{\mathrm{i}}$ (km/h) = average speed of the i-th travel behavior; ${\mathrm{T}}_{\mathrm{i}}$ (h) = average time of the i-th travel behavior; $\mathrm{E}{\mathrm{F}}_{\mathrm{i}}$ (kgCO₂/km) = unit carbon emission factor of the i-th travel behavior; ${\mathrm{V}}_{\mathrm{i}}\cdot {\mathrm{T}}_{\mathrm{i}}$ (km) = average travel distance.

2.3. Research Data

2.3.1. Reliability and Validity Verification

The scale used in this study is based on the NEWS-CFA scale. To adapt to the actual situation of old communities in Chinese cities, a preliminary adaptability test was conducted after translation and content adjustment. Through reliability analysis and KMO/Bartlett’s sphericity test on the six built environment dimensions in the questionnaire, it was found that the scale has high internal consistency and structural adaptability. Except for individual modules, the Cronbach’s α coefficient of each dimension is above 0.75. Among them, the reliability indicators of diversity (Q9) and street connectivity (Q11) exceed 0.94, showing extremely strong stability. All KMO values exceed 0.65, and Bartlett’s test is significant (p < 0.001). This indicates that the correlation between these items is suitable for subsequent factor analysis and has a good statistical basis (Appendix A.1, Appendix A.2, Appendix A.3 and Appendix A.4).

2.3.2. Exploratory Factor Analysis of Built Environment Elements

In the pre-test of comprehensive factor analysis conducted after merging the items of the six modules from Q8 to Q13, the KMO test result showed that the overall KMO value was 0.924, which met the standard of “very suitable for factor analysis”. The approximate chi-square value of Bartlett’s sphericity test was 271.701, and the p-value was less than 0.001. This indicates that the correlation between variables is statistically significant. It can be considered that there is a strong correlation structure between the current scale items, which is suitable for further exploratory factor analysis. The above results initially confirm the rationality of the scale as a measurement tool from a statistical perspective (Appendix A.5).

In the variance explanation results, the first seven principal components extracted before rotation all had eigenvalues greater than 1, with a cumulative explanation rate of 51.756%. After rotation, the cumulative explanation rate was 51.724%. Although it has not reached the ideal threshold of 60%, considering that the scale covers a wide range, has a large number of items, and involves multiple different types of built environment perception dimensions, the current explanatory power is still within an acceptable range. In the scree plot, it can be observed that after the seventh principal component, the curve tends to be flat, which conforms to the inflection point principle of factor extraction. It is relatively safe for this study to conduct factor naming and interpretation based on the seven-factor structure in subsequent analyses.

From the perspective of rotated factor loadings, each principal component has a clear thematic attribution. For example, Factor 1 mainly reflects the items of the Q9 diversity module, with most factor loadings above 0.65, showing a clear and stable structure. Factor 2 well aggregates the street connectivity items in Q11, such as sidewalk configuration, lighting, and traffic facilities. The overall factor loading distribution is reasonable. Except for individual items (e.g., Q11_6) with low explanatory power, other items have good attribution. Factor 3 clearly corresponds to aesthetics (Q12), and Factor 4 focuses on residential density (Q8), both of which have strong structural consistency. It is worth noting that almost all four items of Q10 (accessibility) are clustered on Factor 6, with factor loadings greater than 0.73, showing strong consistency. Q13 (infrastructure and safety for walking) is highly concentrated on Factor 5, especially the items Q13_2, Q13_3, and Q13_5, which have significant loadings, which provide support for the independence of this dimension.

2.3.3. Exploratory Factor Analysis of Carbon Awareness

For the three modules of low-carbon attitude (Q14), low-carbon travel intention (Q15), and policy advocacy (Q16), this study conducted an adaptability test before factor analysis after merging them. The KMO test result showed that the KMO value of this module was 0.661. Although it is slightly higher than the minimum standard of 0.6, it is still at the “weakly suitable” level. The approximate chi-square of Bartlett’s sphericity test was 546.456, with a p-value of 0.000, reaching the 1% significance level. This indicates that there is a significant correlation between items and still a certain basis for factor analysis. The low KMO value also reflects that the correlation between some items in the current module is not close enough, or there is a problem of unclear dimension division, which provides a warning for subsequent structural optimization (Appendix A.5).

In terms of factor extraction, based on the criterion of eigenvalue greater than 1, three principal components were finally extracted. The cumulative explained variance rotation was 67.787%, which has reached an acceptable level in social science research. Further checking the scree plot, it can be observed that the slope tends to be flat after the third principal component, which verifies the rationality of the three-factor structure. The rotated factor loading coefficient table shows that this three-factor structure has clear item attribution and interpretation directions. The first factor aggregates three items in Q15 (Q15_1–Q15_3), all of which are related to self-efficacy and behavioral control. Their factor loadings are all above 0.76, indicating that this factor can be interpreted as “self-cognition and sense of control over low-carbon travel”. The second factor aggregates Q14 items (Q14_1–Q14_3), with loadings concentrated around 0.81, forming a dimension of “low-carbon travel attitude tendency”. The third factor only includes Q16_1, with a factor loading of 0.995, almost forming a component independently. Its communality is 0.99, showing that it is too independent in the existing structure.

From the perspective of structural rationality, although Q16_1 “I am willing to recommend low-carbon travel to others” is related to low-carbon travel in content, its expression is more inclined to social advocacy or communication behavior. It differs in psychological dimensions from the “self-evaluation” and “subjective feelings” emphasized in Q14 and Q15. Therefore, it forms a factor independently in the factor structure, with statistical independence. Although its communality is extremely high, it also reveals that this item may have a problem of “construct drift” in the current module merging structure. Therefore, the items Q14–Q16 are not merged here.

2.4. Questionnaire Adjustment

This study considered that the community type in the original context of the NEWS-CFA scale is mainly suburban low-density residential areas. Some items have problems, such as insufficient cultural adaptability and unclear practical orientation in the high-density built environment of first-tier cities in China. Based on the preliminary reliability and validity analysis and factor loading evaluation, items with weak statistical performance or inconsistent semantics were eliminated to improve the overall structural validity and explanatory power of the questionnaire.

• Inconsistency with Life Scenarios: Q8_1 “How common are detached single-family houses around your home?” is a typical low-density residential form in the original scale. However, it has almost no practical basis in the old communities in the central urban area of Chengdu. In Chengdu, the housing types in old communities are mainly brick-concrete structures with 4–6 floors, and detached houses are extremely rare. Similarly, Q9_7 (post office), Q9_10 (other schools), and Q9_16 (audio-visual store) in the Q9 module also showed low factor loadings and communality. This reflects that the frequency of use of these facilities in residents’ daily lives is low, or they have lost their representativeness. In particular, the audio-visual store is clearly no longer recognized by most residents.
• Insufficient Semantic Adaptability: The item Q11_6 “It is safe to ride a bicycle near my home” in the original Q11 module had the lowest factor loading (0.516) and a communality of only 0.266. During the survey, residents also frequently reported that they “did not know how to judge”.
• Insufficient Label Distinction: Q13_3 and Q13_4 in the Q13 infrastructure and safety for walking dimension are reverse items. If they are not clearly marked or distinguished by typesetting, it is easy to cause misunderstanding among respondents during on-site questionnaire distribution, affecting the stability and interpretability of answers. In subsequent structural optimization, the accuracy of expression and item reconstruction of this module was focused on.

Through the above item deletion and modification, this study retained a more representative and applicable questionnaire to the residential environment characteristics of old communities.

2.5. Data Processing

Due to the design of the questions, Q10–Q13 are different from other scale questions, with only 1–4 points (four options: strongly disagree, disagree, agree, strongly agree, scored 1–4). Based on factors such as the overall distribution variance of the data, a linear mapping adjustment was performed on them.

To test the applicability of the multiple linear regression model, this study first conducted graphical diagnosis on the distribution characteristics of residuals, and sequentially drew P-P plots, Q-Q plots, and residual-fitted value scatter plots. In the Q-Q plot drawn after the initial modeling, it can be observed that the residual points do not fit well along the diagonal, but show systematic deviations in the tail and middle sections. This suggests that the model residuals do not follow a normal distribution. The corresponding P-P plot also shows a certain degree of deviation, and the cumulative distribution curve does not closely adhere to the 45-degree line. This indicates that the phenomenon of middle value bias is relatively obvious. In addition, the residual-fitted value plot shows a “funnel-shaped” distribution, indicating that the variance of residuals is heterogeneous during the change in fitted values. This suggests that the model may have a problem of heteroscedasticity.

To improve the data distribution and enhance the model stability, this study introduced the Yeo-Johnson transformation to transform the dependent variable. Different from the Box-Cox transformation, the Yeo-Johnson method allows input of data containing zero and negative values. Therefore, it is more suitable for non-negative skewed distribution variables, such as travel carbon emissions collected in practice. After the transformation, the residuals were redistributed and modeled, and then the collinearity problem was checked.

The P-P plot and Q-Q plot, redrawn after adjustment, both showed relatively ideal results. The residual points in the Q-Q plot are more closely distributed along the diagonal, indicating that their distribution tends to be normal. The deviation between the theoretical values and observed values in the P-P plot is significantly reduced, indicating that the cumulative distribution curve is close to the normal distribution. The residual points in the residual-fitted value plot are randomly distributed without obvious structural patterns, and the heteroscedasticity problem is also effectively alleviated. After diagnosis and adjustment, the regression model constructed in this study basically meets the basic assumptions required by linear regression analysis, and both the model’s robustness and explanatory power are improved.

3. Data Analysis

3.1. Sample Characteristic Analysis

In this study, the age of respondents was mainly concentrated in 41–60 years old, reflecting that middle-aged people are the main residents of old communities. 55.6% of the respondents were female. High school or technical secondary school education was the main group, accounting for 35.8%. Followed by junior college and junior high school or below, accounting for 24.9% and 18.1%, respectively. Undergraduate and postgraduate education accounted for only 21.2%. This indicates that old communities are mostly populated by groups with medium and below education levels, and the proportion of highly educated residents is limited. Individual and family annual incomes are mainly medium and low. The proportion of individuals with an annual income of 150,000 yuan or below accounted for more than 74%, and the proportion of families with an annual income of 150,000 yuan or below was as high as 71.6%. This shows that not only is the personal income of old communities low, but also the overall economic situation of families is not particularly well-off. 57.2% of residents do not hold a driver’s license, which is closely related to the geographical location, road conditions, and residents’ economic capacity of old communities (

Table 2. Sample Characteristic Analysis.

Category	Option	Frequency	Percentage (%)
Gender	Male	225	42.694
	Female	282	53.510
Age	18–30 years old	44	8.349
	31–40 years old	124	23.529
	41–50 years old	122	23.150
	51–60 years old	145	27.514
	Over 60 years old	72	13.662
Education Level	Junior high school or below	149	28.273
	High school/technical secondary school	132	25.047
	Junior college	115	21.822
	Undergraduate	80	15.180
	Postgraduate	31	5.882
Personal Annual Income	Below 50,000 yuan	106	20.114
	50,000–100,000 yuan	152	28.843
	110,000–150,000 yuan	132	25.047
	160,000–200,000 yuan	56	10.626
	210,000–250,000 yuan	41	7.780
	Above 250,000 yuan	20	3.795
Family Annual Income	Below 50,000 yuan	101	19.165
	50,000–100,000 yuan	142	26.945
	110,000–150,000 yuan	122	23.150
	160,000–200,000 yuan	81	15.370
	210,000–250,000 yuan	41	7.780
	Above 250,000 yuan	20	3.795
Driver’s License Possession	Yes	289	54.839
	No	218	41.366

).

3.2. Descriptive Analysis

There were no missing values in the data for residential density, diversity, accessibility, street connectivity, aesthetics, infrastructure and safety for walking, low-carbon attitude, low-carbon intention, policy advocacy, and travel carbon emissions. After removing outliers, the data quality was high (

Table 3. Descriptive Analysis.

Category	Weight Removal	Maximum Value	Minimum Value	Median	Mean	Coefficient of Variation	Standard Deviation	Variance	S-W Normality Test
Residential density	36	4.7	1.2	3	3.028	0.195	0.592	0.35	Satisfied (p = 0.109)
Land-use mix—diversity (Diversity)	36	5	1	3.1	3.04	0.201	0.612	0.374	Satisfied (p = 0.018)
Land-use mix—access (Accessibility)	35	4.9	1.1	3	3.018	0.209	0.63	0.396	Satisfied (p = 0.239)
Street connectivity	33	4.6	1.1	3	2.988	0.202	0.602	0.363	Satisfied (p = 0.040)
Aesthetics	33	5	1.3	3	3.019	0.194	0.586	0.343	Satisfied (p = 0.062)
Infrastructure and safety for walking	36	4.9	1.2	3.1	3.07	0.205	0.628	0.394	Satisfied (p = 0.069)
Low-Carbon Attitude	37	4.8	1	3	2.994	0.209	0.626	0.392	Satisfied (p = 0.084)
Low-Carbon intention	33	4.6	1.1	3.1	3.079	0.2	0.616	0.38	Satisfied (p = 0.088)
Policy Advocacy	5	5	1	3	2.99	0.376	1.125	1.265	Satisfied (p = 0.000)
Travel Carbon Emissions	507	0.961	0.005	0.522	0.574	0.219	0.003	0	Satisfied (p = 0.096)

).

3.3. Correlation Analysis

This study employed Pearson correlation analysis to systematically examine the relationships between travel carbon emissions and variables including residents’ perception of the built environment, psychological attitudes, and policy advocacy. The results showed that most built environment elements have a significant correlation with travel carbon emissions (Figure 2). The specific results are as follows:

Street connectivity (r = −0.441, p < 0.001) and aesthetics (r = −0.408, p < 0.001) have the most significant impact on travel carbon emissions, and both show a negative correlation. Accessibility (r = −0.35, p < 0.001) and diversity (r = −0.36, p < 0.001) also have a significant negative correlation with travel carbon emissions. Infrastructure and safety for walking have a weak negative correlation with travel carbon emissions (r = −0.113, p < 0.05). Residential density has a significant positive correlation with travel carbon emissions (r = 0.073, p < 0.1). In terms of psychological variables, low-carbon attitude has a weak positive correlation with carbon emissions (r = 0.068, p = 0.126), while low-carbon intention (r = 0.024, p = 0.587) and policy advocacy (r = −0.058, p = 0.194) do not show a significant impact. Travel carbon emissions have a significant negative correlation with multiple built environment elements, especially the effects of connectivity, aesthetics, and land attributes, which are more prominent, while the effect of psychological elements is not obvious.

4. Results

4.1. Built Environment

First, a multiple linear regression model was used to test the impact of built environment, psychological factors, and socioeconomic variables on participants’ travel carbon emissions.

Table 4. Multiple Linear Regression Table (n = 507).

Category	Unstandardized Coefficient		Standardized Coefficient	t	p	Collinearity Diagnosis
	B	S.E.	Beta			VIF	Tolerance
Constant	1.366	0.024	-	56.141	0.000 ***	-	-
Residential density	0.003	0.004	0.011	0.677	0.499	1.035	0.966
Diversity	−0.081	0.003	−0.374	−23.294	0.000 ***	1.060	0.943
Accessibility	−0.090	0.004	−0.377	−23.733	0.000 ***	1.038	0.964
Street connectivity	−0.098	0.004	−0.398	−24.579	0.000 ***	1.077	0.928
Aesthetics	−0.084	0.003	−0.394	−24.602	0.000 ***	1.057	0.946
Infrastructure and safety for walking	−0.006	0.004	−0.026	−1.615	0.107	1.065	0.939
Gender (Male)	0.113	0.005	0.327	20.535	0.000 ***	1.047	0.955
31–40 years old	−0.070	0.011	−0.175	−6.449	0.000 ***	3.030	0.330
41–50 years old	−0.130	0.011	−0.324	−11.765	0.000 ***	3.123	0.320
51–60 years old	−0.180	0.011	−0.476	−16.849	0.000 ***	3.284	0.304
Over 60 years old	−0.245	0.012	−0.500	−20.643	0.000 ***	2.413	0.414
High school/technical secondary school	0.008	0.007	0.021	1.118	0.264	1.491	0.671
Junior college	−0.001	0.008	−0.003	−0.177	0.860	1.401	0.714
Undergraduate	0.005	0.009	0.011	0.631	0.529	1.357	0.737
Postgraduate	0.022	0.012	0.030	1.791	0.074 *	1.191	0.840
Personal Annual Income Below 50,000 yuan	−0.005	0.008	−0.012	−0.610	0.542	1.516	0.660
Personal Annual Income 50,000–100,000 yuan	0.007	0.007	0.017	0.894	0.372	1.568	0.638
Personal Annual Income 160,000–200,000 yuan	−0.004	0.009	−0.007	−0.376	0.707	1.381	0.724
Personal Annual Income Above 250,000 yuan	−0.002	0.013	−0.003	−0.189	0.850	1.222	0.818
Family Annual Income Below 50,000 yuan	−0.010	0.008	−0.023	−1.272	0.204	1.362	0.734
Family Annual Income 50,000–100,000 yuan	−0.003	0.007	−0.007	−0.385	0.701	1.435	0.697
Family Annual Income 160,000–200,000 yuan	0.001	0.008	0.003	0.186	0.853	1.352	0.740
Family Annual Income (210,000–250,000) yuan	−0.010	0.013	−0.013	−0.782	0.435	1.177	0.850
Driver’s License Possession (Yes)	0.000	0.006	0.001	0.044	0.965	1.058	0.945
Vehicle Ownership of Car/Electric Bike	0.007	0.008	0.018	0.880	0.379	1.694	0.590
Vehicle Ownership of 2 Types	−0.003	0.008	−0.006	−0.317	0.751	1.645	0.608
Vehicle Ownership of 3 Types	0.010	0.008	0.025	1.217	0.224	1.761	0.568
R2	0.884
Adjusted R2	0.877
F	F(27, 479) = 134.775, p = 0.000
D-W Value	2.180

Note: *** denotes p < 0.01. * denotes p < 0.1.

shows the model results. The model, after eliminating redundant variables, has a good overall fit (R² = 0.884, adjusted R² = 0.877, F = 134.775, p < 0.001), indicating that the explanatory variables have a strong explanatory power for the dependent variable. As shown in Table 4, among the built environment variables, diversity (B = −0.081, p < 0.001), accessibility (B = −0.090, p < 0.001), street connectivity (B = −0.098, p < 0.001), and aesthetics (B = −0.084, p < 0.001) all show a significant negative relationship. The standardized coefficients (Beta) range from −0.374 to −0.398, indicating that they have a consistent and strong inhibitory effect on travel carbon emissions. Although the infrastructure and safety for walking variable does not reach the 1% significance level, it is marginally significant at the 10% level (B = −0.006, p = 0.107).

In terms of socioeconomic variables, the “male” gender variable has a significant positive impact on travel carbon emissions (B = 0.113, p < 0.001), while the age variable shows a significant negative relationship. Especially the “over 60 years old” group (B = −0.245, p < 0.001) has a standardized coefficient of −0.500, which has the greatest impact. Among the education variables, only the “postgraduate” group is marginally significant at the 10% significance level (B = 0.022, p = 0.074), and the other groups are not significant.

4.2. Age Heterogeneity

Grouped regression were conducted for three age groups, and tests were conducted for differences in coefficients between groups after grouped regression (

Table 5. Age Grouped Regression Table (n = 507).

	Overall		(1) 18–30 Years Old		(2) 31–60 Years Old		(3) Over 60 Years Old
	B	p	B	p	B	p	B	p
Constant	1.290	0.000 ***	1.475	0.000 ***	1.317	0.000 ***	1.153	0.000 ***
Residential density	0.007	0.261	−0.015	0.330	0.002	0.746	0.010	0.463
Diversity	−0.077	0.000 ***	−0.057	0.007 ***	−0.078	0.000 ***	−0.077	0.000 ***
P Stastic	/		0.139		0.907		0.243
Accessibility	−0.085	0.000 ***	−0.100	0.000 ***	−0.092	0.000 ***	−0.088	0.000 ***
P Stastic	/		0.384		0.754		0.342
Street connectivity	−0.104	0.000 ***	−0.134	0.000 ***	−0.097	0.000 ***	−0.107	0.000 ***
P Stastic	/		0.000 ***		0.356		0.027 **
Aesthetics	−0.078	0.000 ***	−0.079	0.000 ***	−0.079	0.000 ***	−0.080	0.000 ***
P Stastic	/		0.967		0.921		0.909
Infrastructure and safety for walking	−0.008	0.159	0.035	0.059 *	−0.009	0.094 *	0.009	0.475
Female	−0.111	0.000 ***	−0.126	0.000 ***	−0.106	0.000 ***	−0.134	0.000 ***
High school/technical secondary school	0.016	0.149	0.046	0.155	0.007	0.499	0.036	0.169
Junior college	−0.005	0.675	−0.009	0.826	−0.010	0.348	0.033	0.176
Undergraduate	0.007	0.592	−0.013	0.702	−0.003	0.796	0.040	0.120
Postgraduate	0.017	0.363	0.100	0.264	0.028	0.110	0.032	0.374
Personal Annual Income 50,000–100,000 yuan	0.010	0.375	−0.007	0.821	0.011	0.301	0.008	0.733
Personal Annual Income 110,000–150,000 yuan	0.005	0.663	−0.030	0.296	0.009	0.394	−0.013	0.585
Personal Annual Income 160,000–200,000 yuan	0.014	0.327	0.022	0.527	−0.004	0.748	−0.009	0.761
Personal Annual Income Above 250,000 yuan	0.023	0.237	−0.028	0.423	−0.011	0.548	0.026	0.610
Family Annual Income 50,000–100,000 yuan	0.013	0.304	−0.097	0.011 **	0.012	0.312	0.022	0.434
Family Annual Income 110,000–150,000 yuan	0.010	0.418	−0.039	0.199	0.022	0.063 *	0.013	0.563
Family Annual Income 160,000–200,000 yuan	0.015	0.268	−0.100	0.011 **	0.013	0.286	0.024	0.440
Family Annual Income 210,000–250,000 yuan	0.024	0.236	−0.087	0.053 *	0.018	0.367	0.004	0.935
Driver’s License Possession (Yes)	−0.011	0.185	0.025	0.231	−0.004	0.603	−0.016	0.379
Vehicle Ownership of Car/Electric Bike	0.012	0.320	0.025	0.432	0.013	0.244	0.008	0.732
Vehicle Ownership of 2 Types	0.012	0.336	0.024	0.517	−0.005	0.639	0.045	0.094 *
Vehicle Ownership of 3 Types	0.035	0.004 ***	0.036	0.251	0.020 *	0.079	0.022	0.380
R2	0.725		0.959		0.791		0.902
Adjusted R2	0.711		0.913		0.778		0.855
F	F(23, 483) = 55.250, p = 0.000		F(23, 20) = 20.570, p = 0.000		F(23, 367) = 60.399, p = 0.000		F(23, 48) = 19.272, p = 0.000

Note:*** denotes p < 0.01. ** denotes p < 0.05. * denotes p < 0.1. The P Statistic refers to the p-value corresponding to the test for the difference in coefficients between groups based on the seemingly unrelated regression model test method (suest test). The specific steps are as follows: first, perform regressions separately on the two groups of samples and store the results; second, conduct SUR estimation on the two sets of results; finally, carry out a test to examine the difference in coefficients between groups. Among these results, the p-values listed under column (1) and column (3). The p-values listed under column (3) represent the p-values for the test of regression coefficient differences between column (3) and column (1).

). To reduce data instability caused by an excessive number of groups in the grouped regression analysis, we merged certain age ranges in the age-grouped regression. The final categories were set as 18–30, 31–60, and over 60 years old. This grouping aligns with shifts in social roles among urban residents in China.

All three models passed the significance test and had good fitting conditions. The R² values were 0.959, 0.791, and 0.902, respectively. The adjusted R² values were 0.913, 0.778, and 0.855, respectively. The F values were 20.570 (p < 0.001), 60.399 (p < 0.001), and 19.272 (p < 0.001), respectively.

In the 18–30 years old group, built environment elements such as diversity (B = −0.057, p < 0.01), accessibility (B = −0.100, p < 0.001), street connectivity (B = −0.134, p < 0.001), and aesthetics (B = −0.079, p < 0.001) all showed a highly significant negative relationship.

In the 30–60 years old group, diversity (B = −0.078, p < 0.001), accessibility (B = −0.092, p < 0.001), street connectivity (B = −0.097, p < 0.001), and aesthetics (B = −0.079, p < 0.001) were still significant. Female (B = −0.106, p < 0.001) were a significant variable.

In the over-60-year-old group, diversity (B = −0.077, p < 0.001), accessibility (B = −0.088, p < 0.001), street connectivity (B = −0.107, p < 0.001), and aesthetics (B = −0.080, p < 0.001) remained significant.

Furthermore, the influence of street connectivity on travel carbon emissions shows significant differences across different age groups.

4.3. Gender Heterogeneity

To explore the heterogeneity of gender in the mechanism influencing travel carbon emissions, this study further established multiple linear regression models for male and female samples, and tests were conducted for differences in coefficients between groups after grouped regression. Both models showed good fitting effects. The R² of the male group was 0.874, the adjusted R² was 0.855, and F = 46.572 (p < 0.001). The R² of the female group was 0.871, the adjusted R² was 0.856, and F = 58.447 (p < 0.001). Both reached the 1% significance level (

Table 6. Gender Grouped Regression Table (n = 507).

	Overall		Male		Female
	B	p	B	p	B	p
Constant	1.394	0.000 ***	1.445	0.000 ***	1.361	0.000 ***
Residential density	0.009	0.076 *	0.003	0.639	0.003	0.545
Diversity	−0.083	0.000 ***	−0.079	0.000 ***	−0.082	0.000 ***
P Stastic	/		0.636
Accessibility	−0.089	0.000 ***	−0.086	0.000 ***	−0.094	0.000 ***
P Stastic	/		0.257
Street connectivity	−0.101	0.000 ***	−0.098	0.000 ***	−0.099	0.000 ***
P Stastic	/		0.851
Aesthetics	−0.091	0.000 ***	−0.088	0.000 ***	−0.080	0.000 ***
P Stastic	/		0.253
Infrastructure and safety for walking	−0.004	0.407	−0.005	0.375	−0.007	0.149
31–40 years old	−0.073	0.000 ***	−0.083	0.000 ***	−0.061	0.000 ***
41–50 years old	−0.130	0.000 ***	−0.132	0.000 ***	−0.126	0.000 ***
51–60 years old	−0.178	0.000 ***	−0.191	0.000 ***	−0.174	0.000 ***
Over 60 years old	−0.247	0.000 ***	−0.245	0.000 ***	−0.245	0.000 ***
High school/technical secondary school	0.021	0.043 **	0.017	0.124	0.004	0.709
Junior college	0.003	0.752	−0.003	0.809	0.000	0.994
Undergraduate	0.017	0.155	0.013	0.297	0.001	0.942
Postgraduate	0.039	0.018 **	0.008	0.646	0.040	0.028 **
Personal Annual Income 50,000–100,000 yuan	0.015	0.146	0.025	0.019 **	0.002	0.881
Personal Annual Income 110,000–150,000 yuan	−0.004	0.704	−0.009	0.420	0.014	0.198
Personal Annual Income 160,000–200,000 yuan	−0.005	0.713	0.016	0.262	−0.005	0.678
Personal Annual Income Above 250,000 yuan	0.018	0.292	0.007	0.672	0.002	0.903
Family Annual Income 50,000–100,000 yuan	0.014	0.225	0.012	0.318	0.005	0.643
Family Annual Income 110,000–150,000 yuan	0.015	0.163	0.012	0.313	0.011	0.314
Family Annual Income 160,000–200,000 yuan	0.016	0.187	0.014	0.288	0.010	0.409
Family Annual Income 210,000–250,000 yuan	0.001	0.960	−0.006	0.762	0.001	0.950
Driver’s License Possession (Yes)	−0.003	0.717	0.009	0.249	−0.008	0.322
Vehicle Ownership of Car/Electric Bike	0.011	0.280	0.012	0.294	0.004	0.718
Vehicle Ownership of 2 Types	−0.001	0.928	−0.000	0.984	−0.001	0.925
Vehicle Ownership of 3 Types	0.018	0.096 *	0.016	0.174	0.008	0.455
R2	0.783		0.874		0.871
Adjusted R2	0.770		0.855		0.856
F	F(29, 477) = 59.415, p = 0.000		F(29, 195) = 46.572, p = 0.000		F(29, 252) = 58.447, p = 0.000

Note: *** denotes p < 0.01. ** denotes p < 0.05. * denotes p < 0.1. The P Statistic refers to the p-value corresponding to the test for the difference in coefficients between groups based on the seemingly unrelated regression model test method (suest test). The specific steps are as follows: first, perform regressions separately on the two groups of samples and store the results; second, conduct SUR estimation on the two sets of results; finally, carry out a test to examine the difference in coefficients between groups.

).

In terms of built environment variables, diversity, accessibility, street connectivity, and aesthetics all showed a highly consistent significant negative impact on the travel carbon emissions of both men and women (p < 0.001). Infrastructure and safety for walking were not significant in either group. However, the influence of the built environment elements on travel carbon emissions does not show significant differences across gender groups.

5. Discussion

This study provides empirical evidence confirming the significant role of the built environment in influencing residents’ carbon emissions. The findings demonstrate that four categories of variables—diversity, accessibility, street connectivity, and aesthetics—consistently exhibit highly significant negative correlations with travel carbon emissions. This aligns with the behavior-environment interaction theory, where the environment provides constraints or support for behavior.

The results show that diversity suppresses travel carbon emissions, which is largely consistent with domestic and international findings [34]. This phenomenon can be explained by the fact that older neighborhoods with higher land-use mix provide more opportunities for “one-stop” services, shorten travel distances, facilitate the choice of low-carbon travel modes, and thereby reduce travel carbon emissions. This aligns with Cervero’s analysis of the impact of diversity in the “5D” elements [35].

There is a negative correlation between accessibility and travel carbon emissions. It is consistent with Nielsen’s findings that higher destination accessibility increases the likelihood of low-carbon travel [36].

Old communities with more compact road networks and better slow-traffic systems have higher rates of walking or cycling and lower travel carbon emissions. This result is consistent with previous research, indicating that better road systems and higher public transport service levels encourage residents to choose green travel modes, resulting in lower daily travel carbon emissions. It also aligns with practical experiences in China regarding the “15 min living circle” and “neighborhood micro-renewal” [37,38].

There is a negative correlation between street aesthetics and travel carbon emissions. Aesthetics is not only about visual appeal but also about comfortable experiences, which can translate into residents’ active preference for low-carbon travel, thereby reducing carbon emissions.

It is further found that identical built environment elements can lead to differentiated perceptions and travel behaviors depending on the individual characteristics of residents. Among the socioeconomic variables, gender (male) exhibited a positive association with travel carbon emissions, while age demonstrated a significant negative relationship. This finding is divergent from previous studies, which generally reported non-significant effects of age and gender on travel carbon emissions [39]. The discrepancy may be attributed to the shift in research scale—from macro-level urban analyses to micro-level old-community focus—where the explanatory power of these variables is magnified and behavioral differences linked to age and gender become more pronounced.

As residents age, their travel carbon emissions gradually decrease. This difference can be attributed to the lifestyle patterns and priority needs at different stages of the life cycle. Younger groups, due to more frequent commuting, socializing, and exploratory activities, tend to rely more on motorized modes of travel. Middle-aged individuals, despite having clear travel purposes and higher trip frequencies, often opt for efficiency-oriented transportation modes under time constraints. In contrast, during older age, as activity ranges shrink and lifestyles slow down, daily travel gradually shifts toward low-carbon modes such as walking and public transit, with non-essential travel needs significantly reduced.

Notably, the influence of street connectivity on travel carbon emissions varies significantly across different age groups, with its impact on reducing emissions being markedly stronger among younger groups than among middle-aged and older adults. This may be because younger individuals’ travel patterns are often more exploratory and multi-purpose. Highly connected street networks provide them with more flexible and efficient route choices, supporting combined use of walking, cycling, and public transit to optimize multi-segment trips, reduce detours, and minimize waiting time. Additionally, younger people use digital navigation tools more frequently. The diverse route options offered by well-connected networks synergize with their tech-savvy habits, further enhancing the feasibility of low-carbon travel.

In terms of gender, males tend to have higher travel carbon emissions than females. diversity, accessibility, street connectivity, and aesthetics are all inversely correlated with travel carbon emissions for both men and women, yet there is no significant difference in the influence of these various factors between genders. This suggests that the absolute difference in travel carbon emissions between genders may stem more from structural factors such as social role division, occupational commuting patterns, and travel distance, while their responsiveness to changes in built environment quality appears similar.

In summary, the study identifies the mechanism by which built environment elements in old communities influence residents’ travel carbon emissions, and emphasizes the importance of individual-level data in understanding these impacts. This conclusion provides strong evidence for understanding the actual operation logic of the “built environment–travel behavior–travel carbon emissions” mechanism at the urban micro-scale, and offers solid theoretical support for the renewal and transformation of old communities.

In areas where large-scale land-use restructuring is not feasible, the goals of reducing residential carbon emissions can be achieved by improving accessibility and optimizing design.

Tailored adjustment strategies for the built environment should be developed to support the low-carbon transformation of different old communities. For example, in response to the “street connectivity” and “aesthetics” elements, it is recommended to implement systematic road pavement renewal on the main community thoroughfares and establish continuous, barrier-free pedestrian walkways within the community to ensure the coherence and comfort of walking paths. This is because poor street pavement quality is widespread in older neighborhoods, which not only affects the continuity and safety of passage but also weakens the overall visual perception and environmental appeal of the space.

In response to the “diversity” and “accessibility” variables, it is suggested to enhance the mix of commercial, office, educational, and medical functions within the community to increase spatial vitality. This can reduce the need for cross-regional travel, ultimately helping to lower residents’ travel carbon emissions.

6. Conclusions

This study examines typical old urban communities in Chengdu, employing multiple linear regression models to reveal the structural impact of built environment elements on travel carbon emissions, and validating the causal pathway of “spatial structure—travel behavior—travel carbon emissions” at the micro-neighborhood scale. Key findings include:

(1)

Diversity, accessibility, street connectivity, and aesthetics consistently demonstrated significant negative effects across demographic groups. Increased land-use mix helps shorten daily travel distances and reduce trip generation. The presence of informal micro-pathways enhances actual pedestrian permeability, reducing reliance on motorized transport. Preserved high-density street networks improve route flexibility and comfort, strengthening the appeal of non-motorized travel. Livable aesthetics significantly increase residents’ walking frequency and willingness.

(2)

Overall, travel carbon emissions decrease with age, and males tend to have higher emissions than females. In terms of the built environment, street connectivity has a significantly stronger effect on reducing travel carbon emissions among younger groups compared to middle-aged and older adults.

(3)

No significant correlation was found between residents’ low-carbon awareness, attitudes, and their actual travel behavior. This indicates that travel choices may be more strongly constrained by socio-economic factors and built environment conditions rather than solely determined by policy advocacy or individual intentions.

These findings provide a theoretical basis for low-carbon retrofitting of the built environment in old communities. The research conclusions are primarily applicable to old communities with built environment shortcomings in high-density core cities. At the same time, we have identified factors that may limit the generalizability of the conclusions: First, urban form and density—the carbon reduction benefits of elements such as street connectivity may vary in intensity depending on different road network structures and mixed-use development intensities. Second, the uniqueness of the public transportation infrastructure—if the public transportation coverage in the target area is significantly higher or lower than that in the study area, the influence of the built environment on travel mode choices may shift in weighting. Third, regional climate characteristics—extreme weather conditions (such as severe cold or intense heat)—may reduce the willingness for slow-mode travel, thereby affecting the emission reduction outcomes of built environment adjustments.

Our study also has a few limitations worthy of further discussion. First, the study is based on cross-sectional data. Although we have constructed reasonable explanations through theoretical frameworks and statistical controls, the self-selection effect of residents’ residential choices and the long-term impact of behavioral adaptation cannot be fully disentangled. Temporal dynamic analysis could be introduced to examine how the built environment influences the evolution of residents’ travel carbon emissions over time. Meanwhile, future research could supplement questionnaires and interviews with behavioral monitoring methods such as mobile signaling data and GPS trajectory data. This would help develop travel pattern recognition models with higher spatiotemporal precision.