Identifying Influence Mechanisms of Low-Carbon Travel Intention Through the Integration of Built Environment and Policy Perceptions: A Case Study in Shanghai, China

Abstract

Promoting low-carbon travel modes is crucial for China’s transportation sector to achieve the dual carbon goals. When exploring the mechanisms behind individuals’ travel decisions, the relationships between factors such as the built environment and transportation policies are often derived from prior experience or subjective judgment, rather than being grounded in a solid theoretical foundation. In this paper, we build on and integrate the Theory of Planned Behavior (TPB) and the Technology Acceptance Model (TAM) by introducing built environment perception (BEP), encouraging policy perception (EPP), and restrictive policy perception (RPP) as either perceived ease of use (PEOU) or perceived usefulness (PU). The integration aims to explain how the latent variables in TPB and TAM jointly affect low-carbon travel intention. We conduct a traveler survey in Shanghai, China to obtain the data and employ a structural equation modeling (SEM) approach to characterize the latent mechanisms. The SEM results show that traveler attitude is the most critical variable in shaping low-carbon travel intentions. Perceived ease of use has a significant positive effect on perceived usefulness, and both constructs directly or indirectly influence attitude. As for transportation policies, encouraging policies are more effective in fostering voluntary low-carbon travel intentions than restrictive ones. Considering the heterogeneity of the traveling population, differentiated policy recommendations are proposed based on machine learning modeling and SHapley Additive exPlanations (SHAP) analysis, offering theoretical support for promoting low-carbon travel strategies.

1. Introduction

In China, the third largest source of carbon emissions, the transportation sector is facing significant challenges in terms of carbon reduction [1,2]. According to the International Energy Agency, most sectors in China are projected to experience peak carbon emissions before 2028, while the transportation sector is expected to peak after 2040 [3]. Road transport generates the largest amount of carbon emissions within the Chinese transportation sector, accounting for 81.1% in 2020, with urban road transport contributing about 70% [4,5].

To address carbon emission issues, the Chinese government declared the “dual car-bon goals”, where China will strive to achieve “carbon peaking” (carbon emissions reaching a national peak) by 2030 and achieve “carbon neutrality” (total carbon emissions balanced by those removed through natural or technological means) by 2060. Non-gasoline-car-based transportation modes such as public transportation, cycling, or driving electric vehicles exhibit considerable importance in promoting low-carbon travel options. However, despite many efforts, the outcome has still been unsatisfactory [6]. A study in Xiamen, China revealed that over 67% of the surveyed travelers regularly use public transportation despite being dissatisfied. They have a high potential to switch to private cars if given the option, indicating that their continued use is due to a lack of alternatives rather than genuine preference [7]. Given this, strategies to enhance low-carbon travel intention among urban travelers are of high interest to policymakers. To formulate effective policies, it is necessary to identify the factors that shape low-carbon travel intentions and their influence mechanisms, which will lead to more effective carbon emission reduction in the transportation sector in China.

Recent studies have developed theoretical frameworks for understanding factors influencing individuals’ low-carbon travel intention [8,9,10]. Factors like built environment [11], policies [12], and socioeconomic attributes [13] are related to the objective environment in which individuals are located and are considered to effectively guide the decision-making of individuals. In addition, some factors are associated with individuals’ subjective awareness and feelings, such as psychological latent variables [14]. These factors may reflect individuals’ internal motivation to engage in a behavior, as well as the logic underlying their decision-making. Some studies have also considered various factors to explore the decision-making process related to individual travel behavior. Manaugh, et al. [15] analyze barriers to the adoption of cycling as a regular transport mode by examining a range of factors, including the built environment, psychological elements, and socioeconomic influences. Sherriff, et al. [16] explore the motivations, barriers, and decision-making processes associated with the use of shared bicycles, while considering individual, social, and environmental factors. However, most current studies either generally examine the impact of specific factors on travel behavior without thoroughly analyzing the underlying mechanisms or empirically define factor relationships with limited theoretical foundation. Mandal, et al. [17] also point out that the influences of the built environment on perceived behavioral control and attitudes are often overlooked in travel behavior studies. To our knowledge, a more comprehensive theoretical framework is needed to establish more reasonable relationships among factors and to develop a more effective mechanism model. Moreover, factors such as the built environment and policies—which are less explored in travel behavior research but have proven valuable in behavioral research in other fields—should be incorporated to enrich the theoretical model.

A unique aspect of our study lies in the use of structural equation modeling (SEM) that integrates insights from the Technology Acceptance Model (TAM) into the conventional Theory of Planned Behavior (TPB), forming an integrated TPB-TAM framework. Within this framework, factors in TAM identified as perceived usefulness or perceived ease of use are directly related to the factors of TPB. The integration establishes more rational internal relationships among factors and yields a more reliable influence mechanism than can be achieved using TPB or TAM alone. Based on the SEM results, we further apply SHAP-based machine learning analysis to segment the travel population according to individual heterogeneity, identifying policy priorities for different traveler groups.

In our study, low-carbon travel refers to (1) public transportation systems (buses, subways, trams, etc.), (2) non-motorized travel options (walking and cycling), (3) shared mobility services (customized shuttle buses and bike-sharing), and (4) clean energy vehicles, all of which collectively reduce transportation-related greenhouse gas emissions. While many types of personal travel exist, this study focuses on urban travel in Chinese megacities, defined as cities with an urban population of over 10 million [18]. These cities exhibit greater total and per capita carbon emissions, along with higher car ownership rates. Consequently, they face more significant challenges in promoting low-carbon transportation [19]. Moreover, it is argued that, for big cities with well-developed traffic systems, increasing the low-carbon travel intention of travelers may encourage a higher proportion of low-carbon travelers and stabilize travel mode choices [20]. Overall, understanding the influencing factors for low-carbon travel intentions in megacities is of practical importance in China.

The rest of this paper is organized as follows. In Section 2, we review the literature related to theories and methods in low-carbon travel research, identifying the research gaps we try to fill. Section 3 introduces the structural equation model and machine learning classification models based on the TPB-TAM framework. Section 4 provides a comprehensive analysis of the model results, which are then discussed in detail in Section 5. Section 6 concludes with the research implications and discusses the limitations of the paper.

2. Literature Review

2.1. The Theoretical Framework

(1)

The Theory of Planned Behavior

The Theory of Planned Behavior (TPB) offers a framework to help understand the psychological factors that shape an individual’s behavior. According to the TPB, the behavioral intention of an individual is influenced by three core factors, namely, attitudes (ATT), subjective norms (SN), and perceived behavior control (PBC) [21]. ATT reflects whether a person views the behavior positively or negatively. SN refers to the extent to which a person believes that people who are important to them, such as family, friends, or colleagues, support or oppose the behavior. PBC is about how easy or difficult a person perceives that it is to carry out a behavior.

A more thorough understanding of human behavior has been pursued through the TPB, which has already been widely used in the study of behavioral intentions and travel behavior. By incorporating additional variables or dimensions, scholars have developed the Extended Theory of Planned Behavior (ETPB), allowing for improved prediction and explanation of travel-related behaviors. For example, Fu [22] examines how travelers perceive the environmental consequences of private car use and how they emotionally respond to it. Han, et al. [23] analyze the behavior of international tourists in the U.S. during the post-pandemic period, taking into account their knowledge of COVID-19. In a related study, Huang and Gao [24] explore how commuters’ low-carbon literacy (low-carbon knowledge, habits, affection, and awareness) shapes their mode choice intentions.

(2)

The Technology Acceptance Model

The Technology Acceptance Model (TAM) is a theoretical framework initially proposed by Davis in 1989 to explain and predict users’ acceptance and usage of technology. It can be regarded as a derivative of the TPB, but it simplifies the variables to better fit the analysis of technology adoption behaviors. This framework primarily focuses on two key constructs that influence users’ behavioral intention to adopt a technology: perceived usefulness (PU), which refers to the extent to which a user believes that using the technology would enhance his or her job performance or efficiency, and perceived ease of use (PEOU), which reflects the degree to which a user believes that using the technology is free of effort. These two constructs influence users’ behavioral intention, which in turn determines their actual use behavior.

TAM has been widely applied in the study of user acceptance across various fields, including information systems, mobile applications, and e-learning platforms. For example, some previous research [25,26,27] investigated the public acceptance of autonomous driving within the framework of TAM. Batouei, et al. [28] explore the public acceptance of ChatGPT as a supportive tool for enriching travel experiences, by integrating constructs including perceived convenience, information quality, personal innovativeness, and compatibility into the TAM framework. A study from Chongqing, China explores the factors influencing travelers’ willingness to adopt travel intention information on social media, revealing different mechanisms at both individual and group levels for transportation demand management [29]. Wang, et al. [30] conduct a study with residents in Beijing, investigating the public acceptance of MaaS (Mobility as a Service) platforms and their impact on the intention to adopt green transportation options. Over time, researchers have also introduced variables from other frameworks into TAM, such as SN and PBC from TPB, to enhance its applicability to the study of more complex travel behavior [31,32,33,34].

Based on our review, attitude is generally found to have a strong and direct impact on behavioral intentions. Jiang, et al. [35] observed the role of subjective norms in affecting both self-practiced and interpersonal intentions, and this effect is mediated by attitudes. A study by Mandal, Johansson, and Lindelöw [17] further highlighted a significant correlation between individuals’ attitudes toward and perceived control over the built environment and their walking behavior. Ru, et al. [36] distinguish between attitudes and subjective norms and investigate how they interact in shaping green travel intentions, reinforcing the importance of attitudes. Not all research, however, supports the dominant influence of attitudes. For example, Juschten, et al. [37] conducted a study on the intentions of urban residents traveling to nearby destinations and found that subjective and social norms were the most powerful predictors of intention, with attitudes appearing to have little influence. The variation in conclusions may be influenced by differences in the study area, research period, target population, and modes of travel. However, it is undeniable that these studies offer valuable insights into and useful frameworks for understanding the influencing factors and mechanisms of travel intentions.

2.2. Travel Decision-Making

Travel decision-making is a complex process influenced by a variety of psychological, socioeconomic, and environmental factors. Understanding how individuals make travel choices is crucial for designing effective transportation policies and promoting sustainable travel behaviors. Behavioral change towards low-carbon travel, such as substituting car trips with green travel modes, has been explored in many studies, particularly those grounded in the TPB [38,39,40,41].

The influence of the built environment on travel behavior has become a topic of considerable interest among researchers. The “3Ds” (density, diversity, and design) and “5Ds” (density, diversity, design, distance to transit, and destination accessibility) concepts proposed by Cervero and Kockelman [42] are classic theories of the built environment. It is confirmed that there exists a strong link between the built environment and travel behavior [43,44,45,46]. In addition to the physical environment, social factors, such as neighborhood relations, have also been found to significantly affect travel behaviors [47,48]. Chen, et al. [49] emphasize the value of interpreting the built environment through an individual perspective. Through the mechanisms of ATT and PBC, Dill, et al. [50] demonstrate that travel behaviors are indirectly affected by demographics, as well as features of the built environment. Moreover, several studies have noted notable gender differences in travel intentions [51,52,53]. However, current research concerning these crucial factors tends to rely on empirical generalizations or subjective assumptions, rather than on solid theoretical foundations. Thus, it is necessary to construct a more systematic conceptual framework to better understand the complex mechanisms underlying travel behavior.

The main research methods in this area include Discrete Choice Models (DCMs), structural equation modeling (SEM), and machine learning (ML). DCMs are primarily used to predict the impact of independent variables on travel intentions, but they cannot directly explain unobserved latent variables. SEM is the most commonly used method in intention studies involving latent variables. In recent years, ML has begun to be applied in travel behavior research, often in combination with interpretability analysis. For example, the multinomial logit model has been used to predict residents’ acceptance of car-sharing services and the behavioral motivations behind their travel decisions [54,55]. The influencing mechanisms of users’ carpooling intentions have been explored using PLS-SEM, providing recommendations for ride-hailing platforms and policymakers [56,57]. Ong, et al. [58] employ machine learning algorithms to identify the latent variables influencing public transport service quality and passenger satisfaction and assess passengers’ future intention to use public transportation. Similarly, Xu, et al. [59] integrate multiple machine learning methods to explore the factors affecting post-pandemic travel preferences, quantifying the relationship between these factors and travel intentions through marginal effects. Furthermore, machine learning has been widely applied in studying the nonlinear relationships between the built environment and travel behavior, as well as identifying the threshold effects of key factors within these relationships. Global studies focus on diverse population groups, including adolescents, the elderly, people with disabilities, and low-income communities, examining travel modes like walking and carpooling, and built environment factors such as population density, road density, land use diversity, and proximity to city centers [60,61,62,63].

2.3. Policies to Promote Low-Carbon Travel

As a key promoter of low-carbon travel, the government plays a crucial role through the design and implementation of transportation policies, as well as enhancement of the existing transit infrastructure. Research has demonstrated that replacing short car trips with walking and cycling offers potential for greenhouse gas reduction, and that the quality of infrastructure significantly influences individuals’ daily travel decisions [64]. The role of active transportation in reducing carbon emissions has also been further evaluated at a micro-level [65,66]. Studies have confirmed that policies, whether encouraging (e.g., improved public transport services, comprehensive infrastructure, bonus–malus taxes, sharing economy, and transit-oriented development) or restrictive (e.g., fewer car quotas, stricter fuel economy standards, higher parking fees, and limited parking supply), are effective in promoting low-carbon travel behavior [43,67,68,69,70,71]. The differential effectiveness of transportation policies across different population groups remains inconclusive and requires further investigation [72]. Identifying the most effective policy mechanisms for promoting low-carbon travel is essential for optimizing strategy design and ensuring targeted, impactful interventions in the transition to sustainable mobility.

2.4. Research Gaps

Based on our review of the literature, two gaps have been identified. First, while factors such as the built environment and transportation policies are vital in promoting low-carbon travel, existing studies often formulate path hypotheses regarding their interactions based on empirical generalizations or subjective judgments, rather than on solid theoretical grounds [73]. There is an urgent need to develop a more comprehensive theoretical framework to explore the underlying mechanisms. Second, policy recommendations fail to address the diverse travel needs, economic conditions, and willingness to adopt low-carbon travel across different traveler groups. For example, the mandatory promotion of electric vehicles may be effective for high-income travelers due to their greater financial flexibility. However, for travelers with lower incomes, electric vehicles may be less attractive for several reasons: (1) low-income individuals may be reluctant to invest in new electric vehicles when their current vehicles are still functional, and (2) the higher costs associated with insurance, maintenance, and battery replacement for electric vehicles can pose significant financial burdens for low-income individuals [74]. Implementing a “one-size-fits-all” policy may marginalize vulnerable traveler groups, exacerbating social inequality.

3. Methodology

The data required to perform the SEM were collected through a traveler survey in Shanghai. A total of 361 valid responses were obtained from the 400 respondents, resulting in an effective response rate of 90.25%. Details about the target respondents, design process, content, platform, and quality control measures of the survey are provided in Appendix A.1.

We propose the following conceptual framework to understand the factors influencing low-carbon travel intention (LI), as shown in Figure 1. Drawing from the TPB, ATT, SN, and PBC are considered core variables of the model, each of which is theorized to influence LI. Built environment perception (BEP), encouraging policy perception (EPP), and restrictive policy perception (RPP), are integrated into the framework as indicators of TAM, with BEP representing perceived ease of use (PEOU) and EPP and RPP reflecting perceived usefulness (PU). It is hypothesized that PEOU will have a positive effect on PU. Both PEOU and PU are expected to positively influence ATT. Additionally, PU is anticipated to directly impact LI. In both the TPB and TAM frameworks, ATT plays an important role. By integrating TPB and TAM, we can establish connections between the latent variables of the two frameworks. This allows us to address questions that a single framework cannot explore, such as how PU and PEOU influence ATT and whether relationships exist between PU/PEOU and SN/PBC.

Building on the TPB-TAM framework, we developed both a structural equation model and a ML model to conduct an in-depth analysis of low-carbon travel intentions. First, by constructing the SEM model, we quantitatively analyzed the mechanisms underlying latent variables and provided directions for policy-making. However, the findings from the path analysis remain general and provide limited guidance for specific policy recommendations. SHapley Additive exPlanations (SHAP)-based machine learning fills the gap and offers more precise identification of key target groups with carbon reduction potential. The integration of the two methods creates a powerful synergy for developing targeted, group-specific policy implementation strategies.

3.1. Structural Equation Model

Urban travel behavior in megacities is often shaped by unobservable psychological factors. Recognizing these factors is important for predicting low-carbon travel intentions. Latent constructs, such as well-being, beliefs, and values, help reveal the hidden social and psychological motivations that affect travel decisions. A commonly used method is the structural equation modeling approach with a five-step process [75], as illustrated in Figure 2. The first step is to construct a theoretical model. Based on the research problem and hypotheses, variables and their relationships are specified in this step to form an initial SEM. The second step is data collection, cleaning, and preprocessing. The third step is model fitting, where statistical software is used to estimate the model parameters. The fourth step is to evaluate the estimated model and make necessary modifications based on the results. In the final step, interpretations of the model parameter estimates are provided, along with discussions of the theoretical and practical implications.

Structural equation modeling characterizes how various aspects of an observable, theoretical phenomenon are structurally related to another one. In SEM, a set of latent constructs is specified, each of which is linked to several observed variables. This method enables simultaneous analysis of both the relationships between latent constructs and observed variables, as well as between latent constructs [76,77]. Structural equation modeling has been widely used in social sciences, economics, political science, psychology, and behavioral sciences. In transportation research, structural equation modeling is mainly used for modeling travel behavior and values.

This study incorporates four latent constructs from the TPB (ATT, SN, PBC, and LI), and three constructs from the TAM (BEP, EPP, and RPP) into the SEM model. Each latent construct is associated with multiple observed variables, whose values are collected from a survey conducted in this study.

The causal relationships among the latent constructs establish the structural model in the SEM, while the relationship between the latent constructs and the observed variables forms the measurement model.

(1)

The structural model

Figure 3 presents our proposed theoretical framework for the variable relationships. Based on the arrows that connect the latent constructs in the figure, 14 hypotheses are formed and will be tested by SEM.

H1. 

Attitudes positively influences low-carbon travel intention.

H2. 

Subjective norms positively influences attitudes.

H3. 

Perceived behavioral control positively influences attitudes.

H4. 

Encouraging policy perception positively influences subjective norms.

H5. 

Encouraging policy perception positively influences attitudes.

H6. 

Encouraging policy perception positively influences perceived behavioral control.

H7. 

Built environment perception positively influences attitudes.

H8. 

Restrictive policy perception positively influences subjective norms.

H9. 

Restrictive policy perception positively influences attitudes.

H10. 

Restrictive policy perception positively influences perceived behavioral control.

H11. 

Built environment perception positively influences encouraging policy perception.

H12. 

Built environment perception positively influences restrictive policy perception.

H13. 

Encouraging policy perception positively influences low-carbon travel intention.

H14. 

Restrictive policy perception positively influences low-carbon travel intention.

(2)

The measurement model

We draw upon the work of Li and Zhang [78] for specifying the observed variables to measure the latent constructs ATT, SN, and PBC. The measurement of latent construct BEP is based on four observed variables following the “3Ds” and “5Ds” theory [42], with a particular focus on the perceived convenience of public transportation infrastructure, such as station accessibility and route availability. For latent constructs RPP and EPP, we introduced five variables related to Transportation Demand Management, including passenger vehicle license auction and restriction, as well as parking price regulation (which are restrictive policies to limit the number of vehicles on the road) and public transport priority (which encourages the use of public transportation through measures like dedicated bus lanes or transit signal priority). These two variables focus on the perceived effectiveness of transportation policies in facilitating low-carbon travel. All the observed variables were measured using a 5-point Likert scale (from “1 = strongly disagree” to “5 = strongly agree”) to help the survey respondents express their opinions. The latent and the observed variables used in the SEM are summarized in

Table 1. Latent and observed variables used in the SEM.

Latent Variables	Observed Variables	Description of the Observed Variables
Attitudes (ATT)	ATT1	Low-carbon travel is comfortable, safe and convenient.
	ATT2	I like low-carbon travel.
	ATT3	Low-carbon travel makes me happy.
Subjective Norms (SN)	SN1	I will make same choice as my friends/colleagues in adopting low-carbon travel.
	SN2	I will make same choice as my neighbors in adopting low-carbon travel.
	SN3	I will choose low-carbon travel because of media promotion and public opinion.
Perceived Behavioral Control (PBC)	PBC1	Time constraints will not affect my decision for low-carbon travel.
	PBC2	Severe weather will not affect my decision for low-carbon travel.
Built Environment Perception (BEP)	BEP1	My nearby transit stations feature smart, advanced technologies (e.g., real-time arrival information, user-friendly interfaces, and integrated payment systems), with safe, punctual buses and subways, as well as efficient transfers.
	BEP2	My surrounding subway and bus networks are dense, with well-planned routes and station placements.
	BEP3	It is convenient for me to get to transit stations.
	BEP4	Subway, bus, and shuttle services adequately meet my daily travel needs.
Restrictive Policy Perception (RPP)	RPP1	License plate auction is effective and encourages my low-carbon travel.
	RPP2	License plate restriction is effective and encourages my low-carbon travel.
	RPP3	Parking price regulation is effective and encourages my low-carbon travel.
Encouraging Policy Perception (EPP)	EPP1	Transit priority measures are effective and encourage my low-carbon travel.
	EPP2	Smartcards are effective and significantly encourage my low-carbon travel.
Low-Carbon Travel Intention (LI)	LI1	I am willing to choose low-carbon travel in the coming weeks.
	LI2	It is common for me to choose low-carbon travel.
	LI3	Low-carbon travel is one of my main transportation modes.
	LI4	I will continue to choose low-carbon travel in the near future.

.

3.2. Machine Learning Classification Model

Existing research [79,80,81,82,83] has demonstrated the use and effectiveness of ML in predicting travel mode choices. To address the “black box” challenge and improve model interpretability, the SHapley Additive exPlanations technique, or SHAP, has been proposed [84], which quantifies the contribution of each feature to model predictions.

Since no prior knowledge existed regarding the most suitable ML model for our problem, five ML classification models were initially trained. The models incorporated respondents’ demographic information and the observed variables in the SEM.

Based on model performance, the best-trained ML model was selected for SHAP analysis. Respondents were divided into two groups based on the average score of LI1, LI2, LI3, and LI4: (1) those with strong low-carbon travel intention (the average score ≥ 4, 84.2%) and (2) those with weak low-carbon travel intention (the average score < 4, 15.8%). Given the moderate data size and the mix of continuous and categorical variables, random forest classifier (RF), gradient boosting decision tree (GBDT), K-nearest neighbor classifier (KNN), logistic regression (LR), and decision tree (DT) models were trained [85,86,87,88].

The imbalanced dataset introduces training bias due to group disparity; therefore, metrics beyond accuracy, including the F1 score, recall, precision, and ROC curve, are essential to mitigate overfitting in the minority group.

For ranking the importance of different features, the mean SHAP value of each sample of the dataset for each feature is calculated. The average importance for feature j is the mean SHAP values over all the samples, as shown in Equation (1).

$${\mathrm{Importance}}_j={\displaystyle \frac{1}{N}}\sum _{i=1}^N\left|{\Phi }_{ij}\right|$$

(1)

where Φ_ij is the SHAP value for sample i for feature j. N is the number of samples used for the SHAP analysis.

SHAP also enables the impact of a feature in each sample to be decomposed into the interaction effects with others [89]. The definition of the SHAP interaction value is similar to that of the standard SHAP value, with the distinction that the object of measurement shifts from a single feature to a feature pair [90], allowing the quantification of interaction effects between two features. Thus, the SHAP-based dependency plots show how the contribution of a specific feature pair to the model prediction changes across different values.

4. Results

4.1. Descriptive Statistical Analysis

The demographic information of the survey respondents is shown in Figure 4. All the demographics are classified into three parts: personal, family, and travel attributes. In the personal part, 49% of the respondents are males, while the rest are females. The gender ratio of the respondents aligns with the population data of Shanghai (49.9% for males and 50.1% for females) [91]. The respondents are relatively young, with an average age of 29.44, lower than that of the city. The majority have received or are pursuing higher education. The respondents consisted mainly of corporate staff (62%), students (24%), and civil servants/public institution staff (5%). As for income, the average monthly income of the respondents is around CNY 11,000, which is comparable to the disposable income of Shanghai residents (CNY 7069.5) [91] after the deduction of personal income tax and social insurance fees. The residential areas of the respondents are dispersed, including downtown (37%), suburban areas (38%), and semi-central semi-suburban areas (25%), which is roughly consistent with the official data of Shanghai (26% for downtown, 51% for suburban areas, and 23% for semi-central semi-suburban areas) [91]. In total, 56% of the respondents live in houses they own, while 44% of them rent houses or live in dormitories.

In the family part, the respondents have an average household size of 3.4 (2.62 for the census data of Shanghai [91]). Specifically, 66% of the respondents have a household size of three or four family members. Based on the current situation of Chinese families, it is suggested that most of them may live with either their children or parents. Given that dropping off and picking up children for school is also part of parents’ daily routines in China, we also included statistics on the number and age of their children.

Regarding travel attributes, 57% of the respondents make one or more trips per day. As for car ownership, the majority have one or two cars, with 62% having gasoline cars and 38% having clean energy cars. In addition, 45% have Shanghai license plates except SH-C plates (a special type in Shanghai with strictly restricted driving access), while 55% having non-local plates or SH-C plates. Regarding car-related expenses, gasoline car owners generally spend over CNY 400 monthly on fuel, while 59% of electric car owners spend less than CNY 200 on charging each month.

Overall, the sample data in this survey covers important groups of residents in Shanghai and reflects the basic situation of the city, making it representative to a certain extent.

4.2. Reliability and Validity Test

Tests of reliability and validity are conducted before SEM, using SPSS 24.0 and MPlus 7.0. All the descriptions and threshold values of the main indicators are listed in

Table A1. Recommended values for structural, convergent, and discriminant validity.

Category	Interpretation			Recommended Value
Structural Validity	whether the theoretical relationships between latent constructs are supported	KMO	a measure of sampling adequacy in factor analysis	>0.7
		p-value	the representation of sample independence	<0.05
		Factor Loading	the correlation between the original variables and the common factors extracted in factor analysis	>0.6
Convergent Validity	the correlation between two measurement tools assessing the same construct	CR	the internal consistency of all observed variables reflecting a certain latent construct	>0.7
		AVE	a measure of convergence for latent construct estimation by observed variables	>0.5
Discriminant Validity	validating the distinctiveness of different constructs or attributes in a test			$\sqrt{\mathrm{AVE}}$ larger than the correlation with other variables

, and the calculated values of the indicators in the study are summarized in

Table 2. Reliability and convergence validity test.

Latent Construct	Observed Variables	Reliability	KMO	Parameters of Significant Test				Item Reliability	Composite Reliability	Convergence Validity
		Cronbach α		Estimate	S.E.	Est./S.E.	p-Value	SMC	CR	AVE
ATT	ATT1	0.745	0.870	0.536	0.043	12.402	***	0.287	0.754	0.512
	ATT2			0.780	0.030	26.178	***	0.608
	ATT3			0.801	0.029	27.862	***	0.642
SN	SN1	0.778		0.707	0.035	20.430	***	0.500	0.780	0.541
	SN2			0.763	0.032	23.509	***	0.582
	SN3			0.736	0.034	21.913	***	0.542
PBC	PBC1	0.709		0.820	0.055	14.975	***	0.672	0.717	0.561
	PBC2			0.671	0.051	13.047	***	0.450
BEP	BEP1	0.831		0.659	0.036	18.563	***	0.434	0.832	0.554
	BEP2			0.765	0.029	26.220	***	0.585
	BEP3			0.762	0.029	26.130	***	0.581
	BEP4			0.784	0.028	27.620	***	0.615
RPP	RPP1	0.759		0.801	0.033	24.524	***	0.642	0.763	0.521
	RPP2			0.741	0.035	21.242	***	0.549
	RPP3			0.609	0.041	14.932	***	0.371
EPP	EPP1	0.708		0.737	0.045	16.225	***	0.543	0.708	0.548
	EPP2			0.743	0.045	16.345	***	0.552
LI	LI1	0.828		0.719	0.032	22.632	***	0.517	0.835	0.559
	LI2			0.716	0.032	22.598	***	0.513
	LI3			0.732	0.031	23.980	***	0.536
	LI4			0.819	0.025	32.171	***	0.671

***: p < 0.001.

.

In the reliability test, all latent constructs here exhibit a Cronbach’s α coefficient over 0.7, indicating a high degree of consistency and stability of the test results, proving the re-liability of the scale data. Confirmatory factor analysis (CFA) is employed for the validity test [92], with 0.870 of the Kaiser–Meyer–Olkin (KMO) measure value confirming the applicability of CFA. In CFA, most of the latent constructs obtain the factor loadings over 0.6, with p < 0.001. In addition, the Composite Reliability (CR) and Average Variance Extracted (AVE) values of each latent construct are both above the criterion. The reliability and validity test confirms that the question set is capable of reasonably reflecting the corresponding latent constructs.

As shown in Figure 5, the square root of the AVE value (the diagonal value) of each latent construct exceeds the corresponding Pearson correlations (the off-diagonal values), which confirms the discriminant validity of the test results [93].

4.3. Path Analysis of the Structural Equation Model

SEM is performed in MPlus 7.0 to test the hypotheses proposed in Section 3.1. The results of the model fit collectively suggest that the model is generally consistent with the sample data. For detailed information on the model fit results of the structural model, please refer to

Table A2. Model fit result.

Indicator	Reference Value	Actual Value	Test
ML χ2	—	345.397	—
df	—	178	—
χ2/df	1 < χ2/df < 3	1.940	Pass
CFI	>0.9	0.940	Pass
TLI	>0.9	0.929	Pass
RMSEA	<0.08	0.051	Pass
SRMR	<0.08	0.070	Pass

and the corresponding description in Appendix A.2.

Table A3. The results from the hypothesis testing.

Hypothesis	Estimate	S.E.	Est./S.E.	p-Value	Significance	Result
H1	0.505	0.061	8.351	<0.001	***	Support
H2	0.511	0.069	7.435	<0.001	***	Support
H3	0.141	0.066	2.129	0.033	*	Support
H4	0.320	0.079	4.058	<0.001	***	Support
H5	−0.038	0.074	−0.512	0.609	—	Reject
H6	0.178	0.076	2.347	0.019	*	Support
H7	0.289	0.062	4.640	<0.001	***	Support
H8	0.361	0.071	5.088	<0.001	***	Support
H9	0.041	0.076	0.534	0.594	—	Reject
H10	0.344	0.069	4.975	<0.001	***	Support
H11	0.367	0.063	5.843	<0.001	***	Support
H12	0.432	0.056	7.683	<0.001	***	Support
H13	0.184	0.069	2.644	0.008	**	Support
H14	0.042	0.071	0.596	0.551	—	Reject

***: p < 0.001, **: p < 0.01, and *: p < 0.05.

reports the testing results for the 14 hypotheses. Based on the p-values, H1, H2, H3, H4, H6, H7, H8, H10, H11, H12, and H13 pass the significance test at a 5% level of significance, while H5, H9, and H14 fail. Figure A1 shows the initial path coefficients. The significance test in Table A3 and Figure A1 leads to the following conclusion:

(1)

ATT has a significant positive effect on LI (β = 0.520, p < 0.001).

(2)

BEP significantly influences both EPP (β = 0.366, p < 0.001) and RPP (β = 0.433, p < 0.001)

(3)

Paths through which EPP and RPP directly influence ATT are insignificant. Instead, paths where EPP or APP indirectly influence ATT through SN or PBC are significant (EPP→SN: β = 0.320, p < 0.001, EPP→PBC: β = 0.178, p < 0.015, RPP→SN: β = 0.361, p < 0.001, RPP→PBC: β = 0.344, p < 0.001).

(4)

Both EPP and RPP can indirectly affect LI by influencing ATT. In addition, EPP rather than RPP directly affects LI (β = 0.184, p < 0.01).

After deleting the insignificant paths, the results of the final model are shown in Figure 6. All the paths in the final model are highly significant with positive coefficients, confirming direct positive effects between variables. Among all tested paths, the path from ATT to LI demonstrates the most obvious influence. Specifically, if ATT increases by one standard deviation, LI will change by 0.520 standard deviations. The magnitude of the path coefficient indicates that an independent variable has a stronger direct impact on the dependent variable as the absolute value of its path coefficient increases.

The reasons behind these significant or insignificant effects will be further discussed in Section 5.1.

4.4. SHAP Analysis of the Machine Learning Model

The selection process of the optimal machine learning classification model is detailed in Appendix A.3. Specifically,

Table A4. Grid search hyper-parameters for the selected algorithms.

Classifier	Hyper-Parameter	Hyper-Parameter Grid	Best Hyper-Parameter
RF	Number of Estimators	5, 10, 20, 30, 40	20
	Maximum Depth	None, 5, 10, 20, 30	20
	Minimum Samples Split	2, 5, 7, 10	7
	Minimum Samples Leaf	1, 2, 3, 5	2
	Maximum Features	“sqrt”, ”log2”	“log2”
GBDT	Number of Estimators	50, 100, 200, 300, 500	50
	Learning Rate	0.01, 0.1, 0.2, 0.3	0.3
	Maximum Depth	3, 4, 5, 6, 7	7
	Subsample	0.7, 0.8, 0.9, 1.0	0.7
KNN	Number of Neighbors	6, 7, 8, 9, 10	9
	Weights	“uniform”, “distance”	“uniform”
	Algorithm	“auto”, “ball tree”, “kd tree”, “brute”	“auto”
LR	C	0.001, 0.01, 0.1, 1, 10	0.1
	Penalty	“l”, “l2”	“l2”
	Solver	“liblinear”, “lbfgs”, “newton-cg”, “sag”	“newton-cg”
DT	Criterion	“gini”, “entropy”	“entropy”
	Maximum Depth	None, 2, 3, 4, 5	3
	Minimum Samples Split	2, 5, 10	10
	Minimum Samples Leaf	1, 2, 4	1
	Maximum Features	“auto”, “sqrt”, “log2”	“log2”

summarizes the optimal hyper-parameters for each algorithm, while Figure A2 provides a comparison of their classification performance.

We choose the RF model for subsequent SHAP analysis, as it best detects and classifies samples among all the ML models. Figure 7 displays the top 20 variables influencing low-carbon travel intentions, which include both the latent variables used in the SEM model and demographics. Figure 7a ranks variables by importance in descending order, while Figure 7b illustrates how they specifically influence low-carbon travel intentions. Colors from blue to red reflect low to high feature values, with higher SHAP values indicating a stronger positive influence on the model output.

Figure 7 shows that ATT2 is the most influential variable, with its importance far exceeding others in affecting low-carbon travel intention. A high ATT value (red) positively influences the prediction, while a low value (blue) has a negative effect. ATT3 and ATT1 rank ninth and tenth, respectively. Consistent with the findings from the SEM, it is confirmed that attitudes have a significant positive impact on low-carbon travel intention.

Gender ranks second, showing that women (red for gender) are more likely to have strong intentions for low-carbon travel than men. Age ranks third, indicating that younger individuals (blue for age) are more inclined to adopt low-carbon travel compared to older individuals. PBC, SN, EPP, and BEP are all significant predictors of low-carbon travel intention and remain among the top 20 variables. In contrast, observed variables related to RPP are not included. Residents’ intentions are more strongly influenced by PBC and encouraging policies than by restrictive measures. This implies that enhancing public awareness of the controllability of performing low-carbon behaviors and promoting them through positive incentives may be more effective than enforcing strict regulations. Other variables suggest that people without private cars, those living in urban areas, and those who rent housing are more inclined to adopt low-carbon travel.

We further study the dependence plots between different feature pairs. As illustrated in Figure 8, the SHAP value gradually declines with age but rebounds after around 32. Additionally, the dependence plots of “age–kids” and “age–trip frequency” pairs reveal that young respondents under 32 are more likely to be childless and to have higher trip frequencies, whereas middle-aged respondents tend to have at least one child and exhibit lower trip frequencies. This suggests that the strong association between low-carbon travel intention and age may be closely linked to travel requirements involving children.

The dependence plots for monthly electric vehicle charging cost and monthly fuel cost with respect to age show that younger respondents tend to have higher charging expenses, while middle-aged respondents are more likely to bear higher fuel costs. This aligns with the current trend, where younger people constitute a higher proportion of clean-energy car owners than gasoline car owners. Furthermore, as shown in Figure 9b,c, there exists a notable overlap between individuals with children and those who spend more on fuel costs.

In recent years, the price of 92# gasoline (with a Research Octane Number of 92) in China has generally ranged from CNY 7 to 8 per liter [94]. For a compact passenger vehicle with a 40 L fuel tank, a full tank would cost approximately CNY 280–320, representing 4.0–4.5% of the average monthly disposable income in Shanghai (CNY 7069.5, according to official statistics [91]). Given the current high oil price, which leads to high travel costs for gasoline car owners, we also generated dependence plots associated with income. Some prior studies show that higher income is linked to lower low-carbon travel intentions [95,96]. In Figure 10, respondents with high incomes (over CNY 20,000) are indeed observed to generally show a lower intention and have a polarized distribution in terms of monthly fuel costs or trip frequency. In addition, middle-income salaried individuals, for whom the proportion of fuel costs to income is relatively high, are also noteworthy.

5. Discussion

5.1. Latent Variable Mechanism

Specifically, as mentioned in the significance test in Section 4.3, ATT has a significant effect on LI, consistent with prior research findings [97,98]. Moreover, ATT mediates the influence of both SN and PBC on LI. A possible explanation is that when individuals hold the awareness of strong support from society for low-carbon travel and the belief in adequate capability to participate in such behaviors, they tend to develop a positive attitude toward low-carbon travel, which in turn serves as a criterion of their preference for low-carbon travel.

BEP, representing PEOU in the TAM framework, is found to significantly influence both EPP and RPP, which serve as indicators of PU. Apart from this, BEP exhibits an essential influence on ATT, aligned with the TAM theory.

The traditional TAM framework indicates that PU can influence attitude, but the specific mechanism behind this effect remains unclear. In the integrated TPB-TAM framework of this research, the significance test in Section 4.3 proves that EPP and RPP are suggested to indirectly, rather than directly, influence ATT through both SN and PBC. This provides a new perspective to reveal the influence mechanism of such PU variables.

In practical contexts, policies usually create objective conditions (PBC) and regulate group behaviors (SN), thus affecting individuals’ attitudes toward the issue policies target. For example, both raising public parking fees (RPP) and offering bus transfer discounts (EPP) regulate the costs of various travel modes economically and simultaneously guide the travel choices of the public, which will influence individuals’ attitudes towards low-carbon travel. Meanwhile, it is EPP rather than RPP that directly influences LI. This may be because EPP can subtly enhance people’s willingness to spontaneously choose low-carbon travel through positive incentives [99], while under RPP, people are more likely to engage in specific behaviors out of passive compliance with restrictions or punitive measures, without forming subjective intention. Therefore, from the perspective of public perception, incentive-based policies are more effective than restrictive policies in fostering the public’s intention to adopt low-carbon travel.

To sum up, as it is influenced by nearly all other variables and has the greatest impact on the willingness to engage in low-carbon travel, ATT remains the most crucial variable in the integrated TPB-TAM framework, just as it is in the single TPB framework [40,100]. Furthermore, the specific influence mechanisms of PU and PEOU on ATT have been uncovered, a distinction that cannot be achieved through the single TPB or TAM frameworks. Consequently, the integrated TPB-TAM framework not only focuses on how latent variables such as EPP, BEP, and RPP influence individuals’ positive attitudes toward low-carbon travel, but also offers new insights for policymakers aiming to foster such attitudes. Moreover, compared to restrictive transportation policies, encouraging and supportive measures are more effective in shaping residents’ intentions toward low-carbon travel.

5.2. Population Segmentation and Policy Recommendations

The SEM results confirm the essential importance of attitude to low-carbon travel intentions. However, while the findings can provide reliable directions for policy formulations, they often fall short in offering specific measures that take into account the characteristics of certain populations. As Li, Zhao, Ma, and Qin [38] proposed, regional and generational differences in influencing factors should be given consideration. Moreover, research indicates that, when designing strategies to encourage low-carbon travel, policymakers need to pay greater attention to population heterogeneity. Failure to do so may lead to policies that are broadly applicable but insufficiently targeted [101]. As a result, we develop interpretable ML models to explore the ways observed variables influence the low-carbon travel intentions of different traveler groups. Combining the questionnaire data from this study with official statistical data from Shanghai, we also compare the size and willingness of different population groups to support the policy priorities decisions of policymakers.

(1)

People of different age groups

Some previous research declares that older residents tend to show a higher intention to choose low-carbon travel modes [102]. However, the SHAP analysis in Figure 8 reveals that respondents in their 20s exhibit the highest intention for low-carbon travel, while middle-aged groups show the lowest, even lower than older respondents. Tamim Kashifi et al. [103] also obtained similar findings. Figure 8 also illustrates an inflection point at the age of 32. The official statistics from the Shanghai Civil Affairs Bureau validate the credibility of our research conclusions. The statistical results show that the average marriage registration age in Shanghai is 35.3 for men and 33.3 for women, and the average age for women with Shanghai household registration having their first child is 31.7 [104], closely aligning with the inflection point identified in our research.

Young individuals without children tend to demonstrate a strong willingness to engage in low-carbon behaviors. This may be attributed to their high travel frequency, diverse travel purposes, and relatively flexible schedules, with convenience typically being their primary concern. In Shanghai, current low-carbon transportation modes—such as subways and shared bicycles—can almost cover areas where young people frequently gather and are well-suited to the travel patterns of young people, who frequently use social media platforms like WeChat, Red Booklet, and TikTok. The government can leverage these platforms to promote low-carbon travel through activities such as topic sharing and offline check-in events to increase their engagement. Participants with high levels of involvement can be rewarded accordingly. Additionally, this group generally exhibits a fashion-conscious consumption preference [105]. Therefore, the government could collaborate with popular brands to launch low-carbon-related products, such as eco-friendly water bottles and canvas bags, to help young people cultivate a positive attitude toward low-carbon travel. These methods have been widely applied in other fields [106,107,108].

In contrast, individuals aged 30–40, who are more likely to be parents, tend to experience a decrease in travel frequency and exhibit more regularized, child-related travel patterns, such as school drop-offs and pick-ups, as well as participation in extracurricular activities. Although their destinations are fixed, they tend to be scattered, making it challenging for existing low-carbon transport services to adequately meet such needs. As a result, they often prefer private vehicles due to the need for safety, convenience, and flexibility, especially for trips among home, work, and school. Existing research indicates that households with children are more likely to own and utilize private vehicles [109,110]. To encourage this group to transition from gasoline-powered vehicles to clean energy vehicles, the government should implement effective measures to address high usage costs and expensive parking, such as purchase incentives, reduced parking fees, insurance, and annual inspection fees. Additionally, increasing the number of charging stations and designating dedicated EV charging areas in residential communities can significantly enhance the convenience of clean energy vehicle ownership [111]. From a manufacturer perspective, it is crucial to prioritize safety and usability features—such as secure power systems, integrated child safety seats, and advanced urban driving assistance technologies––particularly in congested or high-risk traffic environments. Furthermore, community-based family-oriented low-carbon travel campaigns could serve as supportive initiatives to promote sustainable travel behaviors.

It should be noted that, although the clean energy vehicle industry in China has expanded rapidly in recent years, it remains relatively new compared to traditional gasoline vehicles [112]. As a result, most family-owned private vehicles remain gasoline-powered. Figure 9a,b show that younger people are more likely to own electric vehicles, a trend that can likely be attributed to their greater openness to new technologies. Another plausible explanation derived from the dependency plots is that many gasoline vehicles are still within their normal service life and do not need immediate replacement. To investigate this, we further analyzed Figure 9c and found a group of respondents with children who have high fuel costs but a strong intention to shift toward low-carbon travel. These individuals are likely to transition to low-carbon mobility behaviors in the near future when they are faced with the need to replace their vehicles.

(2)

People from different income levels

In our study, most respondents have a monthly income between CNY 8000 and 20,000, which is consistent with the income distribution of the population in Shanghai [91]. The findings from Sahari, et al. [113] suggest that high-income individuals generally have the greatest potential for carbon emission reduction, which means that they are the most likely to transit to low-carbon travel in the future. Figure 10a,b, indeed, shows that their intention toward low-carbon travel is generally low. Theoretically, their intentions could be enhanced through commercial promotion, policy incentives, or other means, thereby fully tapping into their potential for carbon emission reduction.

However, our study further shows that not all high-income individuals possess high emission reduction potential. Some bear almost no fuel expenses, while others have low trip frequency. These individuals have limited travel demand, and therefore limited potential for emission reduction. Hence, we argue that only high-income individuals with a high monthly fuel cost or frequent travel exhibit substantial carbon emission reduction potential. Targeted policies are thus necessary to guide these groups toward low-carbon travel. Given their financial capacity, sophisticated and personalized low-carbon travel services may be tailored to meet their specific needs. For instance, encouraging automakers to develop premium electric vehicles in line with their preferences for high-quality, eco-friendly alternatives could be beneficial. At present, countries such as China, Singapore, and Thailand have established carbon exchanges that allow enterprises to trade carbon emission allowances [114,115,116,117]. With China’s interim regulations for carbon emissions trading management, local governments and enterprises can develop financial mechanisms to incentivize high-income individuals to participate in low-carbon travel. For example, clean energy vehicle manufacturers and buyers may receive additional carbon emission allowances based on annual sales or purchases through these platforms. Moreover, in addition to being traded as financial instruments, carbon emission allowances can also serve as assets for purchasing clean energy vehicles.

For individuals with a monthly income below 12,000 CNY, the government may consider implementing financial incentives, such as subsidies or fare discounts, for public transportation use. Partnerships with ride-hailing companies could further support this by offering additional discounts for trips that originate or terminate at bus or subway stations. Such collaborations would make public transit more attractive, especially during peak hours, and encourage a shift from private vehicle use to more sustainable modes of transport. A combined discount package, which is more cost-effective than buying a fuel card and a public transportation card separately, could further incentivize this transition economically.

Figure 10c focuses on individuals for whom fuel costs constitute a significant proportion of their monthly income, yet whose interest in low-carbon travel remains low. This may result from their livelihoods and income being closely tied to gasoline vehicles, such as long-distance truck drivers or online ride-hailing drivers. Additional rewards should be provided to ride-hailing drivers who complete orders using clean energy vehicles. Intelligent algorithms can recommend low-carbon routes (e.g., avoiding congestion and idle periods) and estimate energy consumption. Ride-hailing platforms could display drivers’ real-time carbon emissions and savings, thereby increasing their environmental awareness.

(3)

Policy priorities

For the government, initiatives with short-term, visible results are more likely to gain public support. Thus, it is crucial to prioritize policies that address the needs of the majority. Based on the survey statistics from this study, combined with the groups identified through SHAP analysis, we found that the middle-aged and young population (23–45 years old), people with medium to low income (lower than CNY 12,000), and gasoline car owners account for 75%, 68%, and 62% of the respondents, respectively. Consequently, these groups should be given priority in policy design and implementation.

Since EPP is more effective in promoting individuals’ intention to low-carbon travel, according to the analysis results based on the TPB-TAM framework, the following policy measures should be prioritized by relevant government departments:

(1)

Promotions on popular media platforms and co-branded merchandise (targeting young people),

(2)

Subsidies on the purchase, parking, insurance, and annual inspection of clean energy vehicles (targeting middle-aged people and individuals with medium-to-low income who currently own gasoline cars),

(3)

Enhancing the safety performance and quality inspection of clean energy vehicles (targeting childbearing middle-aged gasoline car owners),

(4)

Subsidies for public transportation (targeting individuals with medium-to-low income).

In addition, appropriate restrictive policies can also produce noticeable short-term effects. Therefore, measures such as moderately increasing fuel prices or gasoline vehicle purchase taxes could also be considered as temporary priority policies.

6. Conclusions

6.1. Summary and Implications

Promoting low-carbon travel modes is of great significance for the transportation sector to achieve China’s dual carbon goals. This study contributes to the literature by offering a novel theoretical framework combining the Theory of Planned Behavior (TPB) and the Technology Acceptance Model (TAM) to explain the psychological mechanisms underpinning individuals’ willingness to adopt low-carbon travel.

Attitudes toward low-carbon travel (ATT) are identified as the most central construct, not only directly influencing low-carbon travel intentions (LI), but also mediating the effects of both subjective norms (SN) and perceived behavioral control (PBC). This finding underscores the importance of shaping positive low-carbon attitudes through perceived capability and social means.

Moreover, built environment perception (BEP), functioning as perceived ease of use (PEOU) in this research, significantly influences both ATT and perceived usefulness (PU), which encompasses encouraging policy perception (EPP) and restrictive policy perception (RPP). Notably, while EPP directly affects LI, RPP does not, suggesting that positive incentives may be more effective in fostering voluntary low-carbon travel intentions than restrictive measures.

Methodologically, this study employs both structural equation modeling (SEM) and a SHAP-based machine learning approach to explore the influence mechanisms of various constructs on travel intentions. These techniques not only confirm the robustness of the integrated TPB-TAM framework, but also reveal the differing impact levels of variables across traveler groups. Notably, variables such as age, travel frequency, income, number of children, and car-related costs are shown to significantly influence low-carbon travel intentions, calling for more nuanced and data-driven policy design tailored to diverse user groups.

From a policy perspective, these findings emphasize the value of incentive-based, attitude-oriented interventions. Improving public transport accessibility, developing pedestrian- and cyclist-friendly infrastructure, and promoting low-carbon travel knowledge at the community level could be key steps in nurturing a culture of sustainable travel.

6.2. Limitations and Future Research

Several limitations should be considered when interpreting the findings of this study. First, the concept of “low-carbon travel” in this study encompasses various transportation modes, including public transportation, non-motorized travel options, shared mobility services, and clean energy vehicles. While all these modes contribute to reducing carbon emissions, the barriers to promoting these modes may be different. In recent years, the increasing share of Sport Utility Vehicles (SUVs) in the global passenger vehicle market also deserves attention [118,119]. Second, the research used online questionnaires for data collection. Due to the characteristics of the online platform audience, the proportion of elderly respondents and the distribution of occupations and educational backgrounds seemed to be limited. Finally, the empirical data originate from Shanghai—a megacity characterized by high population density, elevated car ownership rates, and unique local policies (e.g., license plate auction systems). While the findings provide insights for similar metropolises, their applicability to smaller cities or regions with divergent transportation infrastructures requires rigorous validation. Future research should consider the impacts of larger vehicle types such as SUVs on carbon emissions and energy consumption and explore the differences between various low-carbon travel modes through more detailed categorization and analysis. Additionally, efforts should be made to expand the sample to include a wider range of respondents, ensuring a more representative dataset for comprehensive analysis. Policymakers should also exercise caution by integrating context-specific adjustments when implementing recommendations from this research.