How Do Information Interventions Influence Walking and Cycling Behavior?

Abstract

In the context of promoting sustainable mobility, walking and cycling have been widely recognized for their environmental and health benefits. However, a notable gap often exists between residents’ motivation to engage in these modes and their actual behavior. This study focuses on this motivation–behavior discrepancy and explores how heterogeneous information interventions—within the constraints of the existing built environment—can effectively influence residents’ travel psychology and behavior. Drawing on Protection Motivation Theory, this study aims to uncover the psychological mechanisms behind travel-mode choices and quantify the relative impacts of different types of information interventions. A travel survey was conducted in Yangzhou, China, collecting data from 1052 residents. Cluster analysis was performed using travel psychology data to categorize travel motivations and examine their alignment with actual travel behavior. A random forest model was then employed to assess the effects of individual attributes, travel characteristics, and information intervention attributes on the choice of walking and cycling. The results reveal a significant motivation–behavior gap: while 76% of surveyed residents expressed motivation to walk or cycle, only 30% actually adopted these modes. Based on this, further research shows that informational attributes exhibit a stronger effect in terms of promoting walking and cycling behavior compared to individual attributes and travel characteristics. Among these, health-related information demonstrates the maximum efficacy in areas with well-developed infrastructure. Specifically, health-related information has a greater impact on cycling (21.4%), while environmental information exerts a stronger influence on walking (7.31%). These findings suggest that leveraging information to promote walking and cycling should be more targeted.

1. Introduction

Rapid economic development, the popularization of automobiles, urban expansion, and other factors have made motor vehicle travel the dominant and fastest growing mode of travel [1,2]. This transformation has brought about many challenges (congestion, pollution, and abnormal climate) [3,4]. In the context of sustainable urban development, walking and cycling have become cornerstones of sustainable urban mobility, providing effective solutions to challenges such as traffic congestion, air pollution, and public health concerns [5,6,7]. Policy document [8] advocates for the establishment of a travel system integrating “rail transit + public transport + non-motorized modes”, aiming to guide residents toward walking and cycling and contribute to the realization of China’s “dual carbon” goals. In addition to alleviating traffic congestion and emissions, walking and cycling can also bring about profound benefits with respect to health and well-being. Empirical research has confirmed that regular walking or cycling can promote physical health (such as by reducing the risk of cardiovascular diseases and controlling weight) and mental health (such as by relieving stress and improving mood) and, at the same time, enhance social cohesion by increasing neighborhood interaction [5,9]. In the context of rapid urbanization in China, a sedentary lifestyle poses a threat to health. Therefore, promoting walking and cycling is of vital importance.

Given their close interaction with built infrastructure—such as continuous sidewalks, segregated bike lanes, and pedestrian crossing [7,10]—walking and cycling are often used as practical indicators of sustainable mobility and urban vitality. A substantial body of research has established that these infrastructure elements fundamentally enable walking and cycling by enhancing accessibility, safety, and comfort [11,12,13]. Specifically, street network density and land-use mix create opportunities for active travel, while the accessibility and quality of pedestrian/cycling facilities affect behavioral adoption [14]. However, even in areas with favorable built environments, a significant discrepancy remains between individuals’ motivation to walk or cycle and their actual behavior [15,16,17]. This paradox indicates that improving infrastructure alone may not be sufficient to drive behavioral change. Instead, attention must also be paid to the internal decision-making processes that mediate this transformation [16,17].

To gain deeper insight into this motivation–behavior discrepancy, it is essential to consider the psychological mechanisms that guide individual travel decisions. In particular, how residents perceive risks and evaluate their coping capacity plays a crucial role. To address this need, this study adopts Protection Motivation Theory (PMT) [18], a psychological model that has been widely applied to explain health-related and environmentally responsible behaviors. PMT comprises two core dimensions—threat appraisal and coping appraisal, which, together, offer valuable insights into the cognitive mechanisms that underlie the transition from motivation to actual travel behavior. As a form of soft intervention, information can influence travel decisions by altering individuals’ risk perceptions and beliefs about behavioral efficacy [19,20]. Recent studies have shown that targeted information interventions, such as those emphasizing the health benefits of cycling, can effectively promote the adoption of walking and cycling. However, two critical limitations persist in existing research: first, there is an insufficient understanding of the mechanisms underlying the motivation–behavior transformation; second, most studies examine built environment factors and psychological interventions in isolation, failing to uncover their potential synergistic effects.

To bridge these gaps, this study concentrates on three key research questions: (1) How do information interventions influence the transformation of walking and cycling from motivation to behavior within the context of the existing built environment? (2) How do heterogeneous information interventions (e.g., health-related, environmental, and policy-related interventions) differ in their effectiveness? (3) How can information intervention strategies compensate for deficiencies in the built environment to increase the proportion of walking and cycling? To address these questions, this study takes Yangzhou, China, as a case study. The city is characterized by features that are conducive to walking and cycling, such as flat terrain, a dense street network in the downtown, and mixed land use. However, it still exhibits a high proportion of motorized travel. This research incorporates variations in the built environment through survey variables such as travel distance. It constructs an analytical framework based on PMT; employs K-means clustering to identify travel motivation types; and applies a random forest model to quantify the effects of individual attributes, travel characteristics, and information interventions on travel behavior. The conceptual framework is presented in Figure 1.

This study is organized as follows. Section 2 reviews literature related to travel behavior and information interventions. Section 3 introduces the data sources, descriptive statistics, and modeling methods. Section 4 presents and discusses the empirical results. Section 5 concludes with key findings and implications.

2. Literature Review

2.1. Travel Behavior and Built Environment

Extensive research has been conducted on travel behavior, primarily focusing on dimensions such as mode choice, travel duration, destination selection, and route choice. In this study, travel behavior is primarily reflected through mode choice. A substantial body of evidence indicates that travel behavior is influenced by multiple factors, including individual attributes, travel characteristics, the built environment, and subjective perceptions [21,22,23]. The built environment, in particular, has received significant attention as a foundational factor. The 3Ds framework—Density, Diversity, and Design—proposed by Cervero and Kockelman [24] provided the foundation for examining the impact of the built environment on travel behavior. This framework was subsequently expanded into the 5Ds model, incorporating Distance to Transit and Destination Accessibility [25]. Widely adopted in empirical research, this model elucidates the relationship between the built environment and travel behavior. For instance, both population and building density have been shown to significantly increase cycling volumes [26,27]. Factors such as land-use mix and the density of transit stations are negatively associated with automobile commuting [28], while the accessibility of facilities for shopping, daily services, and recreational activities significantly promotes cycling [26,29].

As the field has evolved, scholars have increasingly recognized the dynamic and heterogeneous nature of built environment effects. Zhou et al., utilizing two decades of longitudinal data, demonstrated that the direction and strength of these effects vary over time [30]. Panel data from Beijing suggests that changes in residents’ travel attitudes following relocation may exert a stronger influence than the built environment itself [31]. Moreover, the impact of the built environment varies across demographic groups and travel purposes. For example, family members may respond differently to environmental factors [32], and shopping travels tend to be more sensitive to the built environment than work-related travel [33]. In a systematic review, Blitz and Lanzendorf [34,35] noted that the influence of the built environment on walking and cycling exhibits a dose–response relationship. Similarly, Lu et al. [36] found that density-related factors significantly promote such travel by altering temporal and spatial activity patterns. Although the 5Ds framework is well established, the interactions among built environment variables [14] and the contextual variability across different urban settings require further exploration. These gaps present theoretical opportunities for this study to enhance current understanding.

2.2. The Impact of Information Interventions

Information interventions were initially employed in psychological research [37,38,39] and have since been categorized as a form of soft transportation intervention that influences travel behavior by altering individuals’ cognitive perceptions [19,20]. As public awareness of the health benefits [5,9], air pollution reduction [40], and improved traffic safety associated with walking and cycling has grown, research on the role of information interventions in promoting these modes has expanded significantly [41,42]. Most existing studies utilize experimental or quasi-experimental designs to compare travel-mode choices before and after intervention, consistently finding that information interventions significantly encourage individuals across various demographic groups to adopt walking or cycling [17,19]. However, the effectiveness of these interventions is influenced by multiple factors. Specifically, multi-type interventions targeting non-student populations and intensive short-term campaigns have been found to generate the most significant results [43]. It is worth noting that while Stark et al. [44], drawing on the Theory of Planned Behavior, confirmed that information interventions can enhance adolescents’ attitudes and perceived behavioral control regarding walking and cycling, the actual behavioral change may fall short of expectations.

In terms of content, studies have shown that heterogeneous information exhibits varying degrees of effectiveness. For example, information from social networks can directly influence travel behavior, including destination selection, mode choice, and route planning, and may even trigger mimicry or impulsive trips [45]. Targeted interventions also show promise: Mackett found that customized information for individuals with travel-related anxiety significantly increased their willingness to travel [46], while housing and transportation accessibility information helped university students make more rational residential choices, ultimately reducing vehicle miles traveled by up to 68% over the long term [47]. These findings highlight the potential of information interventions, but their effect often requires the support of a policy combination [20]. Future research should pay greater attention to the built environment and refine the classification of information types to enhance the precision and effectiveness of intervention strategies.

In summary, existing studies have provided valuable insights into the relationship between travel behavior and the built environment, as well as the significance of information interventions. However, two critical gaps remain: (1) insufficient understanding of the mechanisms underlying the motivation–behavior transformation process and (2) a tendency to examine the built environment in isolation from the information interventions, making it difficult to explain behavioral heterogeneity under similar infrastructural conditions. To bridge these gaps, this study innovatively integrates the two perspectives. Grounded in Protection Motivation Theory, it characterizes the foundational role of the built environment using variables such as travel distance, identifies motivational types through K-means clustering, and employs a random forest model to quantify the effects of heterogeneous information interventions on travel behavior.

3. Data and Methods

3.1. Data

Yangzhou is a medium-sized city located in eastern China, within Jiangsu Province, adjacent to the Yangtze River and near Nanjing, the provincial capital. It is an important tourist city with strong economic performance and urban vitality. According to the 2023 Yangzhou Statistical Yearbook, the city had a resident population of approximately 4.59 million, received more than 103 million tourist visits throughout the year, and recorded a regional GDP of CNY 742.3 billion. The downtown of Yangzhou is characterized by flat terrain, a high-density road network, and mixed land use, all of which constitute a built environment favorable for walking and cycling. Nevertheless, the proportion of motorized travel remains disproportionately high. Motorized trips account for over 34% of total travel, and the proportion of private car travel has increased by 15% year on year, indicating a growing reliance on motor vehicles despite the city’s favorable conditions for walking and cycling.

This study focuses on residents in the downtown of Yangzhou (see Figure 2) and includes a travel survey. As shown in

Table 1. Classification of survey questionnaire items.

Category	Characteristic Indicators
Individual Attributes	Gender, Age, Car Ownership, Disposable Income
Travel Characteristics	Travel Frequency, Distance, Duration, Weekly Walking and Cycling Frequency
Travel Psychological Attributes	Perceived Severity, Vulnerability, Rewards, Self-Efficacy, Response Efficacy, Response Costs
Information Intervention Attributes	Environmental Information, Health Information, Policy Guidance Information

, the questionnaire design includes four sections: individual attributes, travel characteristics, travel psychology attributes, and information intervention attributes. Travel psychology, as a latent variable, is represented by the six elements of the Protection Motivation Theory, with two questions assigned to each element. On the basis of Geng’s research [48], we classify the information intervention attributes into three categories according to the research purpose and Chinese background, with five questions corresponding to each category, as detailed in

Table 2. Information intervention section of the questionnaire.

Intervention Information Type	No.	Corresponding Choice
Environmental Information	1	Motor vehicle emissions have become one of the main sources of air pollution in many cities in China. If choosing green and low-carbon transportation can save energy and reduce pollution, I am willing to choose non-motorized transport.
	2	If road traffic noise is a major component of urban environmental noise, and the traffic noise mainly comes from motor vehicles, I am willing to choose non-motorized transport to reduce noise pollution.
	3	If open-air parking lots alter the direction and speed of urban airflow, exacerbating the urban heat island effect, I am willing to choose non-motorized transport.
	4	If a person changes five trips of no more than 2 km per week from driving to walking, it can reduce the carbon footprint by 86 kg annually. If cycling to work four days a week for 8 km per day, it can reduce car usage by 3220 km a year, equivalent to saving 380 L of gasoline and reducing 750 kg of CO2. This makes me more inclined to choose walking or cycling as my mode of transportation.
	5	If 20% to 35% of the PM2.5 emissions in large cities come from vehicle exhaust, especially during heavy smog periods when vehicle exhaust accounts for up to 47% of air pollution, I am willing to choose non-motorized transport to reduce pollution.
Health Information	1	If regular physical exercise can lower mortality, cardiovascular diseases, coronary heart disease, and stroke, as well as reduce the risk of cancer and type 2 diabetes, I am willing to choose walking or cycling.
	2	If moderate exercise can shorten sleep onset time, extend sleep duration, improve sleep quality, and enhance quality of life, I am willing to choose walking or cycling.
	3	If cycling helps to strengthen the lower body muscles, enhances overall endurance, and improves cardiovascular function, effectively preventing brain aging, I would be more willing to choose cycling.
	4	If walking is the simplest form of exercise, less prone to injury, and can relieve neuromuscular tension, enhance the secretion function of digestive glands, and promote regular intestinal movement, I am willing to choose walking.
	5	If walking or cycling in open environments with good air circulation can reduce the transmission of infectious diseases, I am willing to choose walking or cycling.
Policy Information	1	When I see the policy that the government will increase the number of shared bicycles and shared electric vehicles, I will be more willing to choose cycling.
	2	When I see the policy that the government will add green pedestrian pathways, optimize the design of sidewalks, and improve facility provisions to ensure travel safety, I will be more willing to choose walking.
	3	When I see the policy that the government will increase the coverage of the metro network and the frequency of bus stops, I will be more willing to walk or cycle for the last mile.
	4	When I see the policy that the city center will adopt a congestion charging policy, i.e., charge users for road access in specific areas during peak traffic hours, I will be more willing to walk or cycle.
	5	When I see the policy that the government will provide subsidies for walking or cycling commuters, I will be more willing to choose walking or cycling.

. Both travel psychology and information intervention attributes are assessed using a 5-point Likert scale (1 = strongly agree, 5 = strongly disagree), enabling quantitative evaluation.

A combination of stratified and convenience sampling methods was employed to collect data from residents of downtown Yangzhou. The questionnaire was distributed both through on-site surveys conducted at public locations such as parks and community centers and through online channels using neighborhood-based social media platforms. The target population consisted of adult residents familiar with their daily travel routines. To enhance representativeness, deliberate efforts were made to include participants across different genders, age groups, and income levels.

The sample size was determined using the standard formula for minimum sample size in probability sampling. To achieve a 95% confidence level and control the margin of error within 6%, a minimum of approximately 1015 valid samples is required.

$$n={\displaystyle \frac{Z^2{S}^2}{d^2}}$$

(1)

where $n$ is the required minimum sample size, $Z=1.96$ corresponds to a 95% confidence level, $S$ is the estimated standard deviation, and $d$ is the allowable margin of error.

Prior to model construction, systematic data preprocessing was carried out to improve data quality and enhance the robustness of the modeling process. First, invalid or incomplete questionnaires were removed, including those with unanswered key questions, inconsistent responses, or submission durations significantly below a reasonable threshold. Second, IP addresses were examined to eliminate responses originating from outside the target study area. Third, outlier detection was conducted for key variables such as travel distance and travel duration, and responses containing extreme values were excluded. Finally, categorical and ordinal variables were numerically encoded to produce input features suitable for model analysis. These preprocessing steps ensured the accuracy, consistency, and usability of the input data, laying a solid foundation for subsequent modeling tasks.

A total of 1200 questionnaires were distributed, and 1137 were returned. After pre-processing, 1052 valid questionnaires were used for analysis. While the final sample size of this study meets statistical requirements, potential selection bias may exist due to the voluntary nature of survey participation, which could favor individuals more willing to respond. Furthermore, as the data were collected from a single medium-sized Chinese city, the generalizability of the findings to other urban contexts may be limited.

3.2. Descriptive Statistics

As mentioned above, 1052 valid questionnaires were used for analysis. Among the valid samples, 530 were male and 503 were female, resulting in a nearly balanced gender ratio of 1:1. The majority of respondents were aged between 30 and 49 years, accounting for 46.3% of the sample. Most of them (77.8%) owned a private car. The largest proportion of respondents (37.8%) had a monthly household disposable income of between RMB 5000 and 10,000.

Table 3. Statistical characteristics of travel situations.

Variable (Characteristic Indicator)		Description	Value	Sample Size	Percentage
Daily Travel Mode	Mode Choice	Car	1	401	38.1%
		Public Transport	2	393	37.4%
		Cycling	3	218	20.7%
		Walking	4	40	3.8%
	Travel Distance	0–1 km	1	84	8.0%
		1–3 km	2	215	20.4%
		4–6 km	3	233	22.1%
		7–10 km	4	183	17.4%
		11–15 km	5	141	13.4%
		16–20 km	6	123	11.7%
		>20 km	7	73	6.9%
	Weekly Frequency	1–3 times	1	242	23.0%
		4–6 times	2	447	42.5%
		7–10 times	3	226	21.5%
		>10 times	4	137	13.0%
	Single Trip Duration	0–5 min	1	77	7.3%
		5–20 min	2	282	26.8%
		20–40 min	3	286	27.2%
		40–60 min	4	209	19.8%
		>60 min	5	198	18.8%
Weekly Walking and Cycling Frequency		0 times	1	127	12.1%
		1–3 times	2	210	20.0%
		4–6 times	3	310	29.5%
		7–10 times	4	194	18.4%
		>10 times	5	211	20.1%

presents the statistical characteristics of travel patterns. The survey results show that the majority of respondents (60%) primarily travel within a 1–10 km range, a distance favorable for walking and cycling. However, only about a quarter of respondents (24.5%) use cycling or walking for daily travel, indicating that the potential of walking and cycling remains underutilized.

The quality of the survey data was examined from both reliability and validity perspectives. Reliability was assessed using Cronbach’s α coefficient, which indicated that the reliability coefficient for all groups exceeded 0.8, suggesting good internal consistency and high overall reliability of the questionnaire. Validity was evaluated using Bartlett’s Test of Sphericity and the KMO test. The results showed that Bartlett’s Test values were all less than 0.05, and the KMO values exceeded 0.8, indicating strong construct validity and confirming that the questionnaire passed the quality assessment.

3.3. K-Means Clustering

To reduce computational complexity, enhance the practical significance of the classification, and address the limitations of hierarchical clustering methods, this study adopts the K-means clustering algorithm [49,50]. The 12 survey items related to travel psychology are used as variables to cluster respondents’ travel motivations and analyze the discrepancy between motivation and behavior. As an unsupervised learning method, K-means clustering divides a given sample set into K clusters based on similarity. Samples within the same cluster exhibit high internal similarity, whereas those from different clusters show low inter-cluster similarity.

3.4. Random Forest Model

The resident travel survey data exhibit complex distribution characteristics, involving multiple discrete variables, ordinal scales, and considerable noise. While machine learning models such as Support Vector Machines (SVM) and Extreme Gradient Boosting (XGBoost) are widely used and perform well in noisy settings, they present certain limitations in the current context. SVM typically requires standardized and numerically encoded inputs, but our dataset includes a wide range of mixed-type variables, which would necessitate extensive preprocessing. In addition, SVM tends to be computationally inefficient when handling large samples with high-dimensional features and lacks interpretability—offering neither variable importance rankings nor tools for visualizing non-linear effects [51,52]. XGBoost, though powerful in predictive accuracy, demands meticulous parameter tuning and provides less transparent interpretations [53,54]. In contrast, the Random Forest (RF) model is well-suited for mixed-type data, requires minimal preprocessing, and offers intuitive measures of variable importance and non-linear effects [55,56,57]. These advantages are particularly valuable for identifying the key determinants of travel behavior and for understanding how different types of information interventions influence walking and cycling decisions. Therefore, RF is adopted in this study to quantify the influence of individual attributes, travel characteristics, and intervention attributes on travel behavior.

To effectively model the influence of these factors, the following independent and dependent variables are defined. Specifically, the independent variables include individual attributes, travel characteristics, and information intervention attributes. Individual attributes consist of gender, age, monthly disposable income, and car ownership. Travel characteristic variables comprise travel frequency, travel duration, walking and cycling frequency, and travel distance. Information intervention variables include environmental information, health information, and policy guidance information. The dependent variable is residents’ travel behavior, represented by the primary travel mode reported in the questionnaire.

Based on this variable framework, an RF model is constructed using MATLAB R2023b, with the dataset partitioned into subset of 70% for training and 30% for testing. To prevent potential multicollinearity among variables from distortion of the feature importance ranking, we further conducted a variance inflation factor (VIF) analysis. After parameter tuning, the model is configured with the following settings: the number of trees (n_estimators) is set to 800, the Gini index is selected as the criterion for node splitting, the minimum number of samples required to split a node is set to 5, the minimum number of samples for leaf nodes is set to 8, and the maximum tree depth is restricted to 30.

4. Results and Discussion

4.1. Analysis of K-Means Results

The algorithm was implemented using SPSS 26.0, and the respondents were classified into three groups based on the intensity of their walking and cycling motivations: strong, neutral, and weak. Cluster 3 represents individuals with the strongest motivation, with the cluster center concentrated around a score of 2 on the Likert scale. Cluster 1 represents a neutral level of motivation, with the cluster center concentrated around a score of 3. Cluster 2 represents individuals with weak motivation, with the final cluster center positioned around a score of 4 on the Likert scale. The t-SNE (t-distributed Stochastic Neighbor Embedding) algorithm was employed to reduce the dimensionality and visualize the clustering results, enabling a clear observation of the relative relationships and clustering effectiveness. As shown in Figure 3, different colors and markers represent different clusters, with the horizontal and vertical axes corresponding to the two principal dimensions of the t-SNE-reduced data. These two dimensions capture the major variability in the data. Although they do not directly correspond to any specific features of the original data, they provide a clear visual representation of the clustering structure, confirming the existence of three distinct clusters.

After verification, the 1052 respondents were divided into three categories: 542 in the “strong walking and cycling motivation” cluster (Cluster 3), 231 in the “neutral motivation” cluster (Cluster 1), and 279 in the “weak motivation” cluster (Cluster 2). The mean values of individual attributes and travel characteristics for each cluster are presented in

Table 4. Statistics of individual attributes and travel characteristics after clustering.

Feature	Mean
	Cluster 1	Cluster 2	Cluster 3
Gender	1.51	1.47	1.50
Age	2.39	3.70	3.06
Ownership of private car in the household	1.28	1.29	1.16
Monthly disposable income of the household	2.77	3.14	2.94
Daily travel mode	2.07	1.80	1.88
Travel distance for daily travel mode	3.57	3.75	3.74
Weekly frequency of daily travel mode	2.13	2.48	2.18
Single-trip duration for daily travel mode	3.05	2.78	3.40
Weekly frequency of walking and cycling	2.91	3.27	3.18

.

The results show that 76% of the respondents displayed a positive or neutral attitude towards walking and cycling (Clusters 1 and 3), indicating generally supportive travel motivations. However, the primary travel modes for all three clusters were automobiles and public transportation, with only 30% of respondents engaging in walking or cycling. This reveals a significant discrepancy between residents’ travel motivation and actual behavior. Even for those who are neutral or supportive, various factors may prevent the translation of motivation into behavior. Therefore, it is essential to investigate the factors influencing residents’ walking and cycling behavior to bridge the discrepancy between motivation and behavior.

4.2. Factors Affecting Walking and Cycling Behavior

After testing on the validation dataset, the RF model achieved a prediction accuracy of 74.02% and a precision rate of 71.89%. These results confirm the effectiveness of the RF model in capturing the complex relationships between input features and travel behavior. In addition, the multicollinearity diagnosis based on variance inflation factors (VIFs) indicated that all variables had VIF values below 2. This suggests weak inter-variable correlations and ensures that the variable importance rankings derived from the model are reliable.

Based on these results, the study analyzed the impact of individual attributes, travel characteristics, and information intervention attributes on travel-mode choice. Using the factor with the highest coefficient (Health Information 2) as a reference, its influence was set to 100.00, and the other factors were adjusted proportionally, as shown in

Table 5. Impact of different characteristics on residents’ travel behavior.

Feature Variable	Influence Degree	Feature Variable	Influence Degree
Health Information 2	100	Monthly Income	38.6
Age	92.37	Environmental Information 4	37.66
Travel Distance	78.1	Policy Information 5	33.55
Health Information 1	67.75	Environmental Information 2	33.33
Health Information 3	67.15	Policy Information 3	23.45
Health Information 5	66.37	Environmental Information 1	22.97
Travel Frequency	65.15	Policy Information 4	20.91
Health Information 4	53.69	Gender	20.62
Policy Information 1	48.53	Environmental Information 5	18.52
Travel Duration	47.31	Policy Information 2	17.13
Environmental Information 3	42.09	Walking and Cycling Frequency	13.28
Ownership of Private Car	40.64

. Compared to individual attributes and travel characteristics, information intervention factors have a more significant impact on travel behavior. Among the three types of intervention information, health-related information has the greatest influence on travel behavior, followed by policy-related information. Environmental information also affects residents’ travel behavior, but its isolated impact is insufficient, requiring the support of other factors. Furthermore, travel characteristics also have a significant influence on travel behavior, contributing to 50.96% of the cumulative importance. The prominence of health information in our study suggests that residents prioritize concrete personal benefits like physical well-being over environmental concerns. This aligns with the “coping appraisal” mechanism in Protection Motivation Theory, which suggests that individuals are more likely to adopt behaviors they perceive as personally beneficial and achievable.

As shown in

Table 6. Classification of impact levels of characteristic variables.

Travel Mode	Independent Variable	Importance	Attribute
Car	Age	65.03%	Individual Attribute
	Health Information 3	59.99%	Intervention Information Attribute
	Health Information 2	59.71%	Intervention Information Attribute
	Travel Distance	57.33%	Travel Attribute
Public Transport	Age	55.20%	Individual Attribute
	Policy Information 4	41.63%	Intervention Information Attribute
	Travel Distance	41.27%	Travel Attribute
	Policy Information 2	38.45%	Intervention Information Attribute
Cycling	Health Information 5	21.40%	Intervention Information Attribute
	Health Information 1	20.81%	Intervention Information Attribute
	Health Information 3	18.79%	Intervention Information Attribute
	Travel Duration	17.11%	Travel Attribute
Walking	Age	8.63%	Individual Attribute
	Travel Distance	8.59%	Travel Attribute
	Environmental Information 3	7.31%	Intervention Information Attribute
	Health Information 3	7.66%	Intervention Information Attribute

, the top four influencing factors for each travel mode were listed to analyze the mechanisms through which various variables impact different travel modes. Partial dependence plots were created to examine the nonlinear influence of different independent variables on travel-mode choices. Cycling and walking were used as examples, and a single independent variable was selected for illustration, as shown in Figure 4. Since all independent variables in this study were represented as discrete variables, parallel trends can be observed in the plots.

Age was found to be the most important factor influencing residents’ choice of automobiles and public transportation. As age increases, older individuals are more likely to choose automobiles, and the probability of choosing public transportation decreases. However, this trend reverses after retirement, as declining driving ability leads many older adults to shift back to public transport. Similar findings have been reported in previous studies [58,59]. Furthermore, fare reduction policies for retired seniors have been identified as an important factor contributing to the increase in public transport use. Health information and policy information were found to be the most significant types of intervention information influencing automobile and public transportation choices, respectively.

For cycling behavior, the top four factors are Health Information 5, Health Information 1, Health Information 3, and travel duration. Increased health information intervention significantly raises the probability of cycling, indicating residents’ sensitivity to disease risk reduction. Travel duration has an influence of 17.11% on cycling behavior, with the probability of choosing cycling decreasing as travel duration increases. This finding aligns with existing studies suggesting that longer travel distances increase perceived time costs, thereby reducing the probability of choosing walking or cycling as travel modes [60,61]. However, once the travel duration reaches 40 min, the probability of choosing cycling gradually increases again, driven primarily by the motivation for physical exercise. This pattern is not inconsistent with the conclusions presented above, as health-conscious individuals may be more inclined to walk or cycle even over longer durations, especially when physical activity itself is perceived as a travel benefit.

For walking behavior, the most influential factor is age. Walking probability declines with age. Travel distance is the second key factor. As travel distance increases, the probability of walking initially decreases but then increases at longer distances. Similar to cycling, this increase could be related to the motivation for physical exercise. This is in line with previous findings indicating that there exists a threshold effect in how distance influences individuals’ willingness to walk [62]. Among the intervention information, Environmental Information 3 is the most critical factor, indicating residents’ high sensitivity to the urban heat island effect. Walking probability increases with increases in intervention intensity but declines beyond a threshold of 0.85. This indicates that while environmental information can influence walking behavior, its impact is limited when implemented in isolation. This finding further supports previous conclusions that although environmental messaging can enhance awareness, its behavioral impact is often constrained without supportive physical conditions, such as shade, safety, and walking comfort [63]. Health Information 3 has an importance of 7.66%, and as the health information intervention increases, the probability of walking also increases.

In summary, our findings indicate that among various types of information interventions, health information exerts the most significant influence in terms of promoting walking and cycling. As the intensity of health information increases, so does the probability of individuals engaging in walking or cycling. It emphasizes the priority of residents for their personal health outcomes. The high responsiveness to health information suggests that interventions framing walking and cycling as preventive healthcare may bridge the motivation–behavior gap more effectively than environmental or policy appeals. When combined with infrastructure improvements, the effectiveness of health information interventions is further amplified. Additionally, although environmental information also holds considerable importance, its impact is limited when applied in isolation and requires integration with other intervention strategies. This disparity highlights a misalignment between cognitive awareness and physical constraints—for example, in areas with poor street design, safety concerns may override the persuasive power of environmental messages. Based on analyses of travel duration and distance, the study also reveals that some residents engage in walking or cycling as a form of physical exercise, further reinforcing the strong influence of health information on this segment of the population. These findings suggest that health information and environmental information should be prioritized in promoting walking and cycling, accompanied by parallel improvements in the built environment, particularly in areas where there is high latent demand but low actual usage. Urban design should prioritize shaded routes and rest facilities to support longer health-motivated trips.

5. Conclusions

This study reveals a significant discrepancy between people’s motivations for choosing travel modes and their actual behaviors. Meanwhile, it contributes to a deeper understanding of the psychological mechanisms underlying the motivation–behavior discrepancy in walking and cycling and demonstrates the critical role of information interventions in influencing sustainable travel choices. Grounded in Protection Motivation Theory, the findings reveal several key insights with important implications for both research and policy. (1) Despite Yangzhou’s favorable built environment, a significant discrepancy persists between residents’ motivation and actual behavior in walking and cycling: while 76% of surveyed residents expressed motivation to walk or cycle, only 30% actually adopted these modes. This suggests that physical infrastructure alone is insufficient to drive behavioral change and underscores the need to incorporate psychological dimensions into urban mobility strategies. (2) Compared with individual attributes and travel characteristics, information intervention attributes—particularly health-related information—have a more significant impact on travel behavior. Notably, the effects vary by travel mode: environmental information is more effective in promoting walking, while health information exerts a stronger influence on cycling. This suggests that, compared with policy-related interventions, information interventions emphasizing personal health benefits and environment-related experiential factors are more effective in promoting walking and cycling. (3) Effective interventions should be designed with dual alignment: spatially, to match the characteristics of the built environment, and demographically, to reflect motivational differences among population groups. In particular, combining infrastructure improvements with targeted information strategies can enhance their synergistic effects and more effectively promote walking and cycling behavior.

Based on these findings, this study proposes the following policy recommendations to promote walking and cycling more effectively through integrated psychological and spatial strategies. First, in areas with well-developed built environments, health-related information should be prioritized and tailored to specific population groups. For example, for middle-aged individuals, the messaging could emphasize the benefits of walking and cycling in preventing chronic diseases, while for older adults, it could highlight improvements in sleep quality and enhanced immunity. Second, commonly used mobile applications can be leveraged to deliver context-aware information. For instance, when users plan trips shorter than 3 km using navigation apps, prompts such as “walking is good for your health” could automatically appear. Third, in areas with relatively weak infrastructure, small-scale projects—such as shaded walkways and improved pedestrian signals—should be prioritized, accompanied by synchronized information interventions. Finally, a dynamic evaluation mechanism should be established to continuously track the effectiveness of interventions through methods such as surveys and travel trajectory analysis, enabling timely adjustments to information content and delivery strategies to improve policy precision and adaptability.

Some limitations remain in this study. First, it was conducted in a single city. Gathering more data from other cities to confirm these findings and provide more robust results should be an essential direction of future studies. Second, future studies could consider incorporating more indicators to examine the interaction between the built environment and information intervention measures, such as land use diversity. Third, as the effectiveness of information intervention requires long-term observation, this study is still ongoing. The research team has conducted multiple collaborations with urban planning and construction departments in Yangzhou, collecting feedback results at annual intervals and tracking behavioral changes before and after the information intervention to verify and evaluate the effectiveness of the intervention measures.