Study on Travel Characteristics and Satisfaction in Low-Density Areas Based on MNL and SEM Models—A Case of Lanzhou

Abstract

This study focuses on the challenges of resident mobility in low-density areas. Amid China’s rapid urbanization, rural landscapes and travel patterns are undergoing significant transformation. Using Lanzhou’s rural areas as a representative case study, this research employs questionnaire surveys to collect data. It applies a multi-nominal logit (MNL) model to examine factors influencing travel mode choices and utilizes structural equation modeling (SEM) to assess travel satisfaction—a composite metric derived from residents’ subjective evaluations of convenience, cost, time, and comfort. Findings indicate that private cars and public transportation are the primary travel modes. The MNL model reveals that age and destination accessibility significantly influence travel choices. SEM path analysis further shows that annual household income has a direct positive effect on satisfaction, while age exerts an indirect negative influence through mediating variables. Female satisfaction levels were significantly lower than those of males. Both road density and perceived infrastructure quality significantly enhanced satisfaction, while destination accessibility may exert a slight negative indirect effect by increasing travel expectations. The study theoretically enriches research on rural travel patterns and provides practical insights into rural transportation planning and infrastructure development.

1. Introduction

In the global urbanization wave, the complex interaction between the constantly changing built environment and the development of public transportation in low-density rural areas of China has attracted increasing attention from the academic community. At present, China is at a crucial stage of rapid urbanization, and the changes in the built environment driven by urbanization are significantly reshaping the travel patterns of rural residents. The continuous outward expansion of cities has led to the loss of rural land and population [1]. Infrastructure and roads have a strong and irreversible impact on the built environment of rural areas [2,3]. Against this backdrop, it is crucial to understand the travel behavior of rural residents especially in remote and low-density areas where infrequent public bus services are a lifeline for people and they face challenges of inefficiency and unsustainability [4,5,6,7]; its mechanism in the rural environment has not been fully explored and cannot be directly inferred from urban models [8,9,10]. This research gap is particularly salient concerning the ultimate goal of transportation planning: resident travel satisfaction.

1.1. Theoretical Foundations in Rural Travel Behavior

Research on travel behavior has traditionally been dominated by urban-centric perspectives. A substantial body of the literature has confirmed that dimensions of a built environment—such as density, land-use mix, and design—significantly influence urban dwellers’ mode choices, often promoting walking and public transport use [11]. However, the application of these findings to rural contexts is problematic. Rural areas are defined by low density, dispersed land-use, and relatively weaker infrastructure [12,13,14], leading to fundamentally different travel patterns and constraints. For instance, while high density encourages walking in cities, the essential travel distances in rural areas may render non-motorized modes impractical, resulting in reverse mechanisms [15]. Although some studies have begun to adapt methodologies like the MNL model to analyze rural mode choice [16,17,18], they often remain descriptive and lack a theoretical framework that accounts for the low-density context, leaving the underlying causal mechanisms insufficiently addressed.

1.2. From Mode Choice to Travel Satisfaction: An Overlooked Link in Rural Settings

Travel satisfaction, as a key indicator of service quality and resident well-being, reflects a comprehensive evaluation of travel experiences concerning time, cost, and comfort [19,20,21]. Existing studies, predominantly situated in urban areas, identify determinants including socio-demographics, travel characteristics, and the built environment [22,23,24,25]. Travel behavior includes mode, frequency, purpose, etc. [26], but there is currently a large gap between urban and rural transportation, emphasizing urban–rural integration and coordinated development of urban and rural transportation [27,28]. Choice models jointly construct travel modes and durations to evaluate urban design and planning options from different travel perspectives, resulting in travel-aware urban planning [29]. Crucially, a bidirectional relationship exists where the built environment indirectly affects satisfaction by shaping travel choices, and the chosen mode itself directly influences satisfaction levels [30,31,32]. For example, public transport users’ satisfaction is highly sensitive to reliability and crowding. The vast majority of this evidence base is derived from high-density urban contexts [33,34,35]. Research in rural areas has disproportionately focused on the antecedents of mode choice [36,37,38], while largely neglecting the subsequent outcome of travel satisfaction. While recent years have witnessed a growing interest in rural travel [39,40], significant gaps persist. Consequently, there is a lack of comprehensive analytical frameworks that simultaneously examine the sequential relationship from the built environment to mode choice and, ultimately, to travel satisfaction in low-density rural settings.

1.3. Limitations of the Research and the Research Content of This Article

Through a review of the existing literature, it can be found that although the research on the relationship between the built environment and travel behavior in the urban context has been relatively mature, there are certain limitations when directly applying its theoretical framework and empirical findings to low-density rural areas.

Firstly, the existing research has deficiencies in terms of situational applicability. The mainstream theory is based on the high-density and highly mobile urban form, while low-density rural areas have essential differences in spatial structure, functional layout, and travel distance, and their influencing mechanisms may present different or even opposite characteristics. If the urban model is simply applied, it may be difficult to accurately capture the complexity of rural residents’ travel decisions.

Secondly, in terms of the integrity of the causal chain, most existing studies have limited their analysis to the choice of travel mode and failed to effectively correlate it with the ultimate benefit indicator of travel satisfaction. Travel satisfaction is a key factor in measuring the quality of transportation services and the well-being of residents. However, there is still a lack of systematic empirical tests on how the built environment influences satisfaction by shaping travel choices, especially in the context of rural areas.

This article aims to systematically investigate the sequence relationship among the built environment, travel mode choice, and travel satisfaction in low-density rural areas. Regarding the choice of travel mode as a key mediating variable connecting the objective environment and subjective experience, in order to interpret the internal logic of residents’ travel behavior more comprehensively. This study selects the rural areas of Lanzhou City, Gansu Province as empirical cases. The typical low-density feature, urban–rural dual structure and diversity of transportation modes in this area provide an appropriate context for testing the theoretical framework. The specific objectives of this study include (1) using the MNL model to analyze the unique impact of the built environment in low-density rural areas on residents’ choice of travel modes [41,42]; (2) through the structural equation model, revealing the direct and indirect influence that paths in the built environment have on travel satisfaction, especially verifying the mediating effect of travel mode selection; (3) on this basis, providing targeted policy inspirations for promoting fair and sustainable development of rural transportation.

For this reason, this paper proposes the following research hypotheses:

H1. 

The built environment has a significant impact on the travel mode choices of low-density rural residents, and its mechanism of action may differ from the model in the urban context.

H2. 

The choice of travel mode has a direct effect on travel satisfaction.

H3. 

The built environment will indirectly affect travel satisfaction through the mediating role of travel mode selection.

By integrating the discrete choice model with the structural equation model, this study aims to provide a more coherent analytical perspective for understanding travel behavior in low-density rural areas.

2. Methods

2.1. Research Framework

To systematically investigate the formation mechanisms of travel mode choice behavior and satisfaction among residents in low-density areas, this study constructs a mixed-methods framework integrating MNL and SEM. This framework aims to address the core research questions from two dimensions: objective behavioral decision-making and subjective psychological perception. As shown in Figure 1:

First, to precisely identify key objective factors influencing residents’ discrete travel choices (e.g., selecting public transit, private vehicles, or walking), this study employs an MNL model. This model effectively handles categorical dependent variables, quantitatively analyzing the relative impact of observable variables—such as individual attributes, household characteristics, and travel features—on mode choice. It directly addresses the sub-question: “Which objective factors determine residents’ travel behavior?”

Second, to delve into the internal structure and complex formation pathways of the unobservable subjective latent variable “travel satisfaction,” this study employs SEM. Factor analysis plays a crucial role in this process, serving as the foundation for constructing the SEM measurement model. Specifically, factor analysis is applied to multiple observed items measuring satisfaction, condensing them into a few core latent constructs (e.g., “perceived infrastructure quality,” “destination accessibility”). This operationalizes abstract satisfaction into measurable dimensions. By examining factor loadings between items and latent variables, the structural validity of the scale is validated, ensuring observed variables accurately reflect corresponding latent constructs. Building upon this foundation, path analysis can further test causal relationships among these latent variables, revealing direct and indirect effects in satisfaction formation. This addresses the sub-question: “What are the intrinsic dimensions of travel satisfaction? How do they form?”

This study follows a “parallel-correlational” MNL and SEM analyses conducted in parallel using the same questionnaire data. Their conclusions complement and corroborate each other, jointly providing comprehensive and in-depth empirical evidence from both objective behavior and subjective perception perspectives to enhance rural transportation accessibility and resident satisfaction.

2.2. Research Area and Data

Existing research on the travel behavior of residents can be found to be more concerned about areas such as Shanghai and Jiangsu, in contrast to Western China where little in-depth research has been conducted. Located in the interior of northwest China, Lanzhou is the capital of Gansu Province and the core node city of the “Silk Road Economic Belt”. This paper focuses on the low-density rural area of Lanzhou City, which covers the county and part of the urban–rural transition zone in addition to the central city, accounting for 85.6% of the city’s administrative area. To analyze rural residents’ travel mode preferences and satisfaction, we conducted mixed-method surveys across four representative villages (Hekou, Heishi, Changpo, Baipo), establishing 1 km radius study zones from village centers (Figure 2). Structured questionnaires captured three dimensions: demographic characteristics, built environment perceptions, and mobility preferences. Combining household interviews with digital surveys ensured comprehensive data collection. Following rigorous researcher training and spatial sampling protocols, 204 valid responses were obtained from 212 collected questionnaires [43] (As in Appendix A).

2.3. Model Introduction

2.3.1. Factor Analysis

Factor analysis is a multivariate statistical method based on the idea of dimensionality reduction, which aims to explain the covariance relationship between observed variables using latent factors. Its theoretical origins can be traced back to the field of psychology, where a one-factor model was first proposed by Charles Spearman to explain the underlying structure of variables in intelligence tests. Since then, it has been extended to multi-factor theory, and the principle of “simple structure” has been proposed to optimize factor interpretability. The further development of exploratory factor analysis (EFA) laid the methodological foundation for identifying the structure of latent variable systems by introducing structured modeling of covariance matrices. This study employs EFA to identify the underlying latent factors that influence travel mode choice behavior. The EFA approach is chosen as there are no predetermined hypotheses regarding the factor structure, allowing the data to reveal the natural groupings of variables. The theoretical model of factor analysis is generally expressed as follows: Assuming that there are N samples and P indicators. $X=(X_1, X_2,\dots \dots , X_3)$T is a random vector. The common factor to be determined is $F=(F_1, F_2,\dots \dots , F_m)T.$

$$\begin{array}{c}{X}_1=a_{11}{F}_1+a_{12}{F}_2+\dots +a_{1m}{F}_m+{\epsilon }_1\\ X_2=a_{21}{F}_1+a_{22}{F}_2+\dots +a_{2m}{F}_m+{\epsilon }_2\\ \dots \dots \\ X_p=a_{p1}{F}_1+a_{p2}{F}_2+\dots +a_{pm}{F}_m+{\epsilon }_p\end{array}$$

(1)

The equations above form the factor model. The matrix $A=\left(a_{ij}\right)$ is the factor loading matrix, $a_{ij}$ is the factor loading. The essence of $A$ is the correlation coefficient between the common factor $F_i$ and the variable $X_i$. $\epsilon$ is a special factor denoting a factor other than the common factor [44,45].

2.3.2. MNL Model

The origins of the MNL model can be traced back to the Random Utility Theory of the early 20th century, of which the core assumption is that an individual’s choice behavior is driven by unobservable utility. This model quickly became a central tool in fields such as transportation and consumer behavior due to its simplicity. The MNL model’s applicability is constrained by its assumption of the Independence of Irrelevant Alternatives (IIA), which implies that all alternatives are perceived as independent. To overcome this limitation, subsequent studies have proposed nested logit models and hybrid logit models. These models relax the IIA assumption by incorporating a hierarchical structure and stochastic parameters, respectively [46]. In recent years, scholars have further introduced spatial lag variables or social network variables to promote the localized adaptation of models for rural or regional characteristics [47,48].

Assuming that the set of travel options available to resident n is $G_n$, the utility of a resident $n$ choosing travel mode $i$ is $U_{in}(i\in G_n)$. According to the utility maximization theory, residents choose $i$ when and only when the utility from option $i$ is higher than all other options in the set of options $G_n$, and thus residents n choose mode i under the condition that

$$U_{im}>U_{jn},i,j\in G_n,i\ne j$$

(2)

Random utility theory views utility as a random variable and typically divides the utility function into a fixed-term function and a random-term function, assuming a linear relationship between the two. Thus, the utility $U_{in}$ for a resident n choosing a travel option i can then be expressed as

$$U_{in}=V_{in}+{\epsilon }_{in}$$

(3)

Equation (3): $V_{in}$ is the fixed term; $in$ is the random error term [49].

According to the utility maximization theory, when the utility of an option is greater, the probability of choosing that option is greater, then the probability $P_{in}$ that resident $n$ chooses option i is as follows:

$$\begin{gathered} P_{in}=Prob(U_{in}>U_{jn},\forall i\ne j,j\in G_n) \\ =Prob(V_{in}+{\epsilon }_{in}>V_{jn}+{\epsilon }_{jn},\forall i\ne j,j\in G_n) \\ =Prob({\epsilon }_{jn}-{\epsilon }_{in}<V_{in}-V_{jn},\forall i\ne j,j\in G_n) \end{gathered}$$

(4)

$$P_{in}={\displaystyle \frac{e^{V_{in}}}{{\sum }_{j\in G_n}{e}^{V_{jn}}}}$$

Equation (4): $0\le P_{in}\le 1,\sum P_{in}=1$.

Assuming that $V_{in}$ and ${\epsilon }_{in}$ are independent of each other and that ${\epsilon }_{in}$ obeys a Gumbel (bivariate exponential distribution) distribution, the process is as follows:

Assuming that the error term ${\epsilon }_{in}$ obeys a Gumbel distribution with parameters $(\eta ,\lambda )$, the density function of each error term is distributed as

$$F({\epsilon }_i)=exp(-e^{-\lambda ({\epsilon }_i-\eta )})$$

(5)

Equation (5): Simultaneous derivation on both sides of the equal sign yields the probability density function of ${\epsilon }_{in}$ as

$$f({\epsilon }_i)=\lambda e^{-\lambda (e_i-\eta )}\cdot exp(-e^{-\lambda ({\epsilon }_i-\eta )})$$

(6)

In Equation (6): $\eta$ is a parameter of the Gumbel distribution, which is usually made equal to zero; $\lambda$ is the parameter corresponding to the variance ${\sigma }^2$ of ${\epsilon }_{in}$ and ${\lambda }^2={\pi }^{2}2/{6\sigma }^2$. The expected value of ${\epsilon }_{in}$ is $\eta +\gamma /\lambda$, $\gamma$ is Euler’s constant. Usually when calibrating a logit model, the value of λ cannot be obtained independently, so it can be assumed that λ is calibrated for each parameter by being included in the explanatory variables of $V_{in}$.

2.3.3. SEM Model

SEM evolved from Sewall Wright’s path analysis, which established causal relationship visualization. Karl Jöreskog formalized SEM in the 1970s by integrating confirmatory factor analysis with path analysis, creating its dual framework: measurement models and structural models. Subsequent methodological expansions include Bollen’s multigroup SEM for cross-population comparisons [50] and Bayesian SEM addressing small-sample non-normal data challenges.

$$\begin{array}{c}{x}_1={\beta }_1\eta +{\epsilon }_1\\ x_2={\beta }_2\eta +{\epsilon }_2\\ x_3={\beta }_3\eta +{\epsilon }_3\end{array}$$

(7)

where $\eta$ is the latent variable, $x_1,x_2,x_3$ are the observed variables, ${\epsilon }_1$, ${\epsilon }_2$, ${\epsilon }_3$ are the errors; the regression equation formed is $\eta ={\gamma }_1{x}_1+{\gamma }_2{x}_2+{\gamma }_3{x}_3+\delta$, where ${\gamma }_1$, ${\gamma }_2$, ${\gamma }_3$ are the estimated parameters and $\mathsf{\delta }$ are the error values [49,50,51].

To evaluate how well the model fits the data, the most common indicator of fit is usually the chi-square test of goodness of fit ($x^2$). It can be derived directly from the value of the fit function and its value is related to the size of the sample size. There are also metrics such as the square root of the squared mean residual, the square root of the approximation error, the goodness-of-fit index, and the modified goodness-of-fit index. The square root of the approximation error (RMSEA) is presented.

$$x^2=(N-1)min\left\{F\right\}$$

(8)

$$df={\displaystyle \frac{1}{2}}(n+1)n-t$$

(9)

where N is the sample capacity; $df$ are the degrees of freedom; $n$ is the number of explicit variables; $t$ is the free parameters number of variables.

$RMSEA$ is the root mean square of the approximation error, independent of the size of the sample size and the complexity of the model, and the RMSEA index tends to zero as the model tends to a perfect fit.

$$RMSER=\left\{max\right[(x^2-df)/(N-1),0]/{\left.df\right\}}^{1/2}$$

(10)

3. Result

3.1. Description of Variables Based on Factor Analysis

3.1.1. Socio-Demographic Variables

The main socio-demographic variables considered in this study include gender, age, place of residence, educational background, number of employed household members, annual household income, and number of vehicles per household. Initial analysis of the survey data reveals a higher proportion of female respondents compared to males, with seniors aged over 60 accounting for 23.04% of the rural participants. Due to the scarcity of quality educational resources and the absence of higher education institutions within rural China, many students pursue schooling outside their home regions. Economically, these areas rely heavily on agriculture, and non-agricultural employment opportunities are limited. Consequently, a significant portion of the labor force migrates for work, contributing to an aging local population. In terms of education, 64.22% of the respondents had attained a junior high school education or lower. The overall education level in the sample is relatively low, which corresponds to the older age demographic. Income-wise, most respondents reported an annual household income ranging between $20,000 and $50,000. Regarding health and mobility, a majority of rural participants reported being in good health. Private car ownership was reported by 46.63% of households, whereas 61.27% did not own a motorcycle. Over half of the households possessed either an electric or conventional bicycle, with electric bicycles being the most common, followed by bicycles, cars, and motorcycles. The modal share of travel modes is detailed in Figure 3. The socio-demographic profile of the respondents aligns with the general conditions in rural Lanzhou. Additional details on variable statistics are provided in Appendix B.

3.1.2. Built Environment Perception and Travel Preferences

Data on built environment perception (five variables) and travel preferences (eight variables) were collected using a five-point Likert scale ranging from 1 to 5, comprising a total of 13 measurement items. Exploratory factor analysis (EFA) was conducted in SPSS26 to examine the underlying structures of these constructs.

The EFA for built environment perception demonstrated excellent data suitability, supported by a significant Bartlett’s test of sphericity (p < 0.001) and a KMO value of 0.869. After applying Varimax rotation to improve interpretability, two clear factors emerged, collectively accounting for 88.858% of the total variance. As shown in

Table 1. Building environment perception.

The Component Matrix After Rotation
	Component
	Infrastructure Perception	Accessibility Perception
Satisfaction with the surrounding sidewalks	0.885	0.337
Satisfaction with the surrounding bicycle lanes	0.890	0.341
Satisfaction with the surrounding roads	0.810	0.458
For the convenience of the surrounding bus station	0.304	0.918
Are you satisfied with the comfort during travel	0.570	0.716
Eigenvalue	2.649	1.794
Percentage of variance (%)	52.971	35.887
Cumulative variance percentage (%)	52.971	88.858

, all items loaded strongly onto their respective factors (loadings > 0.5) Table 1. The first factor, labeled Infrastructure Perception, was characterized by high loadings from variables such as satisfaction with sidewalks, bicycle lanes, and roads. The second factor, termed Accessibility Perception, included high loadings from items related to bus station convenience and travel comfort.

Similarly, the EFA for travel preferences also indicated strong factorability (KMO = 0.887, Bartlett’s test p < 0.001). Varimax rotation revealed a two-factor structure explaining 67.637% of the total variance. The first factor reflected Preference for Own Transport, with high loadings from items such as driving, motorcycle, electric bicycle, bicycle, and tricycle use. The second factor, Preference for Walking and Public Transport, was defined by items including public transport use, walking, and other alternative modes. Detailed loadings are presented in

Table 2. Travel preference.

The Component Matrix After Rotation
	Component
	Preference for Your Transport	Preference for Walking and Public Transport
Like to drive to travel	0.624	0.268
Like to travel by motorcycle	0.839	0.270
I like to travel by electric bicycle	0.838	0.223
Like to travel by bike	0.570	0.564
Like riding a tricycle travel	0.819	0.239
Like to travel by public transport	0.383	0.681
Like to walk	0.108	0.874
Like to travel in other ways	0.331	0.745
Eigenvalue	3.059	2.352
Percentage of variance (%)	38.237	29.400
Cumulative variance percentage (%)	38.237	67.637

.

These validated factors were subsequently treated as latent variables in the SEM, providing a robust measurement foundation before examining structural relationships between the constructs.

3.1.3. Daily Travel-Related Variables

Purpose of travel, distance, and time of travel all influence the choice of mode of travel, and the purpose of daily travel, distance to be traveled, and time of travel influence the choice of mode of travel. Daily activities are divided into three categories: work, going to school or picking up children, shopping and participating in recreational activities, and activities such as visiting family and seeing a doctor. The questionnaire statistics show that the mandatory purposes of working and going to school or picking up children account for the highest proportion, while shopping, participating in recreational activities, and visiting relatives and the doctor account for a relatively low proportion. The number of residents traveling within three kilometers and 3–5 km is high. The number of daily trips made by residents is 1–2 times because most of the respondents are working in nearby farms or towns.

3.2. Resident Travel Mode Choice Based on MNL Modeling

In this study, the MNL model was developed using the SPSS software to study the effect of the built environment in low-density areas on residents’ travel mode choices. When there are more independent variables in the actual problem under study, there may be a certain correlation between two or more independent variables, which is called multicollinearity. When the trend of covariance of independent variables is very obvious, it will seriously affect the fitting of the model.

As shown in

Table 3. Multicollinearity test.

Variable		VIF Value
Personal attributes	Gender	1.590
	Age	2.116
	Level of education	2.586
Family and social factors	Number of working people in the family	1.378
	Annual household income	1.800
	Number of family cars	2.368
	Number of family motorcycles	2.002
	Number of household electric bicycles	2.127
	Number of family bikes	1.787
Daily travel-related variables	Daily travel mode	1.893
	Daily travel purpose	1.861
	Daily travel distance	2.328
	Number of trips in the past day	1.684
	Generally waiting for the time of the bus	1.698
Built environment perception	Satisfaction with daily travel modes	3.557
	Satisfaction with the surrounding sidewalks	5.522
	Satisfaction with the surrounding bicycle lanes	7.301
	Satisfaction with the surrounding roads	6.743
	Are you satisfied with the convenience of the surrounding bus station	3.886
	Are you satisfied with the comfort during travel	3.816
Travel preference perception	Do you like driving	2.519
	Do you like to travel by motorcycle	3.613
	Do you like to travel by electric bicycle	3.050
	Do you like cycling	3.135
	Do you like to travel by tricycle	2.848
	Do you like to travel by public transport	2.881
	Do you like walking	1.820
	Do you like other ways to travel	2.305

: before analytical modeling, all variables were tested for covariance. In this study, the variance inflation factor (VIF) was used to test for covariance; the greater the VIF, the stronger the multicollinearity. VIF > 10 indicates that there is severe multicollinearity between the variables, which affects the model fit and cannot be accepted [52,53]. The VIF values of the variables in this study were less than 10 and no multicollinearity was found between the variables.

As shown in

Table 4. Likelihood ratio test.

Variable	Model Fitting Conditions	Likelihood Ratio Test
	The-2 log-Likelihood	Chi-Squared	Degree of Freedom	Significance
Intercept	124.388	0.000	0	0.000
Your gender	133.033	8.645	7	0.279
Your age	300.994	176.606	7	0.000
Your annual household income	276.569	152.181	28	0.000
Your level of education	305.658	181.270	28	0.000
Your daily travel purposes	361.793	237.406	21	0.000
Your daily travel distance	421.525	297.137	21	0.000
Your usual place of residence	328.847	204.460	7	0.000
The number of people working in your family	148.262	23.874	28	0.688
The number of your family cars	164.153	39.765	21	0.008
The number of your family motorcycles	236.786	112.398	14	0.000
Number of Electric Bicycles in Your Home	106.110	0.000	21	0.000
Your number of family bikes	139.017	14.630	14	0.404
Number of trips you made in the past day	322.457	198.069	14	0.000

: the results of the reliability analysis test showed that the overall reliability test’s clombach reliability coefficient was 0.878; the questionnaire has a high reliability. The KMO value was 0.822, the sample is sufficient and is greater than 0.5. The p-value is equal to 0.000, meaning the test result is significant, i.e., there is a certain degree of correlation between the data. After that, the validity analysis was conducted and a KMO value of 0.878 was obtained, a value that indicates that the data has a high degree of fitness and is suitable for factor analysis. Next, when factor analysis was conducted, the results showed that the significance level (Sig.) was less than 0.001, which indicated that the data satisfied the basic assumptions for conducting factor analysis and could effectively extract potential factors. Based on this, we further established the MNL model and conducted the likelihood ratio test to verify whether the model was fitted or not, and the fitted significance result of the model was p = 0.000, with a significance level of less than 0.050, which indicated that the model was fitted very well and was able to effectively explain and predict the main influencing factors of the travel mode choice. Afterwards, based on the output of the model, the relevant parameter estimates were derived as shown in Appendix C.

3.3. Analysis of Residents’ Travel Satisfaction Based on SEM

The first step in structural equation modeling analysis is to construct a hypothetical model. Based on the selection of appropriate statistics, structural equations are built through statistical notation and observation concepts to assist the researcher in decision-making. The path analysis software, AMOS29, is applied to plot the path diagrams, analyze the covariance structure, and calibrate the structural equation model.

The observed variables are depicted, and their data sources are scales or questionnaires. The survey data is imported into the SPSS software after preliminary analysis. Observed variables are drawn in the form of boxes to depict latent variables. Latent variables that are not directly observable are usually abstract concepts that are interpreted by one or more observed variables. After that, interrelationships between variables were established and arrows were drawn. One-way arrows are used to depict causal relationships, where the direction of the arrow is from cause to effect, while the presence of two-way arrows indicates that the two variables are not causally related to each other. Finally, the error variable is added. The SEM implementation protocol involves four systematic phases: model parameter estimation visualizing direct/indirect effects through path diagrams; AMOS-based model specification assigning observed variables to latent constructs with unidirectional and bidirectional arrows; iterative maximum likelihood estimation generating path coefficients, factor loadings, and fit indices; multi-stage validation including modification index-guided adjustments, subgroup stability tests, and standardized coefficient analysis. Model rejection occurs when theoretical specifications mismatch empirical covariance structures, necessitating evidence-based re-specification.

The analytical operations using the AMOS software dedicated to structural equations show that the maximum likelihood estimation method iterations have reached convergence. In the path fit index, the degree of model fit for each data was fully assessed. The p-value of the fitted model is 0.002, which is significantly smaller than the test value of 0.05, and the model is valid. The final chi-square value was 275.588 and the degrees of freedom was 203.RMSEA (0.027) which is less than 0.05, indicating a good fit of the model, and CFI (0.987) was greater than 0.9, indicating good acceptability of the model. The CMIN/DF, NFI, and IFI values all meet the criteria, and the model fit parameters are shown in

Table 5. Fitting results of SEM.

Model Fitting Index	Model Fitting Numerical	Numerical Standard	Value Test Results
CMIN/DF	1.358	<2.0	coincidence
P	0.002	<0.05	coincidence
RMSEA	0.03	<0.05	coincidence
NFI	0.947	>0.9	coincidence
CFI	0.987	>0.9	coincidence
IFI	0.965	>0.9	coincidence

. The direct, indirect, and total impact results of the path analysis are shown in Appendix D.

3.4. Factors Influencing Residents’ Travel Mode Choice

3.4.1. In Terms of Personal Attributes

The coefficients of males choosing to travel by car, motorcycle, electric bicycle, bicycle, tricycle, and public transportation are not significant compared to females, indicating that gender has little influence on the choice of these travel modes. However, in terms of walking, the coefficient for males is −0.916, which tends to have some negative influence. As for 20–59 years old vs. 60 years old and above, the coefficients for age are insignificant when choosing to travel by car, motorcycle, electric bicycle, or bicycle, indicating that the age of 20–59 years old vs. 60 years old and above does not have a significant influence on these modes of travel. For public transportation trips, the coefficient is −3.349; for walking, the coefficient is not significant. As for town vs. rural, the coefficients for driving, motorcycle, e-bike, bicycle, public transportation, and walking trips are not significant for those who have a permanent residence in a town, indicating that permanent residence does not have a significant impact on these modes of travel. On tricycle travel, the coefficient is significant at 5.923, and permanent residents of towns are more likely to choose tricycle travel relative to rural residents. Bachelor’s degree/college has some negative influence on the trends of driving, biking, and walking trips, and some positive influence on the trends of public transportation trips. The coefficients of other educational qualifications are not significant in each mode of travel choice, and the influence on the choice of travel mode is not obvious.

3.4.2. In Terms of Household and Social Factors

The coefficients of the number of people working in the household and the annual household income on the choice of travel by car, motorcycle, electric bicycle, bicycle, public transportation, and walking are mostly insignificant and do not have a significant impact on the choice of travel mode. When the number of motorcycles is 0, the coefficient is not significant when choosing to travel by car, electric bicycle, bicycle, public transportation, or walking; however, the coefficient on motorcycle travel, −27.777, p = 0.028, is highly significant, indicating that the probability of choosing motorcycle travel for residents who do not have a motorcycle is extremely low. The situation is similar when the number of motorcycles is 1. The coefficient −25.732 on motorcycle trips, p = 0.036 is significant, also indicating that residents with one motorcycle rely more on motorcycle trips. The coefficients of the number of cars, the number of e-bikes, and the number of bicycles are mostly insignificant on each travel mode choice, and do not have a significant impact on travel mode choice. Although there is a trend in individual cases, it does not reach a significant level.

3.4.3. On Variables Related to Daily Travel

The coefficients of work, picking up and dropping off children, and recreational activities are mostly insignificant when choosing to travel by car, motorcycle, e-bike, bicycle, tricycle, public transportation, and walking, indicating that the purpose of travel does not have a prominent effect on the choice of these modes of travel. The coefficients of choosing to travel by car, motorcycle, electric bicycle, and bicycle have some trends and reach significant levels in some travel modes, such as the coefficient of driving −6.506, p = 0.057; the coefficient of motorcycle −6.911, p = 0.044; the coefficient of electric bicycle −5.124, p = 0.048; and the coefficient of bicycle −2.936, p = 0.035, which have some negative influence trends, indicating that long-distance trips have some influence on these travel modes. The coefficients are not significant on public transportation and walking trips. The coefficients on the number of trips and time spent waiting for transit are mostly insignificant, with some trends only for individual travel modes.

3.5. Factors Influencing Residents’ Travel Satisfaction

3.5.1. Effect of Annual Household Income (Socio-Demographic Factor) on Travel Satisfaction

As shown in Figure 4, a positive correlation exists between annual household income and travel satisfaction among rural residents. Higher income levels are associated with greater travel satisfaction, largely because improved financial capacity enables families to afford private cars. Consequently, household income exerts a favorable influence on travel satisfaction through enhanced access to private car travel.

3.5.2. Impact of Gender Age (Socio-Demographic Factors) on Travel Satisfaction

As shown in Figure 5, the model analyzes gender variables with women as the reference group, and the results show women’s travel satisfaction is lower than that of men. Rural women in China have less travel experience because they are more focused on household chores, plus they have less travel experience and are less satisfied with public transportation (gender, satisfaction with mode of travel and gender, satisfaction with public transportation, satisfaction with travel negatively affect travel satisfaction), resulting in lower overall travel satisfaction. Age exerts an adverse mediating effect on travel satisfaction, indicating that older individuals generally report lower satisfaction levels compared to their younger counterparts. This disparity can be attributed to the overrepresentation of older adults in rural regions, where limited transportation options are compounded by their greater familiarity with existing transit infrastructure. As a result, increased age restricts mobility choices, thereby reducing satisfaction. The primary pathway for this effect operates through the sequence: age → car access → travel mode satisfaction → overall travel satisfaction.

3.5.3. Effect of Road Density (Built Environment Variable) on Trip Satisfaction

As shown in Figure 6, road density exhibits a significant positive overall effect on rural residents’ travel satisfaction. Path analysis results indicate that road density primarily exerts a positive influence on satisfaction through multiple mediating pathways, including “perceived infrastructure quality” and “travel mode preference.” Specifically, an increase in road quantity directly elevates residents’ evaluation of infrastructure standards (path coefficient: 0.04), thereby enhancing travel satisfaction. Simultaneously, higher road density significantly improves residents’ preference for walking and public transportation (path coefficients ranging from 0.98 to 0.99). This preference then transmits through “travel mode satisfaction” to overall travel satisfaction, forming a complete influence pathway: “road density, preference for walking and public transportation, travel mode satisfaction, travel satisfaction.” This mechanism demonstrates that a well-developed rural road network not only expands travel options and provides more convenient travel conditions but also helps meet residents’ preferences for green travel modes, thereby systematically enhancing their overall travel satisfaction.

3.5.4. The Effect of Destination Accessibility (Built Environment Variables) on Trip Satisfaction

As shown in Figure 7, destination accessibility does not exert a direct effect on rural residents’ travel satisfaction but exerts a significant indirect influence through mediating variables such as perceived accessibility and travel mode satisfaction. Path analysis indicates that destination accessibility indirectly affects overall travel satisfaction through pathways including “perceived accessibility” (β = 0.167) and “travel mode satisfaction” (e.g., preference for private vehicle travel, β = 0.094). Although the New Rural Construction initiative objectively improved travel infrastructure, a notable gap persists between residents’ subjective perceptions of accessibility and the actual conditions. In low-density rural areas, residents’ psychological sensitivity to accessibility is particularly pronounced due to scattered settlements, distant public service facilities, and limited public transportation coverage. This sensitivity significantly influences their evaluation of travel modes, thereby exerting a negative indirect effect on overall travel satisfaction.

3.5.5. Impact of Perceived Built Environment and Travel Mode Preferences on Travel Satisfaction

As shown in Figure 8, from the overall path model perspective, different variables exert markedly distinct influences on travel satisfaction. Preference for walking and public transportation exhibits the strongest positive direct effect (β = 0.223), followed by perceived accessibility (β = 0.167), while preference for private vehicles shows a relatively weaker impact (β = 0.094). This outcome reflects the dual characteristics of travel behavior in low-density areas: on one hand, residents hold high expectations for green travel modes, with their preference significantly enhancing travel satisfaction; on the other hand, while private vehicles play an irreplaceable role in long-distance and flexible travel, their direct contribution to satisfaction remains limited. Additionally, variables like infrastructure perception exert influence through distinct pathways, indicating that travel satisfaction is shaped by a multidimensional interplay of objective and subjective factors. Among these, psychological perceptions and travel preferences play particularly prominent roles. The findings reveal that in low-density rural areas, travel satisfaction depends not only on objective transportation conditions but also critically on residents’ subjective perceptions and psychological expectations regarding travel modes.

Perceived infrastructure has a positive effect on travel satisfaction, in which people perceive that the better the sidewalks, bike paths, and motor vehicle paths around their living environments, the more these amenities are conducive to their daily needs for walking, riding motorized bikes, and driving, thus enhancing their satisfaction with their daily travel modes as well as the effect on travel satisfaction (Perceived Path Infrastructure, Satisfaction with Travel Mode, Travel Satisfaction).

4. Discussion

This study reveals the dominant role of private cars and public transportation, confirming the trend toward a shift from subsistence-oriented to development-oriented travel among residents in urban–rural transition zones [54]. However, unlike some studies indicating the universal influence of socioeconomic variables such as gender and income [55,56], this research demonstrates that the impact of these factors is context-dependent. For instance, gender exerts only a marginal influence on most mode choices. This may stem from the high homogeneity in motorized travel opportunities and constraints experienced by male and female residents within the sample area, suggesting that traditional gender role differentiation in local travel decisions may be weakened by converging economic and social activities. More significantly, the strong predictive power of household motorcycle ownership over motorcycle travel choice highlights the central role of asset ownership as a core determinant of travel “feasibility capability” [57,58].

The SEM model reveals a strong association between perceptions of the built environment and subjective satisfaction, carrying direct policy implications. The strong positive effects of road density and infrastructure perceptions (e.g., sidewalk and lane conditions) suggest that “micro-upgrades” enhancing micro-accessibility within rural areas (such as improving pavement quality and pedestrian connectivity) may be more cost-effective and can immediately boost residents’ well-being compared to investing in large-scale transportation projects. This aligns with findings from [59,60].

Particularly noteworthy is the gendered satisfaction gap. Women’s satisfaction levels are significantly lower than men’s, consistent with most global research findings [56]. Considering our study’s sample context, we hypothesize that underlying mechanisms include the fact that rural women typically shoulder greater caregiving responsibilities (e.g., child pickup/drop-off, household errands), resulting in more complex travel chains with diverse purposes and heightened demands for safety and comfort. The current commuter-oriented transportation system struggles to meet these complex needs, resulting in a significant experience gap. This provides robust empirical support for examining rural transportation equity and advancing inclusive transport from a gender perspective.

The finding that destination accessibility does not yield a direct positive effect, with potential negative indirect impacts, warrants careful consideration. It suggests that merely improving physical accessibility without concurrently enhancing the quality and reliability of travel services—such as frequency and comfort—may not improve subjective experiences. It could even lead to decreased satisfaction as heightened expectations go unmet.

5. Conclusions

This study systematically examines the mechanisms influencing travel mode choices and satisfaction among residents in low-density rural areas of Lanzhou City. Key findings are as follows:

(1)

In terms of travel modes, private cars and public transportation are the primary options. Household-specific vehicle assets (e.g., number of motorcycles) are the most critical predictors of corresponding travel choices, while the impact of certain socioeconomic variables exhibits situational specificity.

(2)

Regarding travel satisfaction, annual household income, road density, and perceptions of infrastructure directly and positively influence satisfaction. Female satisfaction is significantly lower than male satisfaction, revealing gender inequality in travel experiences. Destination accessibility may exert indirect effects through complex mediating pathways.

(3)

This study innovatively integrates MNL and SEM models, revealing intrinsic links between travel behavior choices and subjective satisfaction, providing a holistic perspective for understanding travel patterns in low-density areas.

The study’s conclusions should be interpreted cautiously within the following limitations: First, although the sample size meets basic model analysis requirements, data from a single region limits the generalizability of findings. Future research should expand the sample size to include low-density areas with diverse geographic and cultural contexts for comparative validation, thereby testing the robustness of these findings. Second, cross-sectional data struggles to capture the temporal dynamics of causal relationships. Longitudinal tracking of data could be employed in future studies to reveal the evolving patterns of residents’ travel behaviors during urbanization.

Despite these limitations, this study undoubtedly enriches the theoretical framework of rural transportation behavior research and provides targeted policy insights for promoting equitable urban–rural transportation services and building a gender-inclusive rural transportation system.