Influence of Differentiated Tolling Strategies on Route Choice Behavior of Heterogeneous Highway Users

Abstract

The differential toll policy has emerged as an effective method for regulating expressway traffic flow and has positively impacted the efficiency of vehicular movement, as well as balanced the spatial and temporal distribution of the road network. However, the acceptance of differentiated charging policies and the range of rates associated with these policies warrant further investigation. This study employs both revealed preference (RP) and stated preference (SP) survey methods to assess users’ willingness to accept the current differentiated toll scheme and to analyze the proportion of users opting for alternative travel routes and their behavioral characteristics in simulated scenarios. Additionally, we construct a Structural Equation Model-Latent Class Logistics (SEM-LCL) to explore the mechanisms influencing differentiated toll road alternative travel choices while considering user heterogeneity. The findings indicate that different tolling strategies and discount rates attract users variably. The existing differentiated tolling scheme—based on road sections, time periods, and payment methods—significantly affects users’ choices of alternative routes, with the impact of tolling based on vehicle type being especially pronounced for large trucks. The user population is heterogeneous and can be categorized into three distinct groups: rate-sensitive, information-promoting, and conservative-rejecting. Furthermore, the willingness to consider alternative travel routes is significantly influenced by factors such as gender, age, driving experience, vehicle type, travel time, travel distance, payment method, and past differential toll experiences. The results of this study provide valuable insights for highway managers to establish optimal toll rates and implement dynamic flow regulation strategies while also guiding users in selecting appropriate driving routes.

1. Introduction

In recent years, China’s highway system has undergone remarkable development. As a crucial component of the transportation infrastructure, highways play a vital role in enhancing regional connectivity and driving economic growth, marking a significant milestone in China’s transportation evolution [1]. By 2023, the total length of highways had reached 184,000 km, with continuous improvements in both scale and quality of construction, resulting in a relatively comprehensive highway network. However, the increasing travel demand has led to an overload on its carrying capacity, causing spatial and temporal imbalances in traffic flow. Sections of highways with persistently high traffic not only compromise their performance but also escalate maintenance costs and pose safety and efficiency challenges [2]. While some highways in economically developed areas remain profitable, many others struggle to break even or are even operating at a loss.

In 2019, China initiated the phased elimination of provincial boundary toll stations. In June 2021, three governmental ministries jointly issued the “Implementation Plan for Comprehensively Promoting Highway Differential Toll Collection” [3]. This differentiated toll program considers market demand, resource availability, and user characteristics to develop diverse pricing strategies. Currently, numerous provinces and cities have implemented various differentiated charging measures [4,5], aligning them with local economic conditions and road environments to create preferential rates from multiple angles. For instance, Guizhou Province offers a 10% discount on original toll rates for six logistics corridors connecting Guiyang to the Guangdong-Hong Kong-Macao Greater Bay Area and neighboring provincial capitals. Liaoning Province provides a 50% discount for international standard container trucks passing through designated toll stations equipped with electronic toll collection (ETC) systems. In Guangdong Province, a similar 50% discount is offered for trucks using the Shenzhen section of the Guangzhou-Shenzhen Coastal Expressway. Sichuan Province has increased the toll discount for new energy trucks equipped with ETC from 5% to 20%, and for international standard container trucks installed with ETC, from 30% to 60%. Anhui Province grants a 10% discount on intra-zone tolls for buses making 30 or more trips per month. Meanwhile, the Ningxia Hui Autonomous Region provides a 25% discount on existing tolls for freight vehicles using ETC on specific road sections during the night (21:00–6:00). To promote tourism, the Guangxi Zhuang Autonomous Region has introduced a 50% discount on tolls for a specific type of passenger vehicle using ETC on seven pilot highways from Friday (00:00) to Sunday (24:00). The adjustment of users’ travel times and route selections through economic incentives can optimize resource utilization and enhance efficiency [6]. For instance, following the implementation of the differentiated toll policy on a pilot road network, parallel road traffic flow increased significantly, with local traffic experiencing an approximate 30% rise and traffic from other provinces increasing by about 10%. Truck traffic notably surged throughout the day during the designated preferential hours.

In summary, differentiated tolling has emerged as an effective strategy for regulating expressway traffic flow, positively influencing vehicle traffic efficiency, and balancing the spatial and temporal distribution of the road network [7]. However, while some policies yield positive outcomes with smaller concessions, others necessitate more substantial adjustments, indicating that a uniform standard is not applicable [8]. Thus, further exploration of specific policy effects and the range of rates in the differentiated charging strategy is essential. Given the diverse nature of travel users, it is challenging to conduct a granular individual analysis; however, user segmentation based on subcategories can facilitate the development of more precise charging strategies.

To bridge this gap, this study employs revealed preference (RP) and stated preference (SP) survey methods to assess users’ willingness to accept the current differentiated scheme and to analyze the proportion of users opting for differentiated road adjustments along with their behavioral characteristics in the simulated scenarios. Additionally, a Structural Equation Modeling-Latent Class Logistics (SEM-LCL) model is constructed to investigate the influence mechanisms of differentiated toll policies on travel choice behavior, considering user heterogeneity. This research aims to provide insights for highway operation departments, assisting them in making informed pricing decisions and effectively utilizing transportation resources by influencing users’ travel times and route choices.

The remainder of this study is structured as follows: Section 2 reviews the relevant literature; Section 3 details the modeling framework and methodology; Section 4 outlines the questionnaire design, data collection process, and a case study in a Chinese province; Section 5 presents the modeling results and discussions; and Section 6 concludes with a summary of findings and implications for future work.

2. Literature Review

2.1. User Heterogeneity

Individuals exhibit varying preferences for service quality, and this variability significantly influences their choice behaviors. A thorough understanding of this heterogeneity can aid in developing more effective marketing strategies. There are two primary approaches to addressing perceived and preference heterogeneity: one involves categorization studies based on socio-economic and trip-related characteristics [9]; the other focuses on identifying the factors and their weights that influence service quality in decision-making, grounded in the behaviors of users in specific scenarios [10]. Gholi et al. [11] collected preference data from 360 commuters in Tehran regarding four selected attributes of regular bus services and employed a random parameter mixed logit model to analyze the heterogeneity of user preferences. Their findings indicated that such preferences could be explained by observed characteristics. Bellizzi et al. [12] designed a revealed preference (RP) questionnaire to gather passengers’ evaluations of their last flight experience and a stated preference (SP) questionnaire to assess service expectations under hypothetical scenarios. Their study revealed differences in passengers’ perceptions regarding factors such as cabin crew attitude, waiting times for flights, and seat fares. Yun et al. [13] examined passengers’ preferences for automated mobility on-demand (AMoD) services. Their analysis highlighted the significant impact of two indicators—vehicle type and seat selection—on user preferences. Moreover, they found that increasing the number of service vehicles was more beneficial for competitiveness than merely expanding service coverage. Moller et al. [14] investigated the motivations behind choices by observing patterns of choice behavior within subway carriages. They developed a mixed latent hierarchical choice model and proposed interventions to enhance the passenger experience. Shao et al. [15] constructed a two-tiered network model to study passenger emotions during large-scale flight delays, illustrating the contagion mechanism between online and offline social networks. This model accounts for individual heterogeneity and reveals that negative emotions can be exacerbated among passengers upon entry. Additionally, reasonable pricing of ancillary services should be user demand oriented. However, existing studies often lack the refinement needed for pricing differentiated services, resulting in a failure to adequately address the personalized needs of users.

2.2. Factors Influencing Travel Route Choice

Accurately studying the mechanism of users’ travel behavior, mining their potential preference patterns, and analyzing the influencing factors of route selection can help to better disseminate the induction information [16]. Zheng et al. [17] used a GPS device to classify drivers based on observed vehicle acceleration and deceleration behaviors, and the results showed that there were differences in carbon emissions and fuel consumption among different drivers. Cautious drivers had the lowest fuel consumption and carbon emissions when stopping, while drivers familiar with the route had the lowest acceleration rate, the highest cruising speed, and a shorter distance from the front when driving. Yang et al. [18] investigated the mechanisms influencing drivers’ travel paths and found that factors such as risk-perceived ability, age, and driving experience had a positive effect on driving decisions. Distracted driving behavior and complex road environments can have a negative impact. Yang et al. [19] used the K-means method to classify 1881 drivers participating in the experiment into low, medium, and high stress levels and investigated driver characteristics under different stress levels. Gender was found to have no significant effect on the stress level, but driving experience had a significant effect. Meanwhile, driving stress increased with decreasing driver age. It is worth noting that the stress level of drivers increases significantly when the vehicle in front of them is a large truck. Li et al. [20] found that the probability of drivers detouring in the area of the reconstruction operation is higher during weekdays and during restricted travel hours such as morning and evening peak hours. As the number of lanes occupied in the construction area increases, the probability of detouring also increases. Wei et al. [21] established a road loyalty evaluation index system for truck drivers, where users preferred to travel on familiar roads out of skepticism about the controllable cost and the reliability of unfamiliar roads. Payyanadan et al. [22] found that travelers’ familiarity with the road network and their willingness to choose alternative routes had a negative relationship. Drivers who also received sufficient guidance information were more likely to choose alternative routes.

2.3. Willingness to Pay

Accurate estimation of passengers’ willingness to pay is crucial in pricing decisions. Especially when adjusting fares, it helps to develop a reasonable fare operation strategy. Yang et al. [23] analyzed the influencing factors of parkers’ acceptance of differentiated fare policies, set dynamic price fluctuation ranges for different types of drivers, and proposed attraction measures. Li et al. [24] proposed a route- and time-based differentiated fare strategy for public transit based on the daily group behaviors of socially interacted travelers. Compared with the traditional pricing model, this scheme can reduce congestion in the transportation network. At the same time, passengers can enjoy higher travel utility, and the transit system gains higher profits. Ren et al. [25] considered passengers’ socio-economic characteristics, travel characteristics, flight characteristics, and seat categories and constructed a seat selection willingness-to-pay model to predict passengers’ willingness to pay for selecting a seat in different scenarios, which provided a basis for airlines to set auxiliary service fees. Zhu et al. [26] used a questionnaire survey to obtain the travelers’ willingness to choose the departure time in long and short vacations. Statistics on the impact of highway toll policy on the departure time of long and short vacations. Those who traveled during short holidays or long trips reduced travel rates and preferred to travel earlier. Song et al. [27] analyzed the factors affecting the willingness to pay of bike-sharing users, and the study covered 11 latent variables and a total of 34 measurement items. 502 bike-sharing users in China’s first- and second-tier cities were surveyed. The results show that perceived value, payment awareness, trust, and environmental awareness are the key factors influencing the willingness to pay of bike-sharing users. Alhassan et al. [28] considered the behavioral complexity of commuters and investigated the impact of their daily behaviors on the traffic flow, profitability, and congestion level of the transportation network and found that the average WTP of non-commuting users is 42% higher than that of commuting users. Based on this finding, differentiated fares are developed to guide the rational distribution of traffic flows based on the range of WTP values of different user groups. Ortega A et al. [29] focused on how to determine the optimal combination of tolls for a tolled highway and parallel roads. The higher the average value of travel time (VTT) of users, the higher the optimal pricing, while the higher the dispersion of VTT, the lower the optimal pricing. The study suggests that highways should be set with higher tolls to avoid congestion, while ordinary roads should be set with lower tolls to promote access.

2.4. Modeling

The widely used methods for route choice modeling include machine learning methods and random utility theory [30]. Some studies have combined machine learning methods to capture the degree of influence of variables on travel path selection. Kong et al. [31] explored the factors affecting truck drivers’ route selection based on the XGBoost model, which was validated by a truck trajectory dataset from Baltimore (MD, USA). It was shown that truck drivers are more sensitive to real-time congestion information and reliability information when the difference in travel time and stability of candidate routes reaches a certain threshold. Lee et al. [32] estimated the travel routes of subway passengers using the empirical cumulative distribution function (ECDF). Meng et al. [33] proposed a personalized rideability-based route recommendation method (PBCRR), which uses deep neural networks to statistically characterize the environmental attributes and rideability features and to find the optimal riding routes based on the minimum impedance of road segments. Mou et al. [34] used a personalized recurrent neural network (P-RecN) to capture tourists’ dynamic travel preferences and predict tourists’ typical behavioral intentions by mining historical behavioral trajectory information.

Machine learning methods are suitable for big data environments where high accuracy is required. Although they can improve prediction accuracy, they require a large amount of data for training and are time-consuming. In contrast, discrete choice models apply less data to capture the heterogeneous characteristics of travel users while providing flexibility and interpretability. Hua et al. [35] analyzed North Carolina car accident data using Latent Class Logit to classify them into six categories. In order to explore the heterogeneity and inter-class consistency between categories, a stochastic parametric logit model was developed for each category, and the influencing factors were estimated through marginal effects. Variables such as gender, region, road alignment, and inclement weather were found to have a more significant effect on rollover accidents. Si et al. [36] investigated the relationship between latent variables and behavioral intention by building a structural equation modeling (SEM) and introducing latent variables into the logit model to form a SEM-Logit model. Online survey data was utilized to explore the model choice for services between cabs and online cars. The results of the study show that the SEM-Logit model is superior to the single Logit model in terms of precision and accuracy. Also, the interpretability of the model was significantly improved by adding latent variables. Lu et al. [37] used latent category Logit and traditional Logit models to predict the probability of purchasing air tickets at different prices and statistically found that the estimation results of the latent category Logit model were better than the traditional Logit model in terms of accuracy by using the real data information.

2.5. Summary

Foreign highways commonly use the no-toll model. Domestic studies pay more attention to the use of ETC transaction data and OD data to assess traffic congestion and predict the traffic status at a specific moment based on historical data [38,39]. Differentiated tolling schemes are widely used to achieve dynamic regulation of traffic flow in scenarios such as high-speed railroads, public transportation, and parking lots but are less frequently applied to highways. The travelers and the willingness and sensitivity to accept policies and rates are also less considered in the study of differentiated tolling schemes for highways. At the level of research methodology, existing studies less deeply analyze the key influencing factors of the SEM-LCL integration model, which makes transportation policy-making lack a reliable basis.

3. Methodology

3.1. Travel Option Selection Considering Utility

Route selection for highway users’ travel is influenced by a combination of factors, including on-travel time, travel costs, and rate discounts. According to individual needs, the user decision-making process can be divided into five stages: problem recognition, information acquisition, program evaluation, decision-making, and travel evaluation [40]. (1) Problem recognition stage: Users formulate the initial route by combining their own travel needs through internal subjective evaluation and external trip characteristics. (2) Information acquisition stage: search for or receive internal and external information. The internal information involves their own original travel experience, and the external information includes the road traffic environment and comprehensive transportation costs. At this stage the user forms a preliminary decision. (3) Program evaluation stage: synthesize the feedback information obtained in the second stage and compare the original route and alternative routes. (4) Decision-making stage: Users make subjective decisions in a specific traveling environment. If users are rational, they will choose the route with the greatest utility among the alternatives. (5) Travel evaluation stage: users summarize their travel experience and compare it with their initial wishes to determine whether it meets their expected expectations. By reflecting on and updating their travel experience, users continue to accumulate experience and optimize their decision-making in future trips to better match their personal needs and preferences. The schematic diagram of the user’s travel route selection scheme is shown in Figure 1.

3.2. Questionnaire Survey

This study combined both revealed preference (RP) and stated preference (SP) methods and used simple random sampling to obtain sample subjects. A pre-survey was conducted to improve the quality of the questionnaire in view of the differences in the literacy level of the participants. The formulation of difficult-to-understand questions was modified in response to feedback. The questionnaire was also tested for reliability to correct the ineffective questions, and then the formal survey was conducted. A 5-point Likert scale was used for all questions. A score of 1–5 was used to represent the frequency, likelihood, or accuracy of the formulation of the questions [41]. Where “1” represents low frequency, low likelihood, or inaccurate descriptions, “5” represents high frequency, high likelihood, or very accurate descriptions [42]. Participants were informed that the survey was to be used for research purposes only and were asked to fill it out based on truthfulness. The questionnaires were collected and then initially screened to exclude non-compliant questionnaires, including cases of identical options and inconsistencies.

3.3. Modeling Framework

The influencing factors of travel route selection usually have a complex structural hierarchy. A single regression analysis method is difficult to accurately explain the relationship between users’ attitudes, perceptions, and latent variables. There is a need to use more precise path analysis tools and to estimate and test the relevant parameters in the model. Structural equation modeling combines the advantages of factor analysis and linked equation modeling and is able to integrate multiple observed and latent variables and reveal the relationships between variables [43]. Logit modeling reflects heterogeneity by dividing the overall sample into multiple subgroups with similar characteristics [44]. The SEM-LCL model constructed in this study is schematically shown in Figure 2.

(1) Structural equation modeling

$①$ Construct structural characteristic equations

$$X={\Lambda }_x\zeta +\delta$$

(1)

$$Y={\Lambda }_y\eta +\epsilon$$

(2)

where $\zeta$ is the vector of exogenous latent variables; $X$ is the vector of observed variables for $\zeta$; ${\Lambda }_x$ is the factor loadings; $\delta$ is the measurement error for the exogenous variable; $\eta$ is the vector of endogenous latent variables; $Y$ is the vector of observed variables for $\eta$; ${\Lambda }_y$ is the factor loadings; and $\epsilon$ is the measurement error for the exogenous variable.

Structural equation modeling is used to explain the causality of latent variables, where exogenous latent variables refer to potential factors that influence other variables, and endogenous latent variables refer to “outcome” latent variables that are influenced by other variables.

$$\eta =B_{\eta }+\Gamma \zeta +\zeta$$

(3)

where $B_{\eta }$ is the coefficient matrix of the endogenous latent variable; $\Gamma$ is the coefficient matrix of the exogenous latent variable; $\zeta$ is the error that is difficult to predict or explain in the model.

$②$ Normalization of load factors in SEM

To describe the normalization process of the loading factor coefficients of the exogenous latent variable ${\xi }_1$ in the SEM as an example, let the exogenous latent variable ${\xi }_1$ have n corresponding observable variables, which are denoted as $x_{11}$, $x_{12}$,..., $x_{1\mathrm{n}}$, as shown in Equation (4).

$$\left(\begin{array}{l}{x}_{11}\\ x_{12}\\ \dots \\ x_{1n}\end{array}\right)=\left(\begin{array}{l}{\Lambda }_{x1}\\ {\Lambda }_{x2}\\ \dots \\ {\Lambda }_{xn}\end{array}\right){\xi }_1$$

(4)

The loading coefficients ${\Lambda }_{x1}$, ${\Lambda }_{\mathrm{x}2}$,..., ${\Lambda }_{\mathrm{xn}}$ are used as weights of the observable variables and their normalized weights are denoted by ${\alpha }_{x1}$, ${\alpha }_{x2}$,..., ${\alpha }_{xn}$. ${\alpha }_{xn}$ as shown in Equation (5), the characteristic expression of exogenous latent variable ${\xi }_1$ is obtained as shown in Equation (6).

$$\left\{\begin{array}{l}{\alpha }_{x1}={\displaystyle \frac{{\Lambda }_{x1}}{{\Lambda }_{x1}+{\Lambda }_{x2}+\dots +{\Lambda }_{xn}}}\\ {\alpha }_{x2}={\displaystyle \frac{{\Lambda }_{x2}}{{\Lambda }_{x1}+{\Lambda }_{x2}+\dots +{\Lambda }_{xn}}}\\ {\alpha }_{xn}={\displaystyle \frac{{\Lambda }_{xn}}{{\Lambda }_{x1}+{\Lambda }_{x2}+\dots +{\Lambda }_{xn}}}\end{array}\right.$$

(5)

$${\xi }_1={\alpha }_{x1}{x}_{11}+{\alpha }_{x2}{x}_{12}+\dots +{\alpha }_{xn}{x}_{1n}$$

(6)

$③$ Establishment of utility functions

According to the random utility theory, it is assumed that the user always chooses the alternative with the largest utility $u_{in}$. The utility function is usually divided into a fixed term $v_{in}$ and a random term ${\epsilon }_{in}$, as shown in Equation (7).

$$u_{in}=v_{in}+{\epsilon }_{in}$$

(7)

The utility function includes not only observable variables such as users’ personal socio-economic and travel characteristics, but also latent variables such as users’ perceptions and attitudes towards different options. Therefore, the fixed term $v_{in}$ in the function can be expressed by Equation (8).

$$v_{in}={\displaystyle \sum _{l\in L}{a}_{il}{s}_{lin}}+{\displaystyle \sum _{q\in Q}{b}_{iq}{Z}_{iqn}}+{\displaystyle \sum _{k\in K}{c}_{ik}{\eta }_{ikn}}$$

(8)

$$d_{(a,b)i}=\left\{\begin{array}{l}1,V_{ai}\ge V_{bi}\\ 0,\mathrm{else}\end{array}\right.$$

(9)

(2) Latent class conditional logit model

Assuming $\mathrm{K}$ potential category, the model’s formulation can be divided into two parts: the category assignment model and the Logit model within the category.

The probability that individual $i$ belongs to the latent category $k$ in the category assignment model ${\pi }_{ik}$, is shown in Equation (10).

$${\pi }_{ik}=P(Z_i=k)={\displaystyle \frac{\exp ({\gamma }_k)}{{\displaystyle {\sum }_{j=1}^K\exp ({\gamma }_j)}}}$$

(10)

where ${\mathsf{\gamma }}_k$ is a parameter for category $k$ and $Z_i$ denotes a potential category for individual $i$.

The inter-category choice model is the probability that an individual $i$ chooses option $j$ in category $k$ as in Equation (11).

$$P(Y_{ij}\left|Z_i=k\right.)={\displaystyle \frac{\exp (X_i{\beta }_{jk})}{{\displaystyle {\sum }_{m=1}^J\exp (X_i{\beta }_{jk})}}}$$

(11)

where $Y_{ij}$ is the response variable indicating whether individual $i$ chose option $j$, $X_i$ is the eigenvector of individual $i$, and ${\beta }_{jk}$ is the regression coefficient of the option $j$ in the category $k$.

Overall, the full probabilistic model of the latent category Logit model can be written as Equation (12). The probability that some individual $i$ chooses $Y_i$ is the weighted sum of the probability of each latent category for that individual and the conditional probability of choosing $Y_i$ under that latent category.

$$P(Y_i)={\displaystyle \sum _{k=1}^K{\pi }_{ik}}P(Y_i\left|Z_i=k\right.)$$

(12)

(3) Model parameter estimation

The parameters of the model are estimated by maximizing the likelihood function. Formula is Equation (13).

$$L(\gamma ,\beta )={\displaystyle \prod }_{i=1}^N\left({\displaystyle \sum _{k=1}^K{\pi }_{ik}P(Y_i\left|Z_i=k\right.)}\right)$$

(13)

The parameters are estimated using the expectation maximization algorithm, and the parameter estimates of the model are finally obtained by iterating the E-step and M-step repeatedly until convergence [45]. The formula of the E-step algorithm is shown in Equation (14), and the formula of the M-step algorithm is shown in Equations (15) and (16).

E-step: Compute the posterior probability that each individual $i$ belongs to each latent category $k$.

$${\tau }_{ik}=P(Z_i=k\left|Y_i\right.)={\displaystyle \frac{{\pi }_{ik}P(Y_i\left|Z_i=k\right.)}{{\displaystyle {\sum }_{l=1}^K{\pi }_{il}P(Y_i\left|Z_i=l\right.)}}}$$

(14)

M-step: Update parameters $\gamma$ and $\beta$ to maximize the likelihood function.

$${\gamma }_k=\arg \underset{\gamma }{\max }{\displaystyle \sum _{i=1}^N{\tau }_{ik}\log }{\pi }_{ik}$$

(15)

$${\beta }_{jk}=\arg \underset{\beta }{\max }{\displaystyle \sum _{i=1}^N{\tau }_{ik}\log }P(Y_{ij}\left|Z_i=k\right.)$$

(16)

3.4. Factors Influencing Choice Preferences

The choice of travel options is a complex decision-making process that is influenced by a variety of factors, especially during peak hours or in congested areas. Different users have different expectations and preferences regarding costs, services, etc. They weigh these factors when choosing travel routes, and this choice reflects the heterogeneity among individuals.

In the acceptance degree of the differentiated toll policy, individual perception and attitude are latent variables, which cannot be directly observed and need to rely on relevant observational variables for measurement. This is specifically shown in Figure 3.

4. Case Study

In order to obtain the information of users’ travel mode choice behavior under highway differential toll collection, the RP and SP methods were used to design the survey. RP survey data is based on the actual travel records of the interviewed users, including age, income, itinerary, and other information, which is used to analyze their socioeconomic status and travel characteristics. At the same time, in order to have a more comprehensive understanding of travelers’ preferences, the user’s choice behavior under different charging policies is determined. The SP survey was used to calculate the travel choices of users under the hypothetical scenario.

4.1. Data Collection

This study employs Questionnaire Star to generate links, which are then widely disseminated across various online social platforms, including WeChat, Baidu Post, and Zhihu. To enhance the targeting, these links were specifically distributed to provinces that have pioneered early reforms in differential toll collection, such as Zhejiang, Guizhou, Guangdong, and Jiangxi, based on their respective IP addresses. Additionally, offline, paper questionnaires were distributed at key locations in Wuhan city and its surrounding areas, including toll stations, highway service areas, and freight stations. To ensure the quality of responses, online respondents were prompted to complete the questionnaire based on their vehicle models and preferences. Only those questionnaires where trap questions were answered correctly and where the time spent on completing the survey exceeded the minimum threshold were retained. In the case of onsite distribution, the ratio of large to small vehicles was deliberately set at 1:2. A total of 327 questionnaires were collected, of which 300 were deemed valid, resulting in a retention rate of 91.74%. Given that the minimum sample size required for structural equation modeling (SEM) is 100, it is justifiable to utilize this dataset to analyze the relationship between variables, ensuring that the sample is representative of the broader population of highway travel users. The questionnaire is meticulously structured into four distinct parts: (1) Demographic and socio-economic attributes; (2) Trip-related characteristics; (3) Willingness to choose differentiated toll policies; (4) Hypothetical scenarios of users’ travel option selections.

4.2. Research Scenario

Differentiated toll roads, often referred to as “alternative roads”, cater to the diverse needs of travelers by offering options with varying departure times, route durations, toll amounts, and other characteristics. This flexibility allows users to select the optimal route that best aligns with their personal preferences and requirements. In this study, two primary factors are considered: travel distance and travel time. The travel distance is categorized into short and long distances, while travel time is segmented into peak, off-peak, and night periods to accommodate the specific travel needs of truck drivers. To streamline the research process, an orthogonal experimental design was conducted using SPSS26 software, whereby various experimental scenarios were screened to eliminate substitutable parts, ultimately identifying 24 distinct research subjects. A randomized group design method was employed, providing each respondent with six different scenarios to prevent information overload and ensure the accuracy of their responses. Additionally, all scenarios were categorized into short-distance and long-distance sections to mitigate interference from sequential questions. Figure 4 presents a specific scenario to facilitate respondents’ route selection. To enrich the study’s insights into differentiated travel traffic guidance information, the questionnaire also included a survey on users’ habitual methods of obtaining highway travel information, as well as their maximum acceptable detour distances and times.

Vehicles typically follow the routes recommended by navigation maps. The primary objective of this study is to investigate congestion on recommended roads during peak periods and the availability of alternative routes during off-peak periods. It is crucial to note that, as the pricing strategy of differential tolling is still under discussion, this study aims to assess user acceptance and opinions regarding the differential tolling policy. Different rate levels will exert varying degrees of impact. Smaller rate changes align more closely with user expectations and acceptance but have less influence on user choice behavior. Conversely, larger rate changes significantly impact user choice behavior but generate more controversy. To achieve the dual goals of maximizing user benefits and minimizing costs for highway authorities, a comprehensive consideration of rate changes is essential. Drawing upon the current toll policy interval range, multiple gradient rates were experimentally set up to compare users’ actual travel choices under different pricing scenarios. This comparison aims to comprehensively evaluate the effectiveness of the differentiated toll policy.

4.3. Data Description

Table 1. User personality traits descriptive analysis.

Factor	Attributes	Frequency	Proportion (%)
Individual socio-economic characteristics
Gender	Male	204	68.1
	Female	96	31.9
Age	<20 (Year)	50	16.7
	20–35 (Year)	175	58.3
	35–50 (Year)	52	17.3
	>50 (Year)	23	7.7
Monthly income	<4000 (RMB)	78	26.1
	4000–8000 (RMB)	104	34.8
	8001–12,000 (RMB)	79	26.4
	>12,000 (RMB)	39	12.7
occupation	Student	49	16.4
	Self-employed/retired	40	13.5
	Government employee	91	30.5
	Private employee	120	39.6
Driving experience	<3 (Year)	71	23.7
	3–6 (Year)	88	29.3
	7–10 (Year)	92	30.6
	>10 (Year)	49	16.4
Educational level	High school and below	42	14.2
	Junior college/Undergraduate	179	59.5
	Master’s degree or above	79	26.3
Trip-related characteristics
Vehicle type	Private car	235	78.3
	Passenger car	41	13.7
	Truck	20	6.7
	Special-purpose vehicle	4	1.3
Travel purpose	Business travel	29	9.6
	Leisure tourism	40	13.5
	Visit relatives	80	26.7
	Commute travel	51	50.2
Travel pattern	Rigid travel	236	78.4
	Flexible mobility	64	21.6
Travel time	Daytime	209	69.8
	Night	91	30.2
Travel range	Within the province	197	65.7
	Transit travel	71	34.3
Travel distance	<50 km	31	10.3
	50–100 km	104	34.6
	100–200 km	128	42.8
	>200 km	37	12.3
Travel frequency	Almost everyday	14	4.7
	1–4 times per week	76	25.3
	1–3 times per month	142	47.3
	Less used	68	22.7
Number of peers	Only one	55	18.4
	2	166	55.2
	>3	79	26.4
Payment methods	ETC	218	72.6
	Cash or other	82	27.4
Differentiated tooling experience	Yes	229	76.3
	No	71	23.7

provides a summary of the statistical data regarding the personality traits of the users surveyed. Among the respondents, 68.1% were male and 31.9% were female. The age distribution indicates that over half of the respondents are under 30 years old, while only 5% are over 50. Among the drivers who participated in the survey, private sector employees represented the largest group at 39.6%, followed by government employees at 30.5%. Notably, more than 76.3% of the respondents reported having over three years of driving experience, with 23.7% being novice drivers. The data also reveals that intra-provincial travel is more common than transit travel among the respondents. Most participants preferred to travel in private cars, which aligns with current trends on the highway. However, the proportion of large trucks and special operations vehicles was found to be lower than the actual number of trips made. Over 45% of respondents indicated that they travel on the highway one to three times a month, primarily covering medium to long distances of 50–200 km. Additionally, 70% of the respondents typically travel with two or more people, and most trips occur during the daytime. The overwhelming majority of users utilize electronic toll collection (ETC) systems for payment and have experienced differential pricing.

4.4. Model Fitness

4.4.1. SEM Model

The Amos26 software was used to test the suitability of the structural equation modeling, and the fit index was corrected by the MI value until the standard was met [46,47]. The fitness indicators of the model are shown in

Table 2. Model Fit and Effectiveness.

Fit Indicators	Standard	Measured Values	Judgment
${\displaystyle \raisebox{1ex}{${\chi }^2$}\!\left/ \!\raisebox{-1ex}{$df$}\right.}$	1–3	2.006	Acceptance
GFI	>0.9	0.899	Acceptance
AGFI	>0.9	0.869	Acceptance
NFI	>0.9	0.909	Acceptance
IFI	>0.9	0.952	Acceptance
RFI	>0.9	0.891	Acceptance
CFI	>0.9	0.952	Acceptance
TLI	>0.9	0.942	Acceptance
RMSEA	<0.08	0.056	Acceptance

. The measured values are all within the standard range, indicating that the hypothesis fit is superior and the analysis of the paths between the latent variables can be continued.

4.4.2. LCL Model

Each stochastic parameter was assumed to follow a normal distribution to ensure the best possible fit for the model. All features were tested with random parameters, and only those that were statistically significant were selected for inclusion in the model. The number of user categories was determined using the LCL model, with the model parameters subsequently estimated. Initially, the number of potential categories was established to be between 2 and 5 [48]. Metrics such as log-likelihood (LL), Canonical Correlation Analysis (CCA), and Bayesian Information Criterion (BIO) were calculated for different numbers of potential categories to evaluate the model’s goodness of fit. As shown in Figure 5, with three potential categories, the BIO and CCA values were the lowest, while the LL value was the highest, indicating that the model performs optimally at this level.

5. Model Results and Discussion

5.1. SEM Model Analysis

The distribution of standardized loading factors between each latent variable in the structural equation is shown in Figure 6, and $e_1$ to $e_{25}$ are the residual terms generated by default by Amos software. Multicollinearity issues are identified and addressed by calculating the variance inflation factor (VIF) using AMOS. Additionally, regularization techniques such as ridge regression and Lasso regression are employed to further enhance the model’s stability and ensure the reliability of the results.

Table 3. Path significance analysis.

	Estimate	S.E.	C.R.	p
Differentiated charges by vehicle type (category)	0.112	0.061	1.849	-
Differentiated toll collection by section	0.444	0.079	5.603	***
Differentiated tolls for different time periods	0.232	0.085	2.732	**
Differentiated tolls for separate entrances and exits	−0.11	0.066	−1.672	-
Differentiated tolls by payment method	0.187	0.079	2.365	*
Directionally differentiated tolls	0.056	0.062	0.913	-

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

gives the results of the path significance analysis of the model, in which the significance level of sub-section differentiated charging policy, sub-time period differentiated toll policy, and sub-payment mode differentiated charging policy is less than 0.05 (reaching the significance requirement), and the remaining three hypotheses do not reach the corresponding significance conditions.

Existing differential pricing policies have shown the highest acceptance rates for differential pricing based on road sections, which significantly impacts users’ intention to utilize these pricing models (p < 0.001). This is followed by differential pricing based on time slots and modes of payment. These strategies can effectively encourage users to opt for alternative routes by offering lower rates. The findings also indicate a significant variation in preferences for alternative route travel across different time periods and road sections. The more favorable the pricing policy, the stronger the users’ intention to utilize it. The differentiated toll policy based on direction, entry and exit points, and vehicle types should provide higher preferential rates to enhance user uptake. In conversations with some users, it became apparent that they were not aware of the differentiated toll policies based on direction and entry/exit points. As such, further clarification is needed for future policy implementation. Additionally, consideration could be given to integrating this differentiated pricing policy with the policies governing charges by road section and time period. It is also advisable to combine these initiatives with various forms of guidance to improve overall user acceptance.

5.2. LCL Model Analysis

Despite the presence of individual differences that cannot be directly observed, the LCL model effectively segmented the study population into three distinct categories: 47%, 36%, and 17%, by clustering respondents with similar attitudes and responses to the differential toll policy into the same category [49].

To further investigate the differences among these three driver categories, the values of two key parameters—toll rate and detour threshold (additional distance and time required to change routes to the destination)—were expanded separately. As illustrated in Figure 7, when the preferential rate is increased, the impact on the first category of users is more pronounced compared to the other two categories, while the effect on the third category is minimal. Conversely, when the detour distance and time are increased and accompanied by adequate guidance information, there is a negative impact on the first and third categories of drivers. However, this scenario yields a significant positive impact on the second category of drivers.

Based on these observations, users were classified into three categories: rate-sensitive, information-facilitating, and conservative refusal. This categorization underscores the significant heterogeneity among users. These findings provide valuable insights into how different user segments respond to toll rate changes and route guidance strategies, which can inform more tailored and effective policy implementations.

Drivers in the rate-sensitive category are primarily motivated by cost considerations. Even small changes in toll rates can significantly influence their travel decisions. These drivers tend to select the most economical travel options and reassess the cost-effectiveness of their routes when faced with increased detour distances or times. In contrast, information-facilitating drivers actively seek out real-time traffic information, including data on traffic flow and congestion, to accurately determine whether a detour is more beneficial. Providing clear and reliable route guidance enhances their trust in the alternative routes. On the other hand, drivers in the conservative refusal category are generally reluctant to change their driving routes. They are less inclined to take a detour, especially when faced with increased detour distances or travel times. These drivers will only consider altering their routes if there is clear evidence of cost savings. The results of the parameter estimation for the significant variables in the model are presented in

Table 4. Parameter estimation results for significant variables in the model.

Factor	Attributes	Rate Sensitive	Information Facilitating	Conservative Refusal
Individual socio-economic characteristics
Gender	Male	0.54 *** (4.66)	0.23 ** (2.08)	—
Age	20–35	0.64 *** (3.80)	—
	35–50	0.35 ** (2.20)	0.43 *** (2.77)	—
	>50	—	—	1.02 *** (2.00)
Trip-related characteristics
Travel frequency	almost everyday	0.82 ** (4.79)	−0.29 ** (−2.00)	—
	1–4 times per week	0.60 *** (3.49)	0.37 ** (2.29)	—
Type	Private car	2.66 ** (2.26)	2.03 ** (2.10)	−1.1 ** (−2.00)
	Truck	3.02 *** (6.32)	1.84 *** (2.86)
Travel time	Daytime	2.55 *** (10.16)	1.78 *** (3.73)	1.27 *** (3.44)
Travel range	Transit	1.27 *** (4.44)	0.7 *** (4.16)	0.46 *** (2.84)
Travel distance	100–200 km	0.72 ** (6.12)	0.11 ** (2.32)	−0.07 *** (−1.47)
Payment methods	ETC	0.56 *** (4.23)	0.26 ** (2.43)	—
Differentiated charging experience	Yes	0.11 *** (12.13)	0.12 *** (14.85)	−0.08 *** (−11.47)

Note: ** indicates p < 0.01, and *** indicates p < 0.001.

.

5.2.1. Analysis of Socio-Economic Characteristics

Figure 8 illustrates the distribution intervals of characteristic indicators for the three types of users. Specifically, Figure 8a,b encompass individual socio-economic characteristics, while Figure 8c–g cover trip characteristics.

(1) Gender

The experiment revealed that male drivers exhibit greater sensitivity to selecting alternative trips compared to female drivers. Male drivers consider factors such as cost, time, and efficiency, often opting for alternative routes to avoid high costs or traffic congestion. In contrast, female drivers prioritize safety, road conditions, and comfort, tending to maintain their original travel route to mitigate potential risks. Additionally, if the alternative road is narrower than the original route or the area has a high incidence of historical accidents, female drivers may feel uncomfortable, thereby reducing the likelihood of choosing an alternative trip.

(2) Age

Younger drivers are more inclined to select differentiated toll roads, are more cost-sensitive, and possess greater travel flexibility. As drivers age, they develop fixed travel habits and route selection preferences, becoming less receptive to complex and unfamiliar transportation environments. The probability of transition increases by 15% for drivers aged 20–35, while the probability of selection decreases by 20% for those aged 50 and older. However, older drivers may also alter their travel routes if the alternative route offers superior road conditions and scenic views.

For user groups with higher rate sensitivity, such as young male drivers, providing immediate and accurate road condition information can assist them in quickly and flexibly adjusting their route choices. Conversely, for safety-conscious and conservative user groups, such as female and elderly drivers, policies should emphasize service facilities and road conditions along alternative routes. It may even be advisable to recommend routes with higher tolls but better conditions to ensure they can enjoy a comparable or even superior travel experience compared to the original route.

5.2.2. Analysis of Trip Characteristics

(1) Vehicle type

The failure to distinguish between car models in the survey on willingness for differential tolls has led to the inference that this policy has a minimal impact on travel choices based on vehicle type. However, in our differentiated scenario experiment, we targeted specific car models and reminded participants to consider not just their own vehicle but also how the policy might affect other types of cars. Our findings reveal that when considered comprehensively, users are receptive to a tolling approach that varies based on vehicle type.

Passenger cars and special-purpose vehicles show less concern regarding differential tolls. Passenger cars, primarily used for transporting people, prioritize route safety and time predictability, preferring stable travel routes. On the other hand, special operation vehicles have distinct travel requirements, often needing to pre-report their routes, and are thus less influenced by variable tolling policies. Private vehicles, seeking convenience and cost-effectiveness, are more likely to opt for differentially tolled routes. In contrast, larger vehicles, particularly freight vehicles, demonstrate the highest sensitivity to alternative travel options. These vehicles typically incur higher toll costs and are significantly impacted by traffic congestion. Additionally, due to constraints such as road conditions, vehicle height, weight, and other factors, large trucks require more meticulous route planning. Consequently, they prioritize transportation costs and efficiency, favoring routes that offer economic benefits and reduced traffic congestion. Policy development should primarily consider these two user groups and be tailored to the specific characteristics of each vehicle type.

(2) Travel time

Drivers’ behavior in choosing alternative routes is notably influenced during weekday morning and evening peak hours. The intricate traffic situations and the high volume of trips during these times heighten the necessity for drivers to seek alternate routes. Interestingly, drivers appear more sensitive to selecting alternate routes during the morning peak compared to the evening. This could be attributed to the relatively stable traffic conditions and fewer time constraints in the evening, allowing drivers more time to deliberate on road conditions and the potential extra costs associated with alternate routes. As a result, they are more inclined to stick to their original travel route.

During peak hours, the focus should be on optimizing traffic flow by providing real-time road condition updates and intelligent navigation services. These tools can guide drivers to select more efficient travel routes. Additionally, emphasizing information transparency and offering detailed insights into road conditions and cost-effectiveness can assist drivers in making informed decisions.

(3) Travel frequency

Drivers who are well-acquainted with the road network tend to prefer easy-to-navigate routes to minimize complex maneuvers. For those who frequently travel on a specific highway, switching routes often entails navigating a more complicated road environment. As a result, they are more likely to stick to their original route. However, experienced drivers may become frustrated with congestion and delays and may opt for alternative routes in search of faster travel times. Lack of experience makes novice drivers usually rely more on traditional driving routes. Therefore, policies should consider implementing incentives to encourage users to explore these alternative routes.

(4) Traveling distance

Traveling short distances within the province generally incurs low costs, with travelers prioritizing time savings. They are open to exploring alternative routes if these can significantly alleviate congestion; otherwise, they will adjust their travel routes primarily for substantial rate discounts. For intra-provincial long-distance and transit journeys, tolls play a crucial role, and users are keenly attuned to the price variations among different routes. Transit users, in particular, consider additional factors such as route connectivity and the accessibility of services. In essence, even with favorable rates, transit users will not opt for routes that do not meet these prerequisites.

For short-distance users within the province, strategies to attract them to alternative routes should focus on offering rate concessions. For long-distance users, a tiered charging system or other incentives could be employed. For transit users, the priority should be enhancing the service level of alternative routes and ensuring the consistency of trip planning routes. This multifaceted approach aims to optimize route selection and improve overall user satisfaction.

(5) Payment Methods

Users who pay with ETC are more likely to accept differential tolls as an alternative to their usual travel routes. They appreciate the convenience of seamless, non-stop payments and are therefore more open to trying new routes. In contrast, users who rely on cash or other payment methods tend to be more cautious. The proportion of ETC payers willing to change routes is approximately 30% higher than that of non-ETC users. Policymakers can encourage more drivers to adopt ETC payments by offering incentives, such as favorable conditions that enhance the appeal of this payment method.

(6) Differentiated tolling experience

Positive experiences can significantly mitigate the impact of uncertainties. Individuals who have enjoyed favorable experiences with differential tolls are more likely to use them again, especially if they benefited from saved travel time or encountered improved road conditions during previous trips. Conversely, if their experiences were negative—such as facing traffic congestion or incurring unexpected costs—they are likely to react adversely to alternative routes in order to avoid similar issues.

For users with positive experiences, service policies should be further promoted and optimized to enhance the likelihood of their choosing alternative travel options again. Meanwhile, for users with negative experiences, the focus should be on addressing their dissatisfaction by improving service quality and ensuring transparency in information. This approach aims to reduce uncertainty and rebuild user trust.

5.3. Policy Implications

5.3.1. Transfer Intentions Under Different Rates

There are differences in the attractiveness of the different discount programs for passenger cars and trucks. Most vehicles are more satisfied with a 20% discount. An average discount rate of 20% can significantly increase the transfer probability of vehicle trips. As shown in Figure 9a, the most selected rate discount scheme for passenger cars is an 8.5% discount, accounting for 35.1%. When the discount reaches 8.5%, its cumulative transfer probability reaches 62.2%. As shown in Figure 9b, the rate discount scheme chosen most by trucks is 20% off, accounting for 23.2%. When the discount reaches 80% off, its cumulative transfer probability reaches 44.6%. It is also found that all models are less sensitive to discounts below 90%.

A review of vehicles opting for discounts of 60% or less indicates that they are primarily Type 1 passenger vehicles, Type 1 vans, and Type 6 vans. Although highways already provide higher discounts, drivers of Type 1 passenger vehicles and vans tend to prioritize controlling operating costs, making it more economically viable for them to choose free or cheaper travel routes. In contrast, Type 6 vans are well-suited for long-distance and large-scale transportation due to their high carrying capacity.

5.3.2. Information Pathways and Bypass Thresholds

The primary method for users to receive information about differentiated toll roads is through navigation software that offers route planning schemes, accounting for 75% of the statistics. This is followed by guidance provided through road traffic signs and markings. Users’ acceptable detour distances are predominantly centered within 30 km, with the next most common range being 30–45 km. As for acceptable travel time adjustments, these are heavily concentrated within 15–30 min. These time adjustments include both traveling earlier and arriving later. Among these, users prefer to opt for an earlier departure rather than delayed travel, though the dynamic time threshold for those choosing delayed travel is higher than for those opting for an earlier start. Figure 10 showcases the access and proportion of users to differentiated information. Figure 11a,b, respectively, illustrate the range of detour distances acceptable to users in comparison to their original trip and the range of trip time adjustments.

Providing road information during a trip interferes with drivers’ planning and inhibits their willingness to choose differentiated toll roads compared to providing such information before departure. The statistical results, depicted in Figure 12, indicate that giving information during the trip reduces the likelihood of drivers opting for alternative routes by an average of 21.7%. The overall impact of providing information before the trip is notably more significant in diverting users’ routes than doing so during the trip. However, offering reminders too far in advance did not significantly increase the proportion of users who diverted. For instance, providing information one hour in advance increased the proportion of transfers by only 3% compared to half an hour before the trip.

This phenomenon aligns with users’ behavioral inertia and their high priority for handling urgent information, reflecting users’ tendency to avoid inconvenience and their fear of the unknown when traveling [50]. Nevertheless, if real-time travel information is provided, users are also willing to adjust their travel plans according to actual conditions. As illustrated, providing real-time information increases the average probability of diversion by 17%. Therefore, drivers should be provided with route information as early as possible, and information push during the trip should be minimized. The information push during the trip should also consider users’ access to information channels, such as the occurrence of emergencies or other situations where real-time information can be pushed through highway variable message boards.

Additionally, during the morning and evening rush hours on weekdays, when highways carry a large number of urban commuter vehicles, traffic flow increases significantly, and the free travel of vehicles is restricted. Travelers in this context are more receptive to guidance information. More frequent releases of information on adjacent parallel roads or alternative routes could help alleviate traffic pressure.

6. Conclusions

To gain a comprehensive understanding of users’ acceptance and willingness to embrace the differentiated toll policy, this study examines the factors influencing their selection of differentiated routes for alternative trips, as well as assessing the impact of rate discounts on vehicle transfer intentions. By employing both revealed preference (RP) and stated preference (SP) survey methods, this research investigates users’ willingness to accept the current differentiated scheme and quantifies the proportion of users opting for differentiated road adjustment trips alongside their behavioral characteristics within simulated scenarios. Furthermore, a Structural Equation Modeling-Latent Class Logistics (SEM-LCL) model is constructed to delve into the influence mechanisms of differentiated toll road alternative travel choice behavior, considering user heterogeneity. The key findings of this study are as follows:

(1)

Different charging strategies and rate discounts appeal to users in distinct ways; lower rates in certain programs can encourage users to select alternative routes, while other programs require more substantial discounts to sway user choices. The current differentiated tolling scheme, which varies by section, time period, and payment method, has a more pronounced effect on users’ decisions regarding alternative travel routes, particularly influencing the choices of large trucks.

(2)

Users’ preferences for selecting differentiated routes for alternative trips exhibit significant heterogeneity. Based on the analysis of the SEM-LCL model, users can be categorized into three distinct types: rate-sensitive, information-promoting, and conservative-rejecting.

(3)

Users’ willingness to travel via differentiated route alternatives is significantly affected by various factors, including gender, age, experience, type of vehicle, travel time, travel distance, payment method, and familiarity with differentiated tolls.

By analyzing the underlying mechanisms that drive users to choose alternative travel routes, this study aims to provide valuable insights for highway managers in designing personalized toll rate strategies that encourage users to select more suitable driving routes. Additionally, it offers valuable insights for determining optimal toll rates and implementing dynamic traffic flow regulation. This, in turn, facilitates more strategic road network planning and helps alleviate traffic congestion, ultimately improving the overall efficiency of road operations.

Restricted by the data capacity and presentation of the questionnaire, the respondents may conceal their own relevant information and true thoughts when collecting data. In addition, the relatively small amount of data on special operation vehicles collected in the survey may lead to biased estimation results for these groups. The scope of the survey and user groups will be expanded in the follow-up to obtain more comprehensive data. Meanwhile, the survey could not fully simulate the route selection under real travel. Subsequent research suggests the use of driving simulators and other technologies to obtain more comprehensive behavioral data and to explore the selection behavior of users for alternative trips in terms of route characteristics, traffic conditions, and weather characteristics.