Impact of Carpooling under Mobile Internet on Travel Mode Choices and Urban Traffic Volume: The Case of China

Abstract

This paper firstly analyzes the comparative advantages of carpooling under the mobile Internet and traditional travel modes, including buses, private cars and taxis, as well as the differences between carpooling under the mobile Internet and traditional carpooling, so as to obtain the factors that affect travelers’ mode choices. Secondly, the mixed logit model is used to describe the travelers’ travel mode choice behavior, which effectively avoids the limitations of the IIA characteristics and preference randomness of the logit model. Finally, we conducted an SP survey on 1077 samples online and offline. After eliminating some invalid samples, 984 valid ones were obtained. Based on these survey data, we analyze the impacts of carpooling under the mobile Internet on the mode shares of traditional travel modes. The results show that for different trip lengths, carpooling under the mobile Internet has different degrees of substitution for buses, taxis and private cars. That is to say, travelers who previously chose buses and other modes will shift to carpooling due to the mobile Internet. In addition, in most cases, the emergence of carpooling under the mobile Internet is helpful to reduce the traffic volume in the urban road network, thus alleviating the urban congestion. However, when the trip length is short and the seat utilization ratio of carpooling under the mobile Internet is low, carpooling under the mobile Internet will increase the traffic volume.

1. Introduction

Carpooling (i.e., the act in which two or more travelers share the same car for a common trip) is one of the possibilities brought forward to reduce traffic and its externalities [1]. In some countries, the demand for carpooling fluctuates with oil prices and energy sources, while in China, the traditional method of carpooling only occurs between neighborhoods in the early stage of carpooling development, and few people choose this mode. However, carpooling began to gradually enter the life of people in China following the emergence of intelligent car-hailing APPs under the mobile Internet. As car-hailing software (e.g., the Didi APP) was gradually developed and put onto the market, China’s Internet-based car-hailing market officially presented “blowout” growth. Figure 1 shows the scale of the carpool users in China from 2016 to 2020. This is because carpooling under the mobile Internet can integrate the information of car owners and travelers through smart phone car-hailing APPs, which thus realizes the efficient matching of supply and demand and facilitates the carpooling of travelers with one click [2,3].

The emergence of the mobile Internet has enabled carpooling to take advantage of lower prices than taxis, and it provides a better user experience than buses. In addition, since 2015, transport commissions in some provinces and cities have granted business licenses to major mobile Internet carpooling platforms. On 28 July 2016, the Ministry of Transport of China issued the Interim Measures for the Management of Online Car-pooling Operation and Service, which gradually determined the legality of carpooling under the mobile Internet. In April 2020, the Ministry of Transport of China issued the Interim Measures for the Management of Online Booking Taxi Business Services, which set out the operation standards and specific requirements for the online booking of taxis (including ridesharing services) and license application. In January 2021, the Ministry of Transport of China issued the Regulations on the Management of e-Hailing Drivers, which set out detailed regulations and requirements on the qualifications, safety, insurance and other aspects of e-hailing drivers in ridesharing services. The introduction of these policies has further improved the driver qualification, safety risks, insurance and other aspects of ridesharing services, and further promoted the development of carpooling under the mobile Internet.

In the information society, the popularity of the mobile Internet and mobile payment provides a new physical basis for carpooling. Carpooling under the mobile Internet not only provides convenient services for travelers, but it also effectively improves the seat utilization of cars. Therefore, it is considered an important way to effectively reduce the volume of traffic in the urban road network and alleviate the congestion [4,5,6]. However, the users attracted by carpooling under the mobile Internet are not only those who were used to choosing taxis or private cars, but also those who were used to choosing buses, which means that ridesharing changes travelers’ mode choice behavior, and the impact of carpooling on traffic is difficult to estimate. Therefore, in order to make better use of this travel mode, it is necessary to describe the impacts of carpooling under the mobile Internet on travel mode choices. Then, the impacts of carpooling under the mobile Internet on traffic volumes can be determined by analyzing the proportion of carpooling users who shift from the bus, private car and taxi modes.

In addition, in order to accurately depict travel mode choice behavior, it is necessary to determine the factors that affect travelers’ mode choices. On the one hand, factors influencing travelers’ choices can be obtained by analyzing the relative differences between carpooling and traditional travel modes, including buses, private cars and taxis. On the other hand, there is a significant difference between travelers’ acceptance of carpooling under the mobile Internet and traditional carpooling, which means that the factors with large differences between the two types of carpooling may also be the key to influencing travelers’ choices.

To sum up, this paper analyzes the comparative advantages of carpooling and buses, private cars and taxis under the mobile Internet, as well as the relative differences between carpooling under the mobile Internet and traditional carpooling, so as to determine the factors that affect travelers’ mode choices. Then, a mixed logit model is established to reflect persons’ mode choice preferences. Based on SP survey data of urban residents’ travel choice preferences under the mobile Internet, the impacts of carpooling on the modal splits in a city and the change in the traffic volume on road networks are studied. Thus, we provide the basis for the formulation of related policies for carpooling.

2. Literature Review

Recently, with the development of the sharing economy and the popularity of mobile Internet technology, Internet-based carsharing has rapidly developed. The existing literature is mainly focused on the impacts on commuting behavior [7,8], and on the overall impact on urban traffic [9,10,11].

2.1. Literature on Travel Choice Behavior

There is much literature on travel mode choice behavior, and the most representative model is the logit model derived by Luce [12]. At present, there are many achievements in the research on the choice behavior of drivers and passengers in commuter carpooling.

Correia et al. found that time convenience and cost saving were the two main factors influencing passengers’ choice of carpooling, while other objective factors (such as age, gender, etc.) affect passengers’ preferences [13,14]. Cao analyzed the influencing factors on travel strategies and travel mode choices, and argued that travelers’ travel decisions are mainly influenced by subjective factors, such as subjective willingness, satisfaction, loyalty and lifestyle, and objective factors, such as socioeconomic and demographic variables [15]. Studies by Long and Gardner et al. showed that travelers will rely on the reliability, extra time and physical exertion of carpooling (e.g., the time and physical consumption of walking or riding a bike from the starting point to the carpooling point) to consider whether to choose carpooling [16,17]. Amirkiaee and Evangelopoulos found that the most important factors influencing people’s participation in ridesharing relate to cost savings, environmental concerns and the desire for more social and pleasant modes of travel. When constructing the behavioral model of the carpooling choice, we should consider not only the driving cost and time cost in the car, but also the physical and psychological costs. Some scholars have conducted studies on the extra costs incurred in carpooling [18]. Correia and Delhomme et al. believed that a key factor that hinders private car drivers from choosing carpooling is the psychological loss caused by it. The psychological loss cost is induced by the loss of privacy and the uncertainty of the travel schedule [5,13].

Neoh first found that the factor that affects carpooling that residents are most concerned about is safety; that is, when urban residents share a car with strangers, they will bring psychological barriers due to safety concerns [19]. Vanoutrive et al. believed that although carpooling has the advantage of saving travel costs, in many cases, the cost advantage of saving is not enough to make up for the loss of comfort and flexibility caused by carpooling, and many private car drivers are reluctant to share their car space with strangers due to safety considerations [20]. In addition, these scholars studied the impact of the difficulty in obtaining travel information before the popularization of the mobile Internet, and they found that if the trust between drivers and passengers increases, then the flexible matching of information between ridesharing service providers and customers will increase the sharing rate of ridesharing trips [19,20].

These studies explored the various factors that affect travelers’ choice to use carpooling, such as cost savings, security, reliability and personal preference. However, this field lacks research on the dissimilarities between carpooling under the mobile internet and traditional carpooling. At the same time, it does not consider the differences between the carpooling travel mode and other travel modes under the mobile Internet, such as information symmetry, designated pickup locations and times, diversified payment methods and other aspects.

2.2. Literature on Urban Transportation Systems

The impact of carpooling and ridesharing services on urban transport systems is a topic that has garnered significant attention in the academic literature in recent years. Many studies have explored the various ways in which these services can impact traffic congestion, as well as the fuel consumption and access to transportation.

In terms of traffic congestion, Fagnant and Kockelman found that ridesharing services can either reduce or increase congestion, depending on the specific context and usage patterns [21]. Wang analyzed the development of urban carpooling policies in China and the differences between China and Western countries, and found that the implementation of effective carpooling policies is conducive to reducing traffic congestion in urban areas and promoting sustainable mobility [10]. Li et al. believe that on-demand ridesharing has the potential to reduce congestion, but, at the same time, stress the need for effective policies to regulate the growth and operation of such services [22]. Shaheen et al. argued that carpooling is a sustainable mode of travel, and that it has the potential to reduce the number of single-occupancy vehicles on the road, thus reducing traffic congestion in cities [23,24]. Schaller found that ridesharing services such as Uber and Lyft can help reduce traffic, but the author also highlights the need for policies to promote and regulate the use of these services [25]. In contrast, Henao and Marshall found that ride-hailing services increase the vehicle miles traveled (VMT) in urban areas but reduce the number of trips and miles traveled by private cars, which may cause traffic congestion [26].

Minett and Pearce found that carpooling reduces transportation energy consumption. In San Francisco, random carpooling saves from 1.7 million to 3.5 million liters of gasoline per year [27]. Seyedabrishami used the case of Tehran, Iran, and the impact of carpooling on fuel consumption was evaluated in terms of increasing the average number of passengers and reducing the distance that passengers travel. The results showed that carpooling would reduce fuel consumption by about 240 million liters per year [28]. Shaheen and Cohengave analyzed an overview of ridesharing services in North America and their potential benefits, including reducing congestion, improving traffic conditions for people without private cars and reducing emissions [29]. Si et al. found that government regulation, carbon emission reduction certification and information disclosure can have positive impacts on ridesharing, which, in turn, can play a role in reducing transport-related emissions and promoting sustainable transport practices [30].

In terms of access to transportation, carpooling and ridesharing services have been found to improve mobility for individuals who previously lacked convenient and affordable travel options. The Shared Use Mobility Center (2016) conducted a survey of 4500 users of shared mobility services and found that frequent users of carpooling services were also often frequent users of public transport and intermodal travelers. Most rides for carpooling services occur during bus-free hours between 10 p.m. and 4 a.m. This suggests that ridesharing services can provide travel options for people who cannot use public transport at night, thus filling a gap that public transport cannot cover [31]. Schwieterman and Smith found that both ride hailing and bus services play a role in meeting the travel needs of Chicago residents, while emphasizing that ride-hailing services such as UberPool can complement traditional travel services, address travel needs and improve urban transportation accessibility [32]. Barajas and Brown found that ride-hailing services are more accessible in areas with convenient public transportation, but are less accessible in cross-regional travel, and that ride-hailing services can help improve the accessibility of cross-regional travel [33]. Young, Allen and Farber found that when the travel time difference between the two modes is small, ride hailing is more likely to be used as a substitute for transportation, and when the travel time difference is large, ride hailing is more likely to be used as a supplement [34]. These findings suggest that ride-hailing services can complement traditional transportation services and improve urban transportation accessibility. Jin, Kong and Sui found that in wealthy areas, car-hailing services are easier to obtain, and online car-hailing services may compete with public transportation to provide better transportation services to urban residents [35]. Jiang, Chen, Mislove and Wilson argued that the competition between ride-hailing services, such as Uber and Lyft, improves accessibility, but they also stress the need for further research into the long-term impacts of this competition on accessibility and sustainability [36]. Public transit may be losing riders as the share of carpooling services increases: a study of seven large U.S. metro areas showed that these services tend to substitute for 6% and 3% of the trips that would have been otherwise made by bus and light rail, respectively [37].

These studies discuss the impact of carpooling on urban transportation systems, and particularly on traffic congestion, fuel consumption and accessibility. Research has found that ridesharing can reduce the number of single-person vehicles on the road, reduce fuel consumption and complement traditional transportation services, improving urban transportation accessibility. However, these studies lack the complementarity and competition between carpooling and other modes of transportation, and they do not consider the impact of carpooling on traffic volumes under different travel distances. Therefore, we aim to determine the impact of ridesharing on the traffic volume under the mobile Internet by analyzing the proportion of ridesharing users who switch from buses, private cars and taxis.

3. Factors Affecting Travel Mode Choice Behavior

In order to determine the factors affecting travelers’ mode choices, this paper analyzes the comparative advantages of carpooling and buses, private cars and taxis under the mobile Internet, as well as the differences between traditional carpooling and carpooling under the mobile Internet.

3.1. Comparative Advantage of Carpooling under Mobile Internet

Firstly, carpooling under the mobile Internet is a travel mode based on private cars, which can replace private cars to a certain extent. This is because, for urban employees, if they choose private cars for daily commuting, the high fuel cost consumed by cars and the daily maintenance cost of cars will put great economic pressure on them [38]. However, carpooling under the mobile Internet has changed this situation to some extent. Private car owners take advantage of the existing free seat resources and search for travelers with the same or similar travel routes on smart phone APPs under the mobile Internet. After completing the travel service, they can receive partial remuneration and relieve the economic pressure. At the same time, due to the convenience of carpooling, some owners of private cars will give them up and choose to use carpooling under the mobile Internet to reduce travel costs. Therefore, reducing travel costs is the primary reason for most private car owners to choose carpooling under the mobile Internet.

Secondly, carpooling under the mobile Internet also affects the demand for taxis. Taxis have always been known for their high costs, and it is difficult to take a taxi during the morning and evening rush hours. However, carpooling under the mobile Internet solves these problems somewhat. First, the cost of carpooling under the mobile Internet can be reduced by at least 30% compared with the cost of taxis. Secondly, carpooling under the mobile Internet can be used to book a car in advance, which can alleviate the difficulty of taking a taxi and the uncertainty of boarding buses. Therefore, some travelers will switch to carpooling under the mobile Internet.

Thirdly, some travelers who previously chose buses will shift to carpooling because of the short time and lower cost. The travel time by bus mainly includes the walking time to the stops, waiting time, on-board time and walking time to the destinations. Buses usually detour due to the requirements of stopping stations and will not choose the shortest path to the destination, and so the travel time is increased, while carpooling reduces the travel time.

In addition, carpooling under the mobile Internet is carried out by smart phone, and information posting and browsing are required; therefore, ages and genders will have great differences in the acceptance of the carpooling mode under the mobile Internet.

Based on the above analysis, it can be seen that factors such as travel time, cost, gender and age are important factors that influence travelers’ preferences for different travel modes. Therefore, this paper first takes time, cost, gender and age as the variables of the travel mode choice. Moreover, private cars, taxis, buses and carpooling through the mobile Internet are selected as the four available modes.

3.2. Difference between Carpooling under Mobile Internet and Traditional Carpooling

In order to analyze the factors of carpooling under the mobile Internet that cause changes in travel mode choices, this paper analyzes the differences between the traditional carpooling mode and the carpooling mode under the mobile Internet.

First of all, before the emergence of APPs on the mobile Internet, carpooling was basically implemented between neighbors, which is a small-range travel mode. Under the mobile Internet, car owners who provide ridesharing services can see the travel demands of travelers on smart-car-booking APPs, and travelers can see the information published by car owners seeking travelers. That is to say, intelligent APPs under the mobile Internet can fully integrate the information of car owners and travelers, improve the symmetry of the information and thereby enhance the convenience of ridesharing services.

Secondly, under the mobile Internet, carpooling can specify the vehicle arrival place and time on the smart APPs according to customers’ travel requirements. In addition, the diversification of payment methods is also a key factor to encourage travelers to choose carpooling through the mobile Internet. Online ridesharing supports various forms, such as online payment and offline payment, which are convenient and fast.

However, although carpooling has many advantages, safety problems associated with carpooling under the mobile Internet have gradually been exposed with the increasing popularity of carpooling [39]. This is due to the unfamiliarity between car owners and passengers, which leads to the inability to guarantee the safety of some passengers. Although the intelligent ride-hailing platform provides one-button alarm service, it cannot completely avoid the occurrence of danger. Therefore, the safety of customers has become a key factor for travelers to consider when choosing this travel mode.

The difference between carpooling under the mobile Internet and traditional carpooling is also the reason for the development of carpooling. Therefore, the above five characteristics are selected as the factors that influence the travelers’ choice of travel mode in this paper.

4. Model

4.1. Mode Choice Model

At present, many researchers have adopted multiple logit models and nest logit models to analyze the mode shares of traditional travel modes [40,41], but the MNL model has IIA characteristics and the limitation of random preference, which is somewhat contradictory to the fact. The so-called IIA feature (that is, the mode shares of two travel modes) is only related to these two travel modes but has nothing to do with other travel modes, which is inconsistent with the actual situation. For example, according to the irrelevance of choice probability, carpooling reduces the choice probabilities of buses and taxis. There is a substantial similarity between carpooling and taxi pooling in convenience, but there is a less obvious similarity between buses and taxis. As a result, the reduction degree of the bus choice probability differs from that of the taxi choice probability. In the nested logit model, the IIA assumption is relaxed within each nest, but it is still assumed to hold across nests [42,43]. This means that the correlation among alternatives within the same nest is allowed, but the correlation across different nests is assumed to be zero.

In addition, in terms of preference randomness, if each parameter to be estimated in the utility function of the logit model is fixed, then the characteristics of the preference randomness of different surveyed persons cannot be expressed. For example, some persons may be more sensitive to time, and some may be more sensitive to price. Then, the sensitivity of the utility of the parameters of these corresponding variables should be higher [44]. On this basis, we use the mixed logit model to analyze the changes in the mode shares of different travel modes in different situations. The mixed logit model assumes that the estimated parameters follow a certain distribution, such as normal distribution or lognormal distribution. This allows for a better consideration of individual heterogeneity [42,45], as well as the ability to handle both continuous and discrete variables [45]. The model also accounts for the interaction between different factors and the heteroscedasticity of probability [42,45], while reflecting the randomness of personal preferences and avoiding the defect of ratio independence [46].

In the mixed logit model, the probability of individual n choosing travel mode i is as follows:

$$P_{ni}={\displaystyle \int \frac{e^{V_{ni}\left({\beta }_n,x_{ni}\right)}}{{\displaystyle \sum _j{e}^{V_{nj}\left({\beta }_n,x_{nj}\right)}}}f\left({\beta }_n\right)}$$

(1)

where $B_n$ is the parameter being estimated; $V_{ni}$ is the utility function; $f(B_n)$ is a distribution function; $x_{ni}$ is an attribute variable.

In this study, we focus on the study of the impacts of carpooling under the mobile Internet on the mode shares of several traditional travel modes. Therefore, the choice set is determined as buses, taxis, private cars and carpooling under the mobile Internet (hereinafter referred to as carpooling):

$$A_n=\left\{bus,taxi,car,carpooling\right\}$$

(2)

Secondly, it is necessary to determine the attribute variables in the utility function ($V_{ni}$). The attribute variables here are mainly divided into three parts. The first is traveler characteristics, which are gender ($x_m^1$) and age ($x_m^2$). The second is travel characteristics, including travel time ($x_m^3$) and travel cost ($x_m^4$). The third is the variables generated under the mobile Internet, which are information symmetry ($x_m^5$), the boarding place can be specified ($x_m^6$), the boarding time can be specified ($x_m^7$), the diversification of payment methods ($x_m^8$), customer security ($x_m^9$), etc.

In this paper, based on the research on the choice behavior of carpooling travelers [16,17,40], income was not included as a variable due to concerns about privacy and the possibility of incomplete or reluctant responses in income-related surveys, which may have reduced the efficacy of the data samples. Nevertheless, the significant findings of the study suggest that the model can effectively explain the carpooling behavior among travelers in the mobile Internet era, even without the inclusion of income as a variable.

Therefore, the utility function of an individual (n) to the alternative (i) is as follows:

$$U_{ni}=AS{C}_{ni}+{\beta }_{ni}^k{x}_{ni}^k+{\epsilon }_{ni}$$

(3)

where $AS{C}_{ni}$ is the alternative inherent constant, representing the impacts of the inherent attributes of the travel mode itself on the utility, and it takes a fixed value; ${\epsilon }_{ni}$ is a random term of utility.

In the classical logit model, $\beta$ is a fixed parameter and $\epsilon$ follows a Gumbel distribution, but in the mixed logit model, $\beta$ has random elements to satisfy individual differences. Therefore, we assume that the travel time, travel cost and five elements under the mobile Internet follow lognormal distribution. Normally, the expected values of the travel time and travel cost parameters are negative (that is, the larger the value, the smaller the utility), while the value of lognormal distribution is in the non-negative interval, and so it is necessary to set the symbol change of the attribute value in the calculation. Therefore, the utility functions of the four different modes are determined as follows:

$$V_{n1}=AS{C}_{n1}+{\beta }_{n1}^1{x}_{n1}^1+{\beta }_{n1}^2{x}_{n1}^2+{\beta }_{n1}^3{x}_{n1}^3+{\beta }_{n1}^4{x}_{n1}^4$$

(4)

$$V_{n2}=AS{C}_{n2}+{\beta }_{n2}^1{x}_{n2}^1+{\beta }_{n2}^2{x}_{n2}^2+{\beta }_{n2}^3{x}_{n2}^3+{\beta }_{n2}^4{x}_{n2}^4+{\beta }_{n2}^5{x}_{n2}^5+{\beta }_{n2}^6{x}_{n2}^6+{\beta }_{n2}^7{x}_{n2}^7+{\beta }_{n2}^8{x}_{n2}^8+{\beta }_{n2}^9{x}_{n2}^9$$

(5)

$$V_{n3}=AS{C}_{n3}+{\beta }_{n3}^1{x}_{n3}^1+{\beta }_{n3}^2{x}_{n3}^2+{\beta }_{n3}^3{x}_{n3}^3+{\beta }_{n3}^4{x}_{n3}^4$$

(6)

$$V_{n4}=AS{C}_{n4}+{\beta }_{n4}^1{x}_{n4}^1+{\beta }_{n4}^2{x}_{n4}^2+{\beta }_{n4}^3{x}_{n4}^3+{\beta }_{n4}^4{x}_{n4}^4+{\beta }_{n4}^5{x}_{n4}^5+{\beta }_{n4}^6{x}_{n4}^6+{\beta }_{n4}^7{x}_{n4}^7+{\beta }_{n4}^8{x}_{n4}^8+{\beta }_{n4}^9{x}_{n4}^9$$

(7)

where i = 1, 2, 3, 4 represent the modes of buses, taxis, private cars and carpooling, respectively.

The vectors of the estimated parameters in the mixed logit model follow a certain distribution form, which reflects the randomness of individual travel mode preferences and avoids the restriction of the logit model’s preference randomness. In the mixed logit model, the rate independence defect caused by the logit model IIA property is also avoided. The cross elasticity between the choice probability ($P_{ij}$) of travel mode j and the mTH variable of travel mode n is shown in Equation (8):

$$E_{ij{X}_{in}^m}=\raisebox{1ex}{$\frac{\partial P_{ij}}{P_{ij}}$}\!\left/ \!\raisebox{-1ex}{$\frac{\partial X_{in}^m}{X_{in}^m}$}\right.=-{\displaystyle \int {\beta }^m{L}_{in}\left(\beta \right)\left[\frac{L_{ij}\left(\beta \right)}{P_{ij}}\right]f\left(\beta |\theta \right)d\beta }$$

(8)

where

$$L_{ij}\left(\beta \right)=\frac{e^{V_{ij}}}{{\displaystyle \sum _j{e}^{V_{ij}}}}$$

(9)

It can be seen that, assuming that $\beta$ follows a distribution function, when the mth variable of travel mode j changes, the change percentages of the choice probabilities of the other travel modes are different, and the difference mainly depends on the variable and the distribution function of $\beta$. Therefore, it can be concluded that the mixed logit model perfectly overcomes the defect of rate independence caused by the IIA.

4.2. Algorithm

To obtain the value of $\beta$, integration is commonly used; however, it was found that the integration operation in this model is very complicated, and so it is convenient to use a simulation method to solve it. Therefore, this paper intends to use the simulation algorithm of maximum likelihood function estimation to solve it (namely, to find the travel mode with the maximum probability of being selected according to the overall probability or sample probability).

The simulation solution steps of the mixed logit model are as follows:

Step 1: Obtain the simulation probability.

(1) On the basis that $\theta$ is known, a random vector ($f(\beta |\theta )$) is randomly selected from the density function ($\beta$), denoted as ${\beta }^r$, and r = 1 when first selected;

(2) Calculate the value of $L_{ij}({\beta }^r)$ according to Equation (9);

(3) Repeat Steps (1) and (2) R times to calculate the mean value of $L_{ij}({\beta }^r)$ as the simulation probability:

$$P_{ij}=\frac{1}{R}{\displaystyle \sum _{r=1}^R{L}_{ij}}\left({\beta }^r\right)$$

(10)

Step 2: Construct the maximum likelihood operator.

(1) Record the sample size as I, select the number of alternatives as J and define the auxiliary variables:

$$y_{ij}=\{\begin{array}{l}1\begin{array}{cc}& \mathrm{Individual}i\mathrm{chose}\mathrm{alternative}j\end{array}\\ 0\begin{array}{cc}& else\end{array}\end{array}$$

(11)

(2) The simulation likelihood function of the sample is shown in Equation (12):

$$SL\left(\beta \right)={\displaystyle \prod _{i=1}^I{\displaystyle \prod _{j=1}^J{P}_{ij}^{y_{ij}}}}$$

(12)

(3) Take the logarithmic form of Equation (12) (that is, obtain the maximum likelihood simulation operator):

$$SLL\left(\beta \right)={\displaystyle \sum _{i=1}^I{\displaystyle \sum _{j=1}^J{y}_{ij}\ln P_{ij}}}$$

(13)

Step 3: Solve parameters ($\theta$).

By changing the parameters ($\theta$) until the maximum likelihood operator of simulation, the required maximum value can be obtained. The Newton–Rapson and other methods can be used to solve the problem.

5. Case Study

5.1. SP Survey

To obtain the data, we carried out an SP survey. The designed questionnaire is mainly divided into three parts: the first is the survey of the travelers’ personal attributes; the second is the SP survey contents, which investigate the choices of travel mode of travelers with different commuting distances under the traditional and mobile Internet contexts; the third is the derivative contents of the survey. Because the above five factors brought about by the mobile Internet need to be quantified, this paper sets the preference survey as follows: for the five elements, give scores of 1–9, and select the value of each option that the persons being surveyed think affects their mode choices, among which 1 represents the least attractive choice, and 9 represents the mode that is most attractive.

The survey was implemented both online and offline for 7 days, and a total of 1077 samples were collected. After eliminating some invalid samples, 984 effective samples were obtained (validating rate = 91.36%). In the sample, the proportion of males was 53.4%, and that of females was 46.6%, which is close to the actual situation. In terms of personal age, the average age of the surveyed group was 33.7 years old, with 14.7% aged from 18 to 20 years; 28.9% aged from 20 to 29 years; 27.7% aged from 30 to 39 years; 17.9% aged from 40 to 49 years; 10.9% aged from 50 to 59 years. All ages conform to the national regulations on motor vehicle driving and the use of mobile phone payment.

Before the calculation, it was necessary to sort out the data of the questionnaire survey to facilitate the assignment of the required attribute variables.

Table 1. Mixed logit model property variable assignment table.

Variable	Meaning	Value
$AS{C}_{ni}$	Inherent variable	The inherent variable of private car utility is 1, that of taxis and carpooling is 0.5, and that of buses is 0.1.
$x_{ni}^1$	Gender	Male = 1; female = 2.
$x_{ni}^2$	Age	Age of 18–20 = 1; age of 20–29 = 2; age of 30–39 = 3; age of 40–49 = 4; age of 50–59 = 5.
$x_{ni}^3$	Travel cost	Bus = RMB 1; taxi = RMB 10 + (distance − 3) × 2; car = RMB 18.6 + 0.52 × distance × 1; carpooling = RMB (18.6 + 0.52 × distance)/3.
$x_{ni}^4$	Travel time	By distance, with 5 km as an example: bus = 0.4 h; taxi = 0.2 h; private car = 0.2 h; carpooling = 0.25 h.

shows the results.

5.2. Data Analysis

The probability of the mixed logit model needs to be solved by maximum likelihood estimation simulation. Here, Nlogit software is used to calculate the parameters. The Halton method is used to complete the extraction of the random parameter ($\beta$), with the intended extraction times at R = 500. The calibration results are shown in

Table 2. Parameter calibration results.

Parameters	Nonmobile Internet			Mobile Internet
	Parameter Value	Standard Deviation	p-Value	Parameter Value	Standard Deviation	p-Value
$AS{C}_1$	−0.245	0.326	0.048	−0.323	0.694	0.037
$AS{C}_2$	0.652	0.539	0.019	0.478	0.562	0.001
$AS{C}_3$	0.972	0.427	0.034	0.875	0.347	0.000
$AS{C}_4$				0.596	0.559	0.000
${\beta }^1$	0.187	0.126	0.065	0.219	0.207	0.078
${\beta }^2$	0.128	0.096	0.010	0.232	0.102	0.032
${\beta }^3$	−2.791	2.523	0.000	−3.014	2.031	0.000
${\beta }^4$	−0.128	0.796	0.000	−0.097	0.652	0.000
${\beta }^5$				0.974	0.069	0.000
${\beta }^6$				0.762	0.106	0.002
${\beta }^7$				0.896	0.178	0.000
${\beta }^8$				0.571	0.184	0.001
${\beta }^9$				0.124	0.253	0.000

.

The results in Table 2 are those of two scenarios: the first is that of traditional carpooling as an alternative; the second is that of carpooling under the mobile Internet as an alternative. The results mainly include the estimated parameters, standard deviations and p-values. The parameters from $AS{C}_1$ to ${\beta }_2$ are the values of the fixed variables.

In these two scenarios, the parameters of the fixed variables, except for gender, are statistically significant. Secondly, the parameter of the inherent variable of buses is negative, indicating that travelers are more willing to choose taxis and private cars than buses under the nonmobile Internet. Travelers are more willing to choose taxis, private cars and carpooling; the parameters from ${\beta }_3$ to ${\beta }_9$ of the variables are statistically significant. Among them, the parameters of travel cost and travel time are negative, indicating that the increase in the travel cost and travel time of a travel mode will reduce its utility to some degree.

In addition, from ${\beta }_5$ to ${\beta }_9$ are the parameters of the unique factors of carpooling under the mobile Internet that cause changes in travel mode choices. It was found that the parameter of information symmetry is the largest, which further verifies the promoting effect of the Internet environment on the growth of the carpooling mode.

5.3. Different Distance Scenarios

In order to clarify the impacts of carpooling under the mobile Internet on the mode shares of traditional travel modes, we calculated the modal splits based on the above parameters under different trip-length scenarios. In different scenarios (namely, when the trip length is 5 km, 10 km, 15 km and 20 km), there are changes in the modal splits of the different modes. The changes in mode shares under different scenarios are shown in Figure 2.

As can be seen in Figure 2, with the increase in the trip length, the mode share of carpooling gradually increases, which means that this mode will bring an obvious convenience to long-distance commuting trips, especially in large or megacities.

Figure 2a–c compares the modal splits under the nonmobile Internet and mobile Internet of trips of different lengths. The numbers in the figure are the reduction rate of the mode share of each mode under two situations. Figure 2d is the mode share of carpooling under the mobile Internet. It can be clearly seen from Figure 2 that regardless of whether the length is 5 km, 10 km, 15 km or 20 km, carpooling under the mobile Internet has an obvious impact on the three traditional modes.

First of all, when the length is 5 km, the mode shares of buses, taxis and private cars under the mobile Internet decrease by 12.72%, 6.78% and 6.59%, respectively, compared with those under the nonmobile Internet, and the decline ratios of the mode shares of these three modes are exactly as the mode share of carpooling under the mobile Internet, which is 26.09%. The reduced ratio of buses is the largest, indicating that when the trip length is 5 km, carpooling under the mobile Internet has the greatest impact on the mode share of buses (that is, carpooling has a stronger substitution effect on buses when the trip length is 5 km). This means that, in cases of short-distance trips, carpooling under the mobile Internet mainly serves the users who were used to choosing buses. In this case, carpooling under the mobile Internet can mainly improve the travel experience of these users, but its effect on reducing the traffic volume is relatively weak, which will be discussed further in Section 5.4.

Secondly, when the trip length increases to 10 km, the mode share of carpooling is 29.62%. In this case, carpooling under the mobile Internet has the greatest impact on taxis, the mode share of which decreases by 12.71% compared with in the case of the nonmobile Internet. However, the influence of carpooling under the mobile Internet on the mode share of private cars changes little compared with when the trip length is 5 km.

In addition, when the trip length is 15 km and 20 km, the private car is the mode that is most influenced by carpooling under the mobile Internet. Compared with that under the nonmobile Internet, the mode share of private cars decreases by 16.34% and 22.01%, respectively. This indicates that when the travel distance is 15 km and 20 km, more than half of private car travelers are willing to shift to carpooling (that is, carpooling has a stronger substitution effect on private cars). Among them, when the trip length is 20 km, the mode share of carpooling under the mobile Internet is the largest, which is 41.69%. This means that carpooling under the mobile Internet can effectively reduce the trips of private cars in cases of long-distance trips. For large or even megacities with heavy long-distance commuting trips, carpooling is a key way to reduce road congestion. Therefore, it is necessary to introduce policies to promote carpooling.

In short, no matter how far the pooling distance is, carpooling under the mobile Internet will attract travelers who were used to choosing buses, taxis and private cars, and in cases without the mobile Internet, travelers will choose carpooling, which also shows that the prevalence of carpooling under the mobile Internet will alleviate some urban traffic problems to a certain degree.

5.4. Impacts on Urban Traffic Volume

In order to analyze the impact of carpooling under the mobile Internet on traffic volumes, we first regarded car and taxi trips as one unit. For example, when carpooling is selected, no matter how many passengers are on board, it will be regarded as one unit of traffic. A trip by private car or taxi is also recorded as a unit of traffic. At this time, we assume that the volume of bus vehicles in the road network does not change, and that the seat utilization ratio of private cars or taxis is 1.36 persons/car [47]. For different trip lengths and seat utilization ratios, the rates of change in the traffic volume in the network are shown in

Table 3. Rates of change in traffic volume under different scenarios.

SUR (Person/Car)	Travel Distance
	5 km	10 km	15 km	20 km
2	10.8%	−10.1%	−21.6%	−22.4%
3	−17.3%	−31.2%	−38.9%	−39.5%
4	−31.4%	−41.8%	−47.6%	−48.0%
5	−39.8%	−69.3%	−73.9%	−74.3%

Note: SUR–seat utilization ratio.

.

It can be seen from Table 3 that, in most cases, carpooling under the mobile Internet will help reduce the traffic volume. With the increase in the trip length, carpooling under the mobile Internet will reduce the traffic volume even more significantly. At the same time, the seat utilization ratio of carpooling under the mobile Internet is also an important factor affecting urban traffic volumes. The higher the seat utilization ratio, the larger the reduction in the urban traffic volume. However, when the seat utilization ratio and trip length are both small (e.g., the seat utilization ratio is two persons/vehicle and the trip length is 5 km), carpooling under the mobile Internet will increase the traffic volume because travelers who choose carpooling under the mobile Internet are mainly those who are used to choosing the bus as their travel mode. Namely, some short-distance travelers who are used to choosing buses will become passengers of carpooling under the mobile Internet and induce additional traffic volume.

6. Conclusions

Based on SP survey data, this paper analyzes the factors that affect travelers’ travel mode choices, constructs a mixed logit model to describe the travel mode choice behaviors and then analyzes the changes in the urban traffic volume due to the development of carpooling under the mobile Internet. It can be found that:

(1) Under the mobile Internet, information symmetry, determined pick-up sites and times, diversified payment methods and customer safety are the five factors that affect travelers’ mode choices;

(2) When the trip distance differs, carpooling under the mobile Internet plays a different role in substituting for other travel modes. With the increase in the trip distance, carpooling plays a stronger role in substituting for the other three travel modes: when the trip distance is 5 km, it is buses that are mostly influenced; when the trip distance is 10 km, it is taxis that are mostly influenced; when the trip distance is 15 km and 20 km, it is private cars that are mostly influenced;

(3) In most cases, carpooling under the mobile Internet will help to reduce traffic volumes on urban roads. With the increase in the length of travelers’ single trips and the increase in the seat utilization ratio of carpooling, carpooling under the mobile Internet will reduce the traffic volume more significantly.

Therefore, the government should consider improving the mode share of carpooling for long-distance trips in various ways, encourage travelers to choose carpooling and improve the seat utilization ratio of carpooling under the mobile Internet, such as by adding urban HOV (high-occupancy vehicle) lanes, or providing subsidies to car drivers who provide carpooling services to ease traffic congestion [48,49].