Understanding Electric Bicycle Users’ Mode Choice Preference under Uncertainty: A Case Study of Shanghai

Abstract

The electric bicycle is considered as an environmentally friendly mode, the market share of which is growing fast worldwide. Even in metropolitan areas which have a well-developed public transportation system, the usage of electric bicycles continues to grow. Compared with bicycles, the power transferred from the battery enables users to ride faster and have long-distance trips. However, research on electric bicycle travel behavior is inadequate. This paper proposes a cumulative prospect theory (CPT) framework to describe electric bicycle users’ mode choice behavior. Different from the long-standing use of utility theory, CPT considers travelers’ inconsistent risk attitudes. Six socioeconomic characteristics are chosen to discriminate conservative and adventurous electric bicycle users. Then, a CPT model is established which includes two parts: travel time and travel cost. We calculate the comprehensive cumulative prospect value (CPV) for four transportation modes (electric bicycle, bus, subway and private car) to predict electric bicycle users’ mode choice preference under different travel distance ranges. The model is further validated via survey data.

1. Introduction

In the past 20 years, the private ownership of electric bicycles has increased rapidly. China has the world’s largest electric bicycle market and is expected to have 300 million by 2025 [1]. Its cycling culture can be traced back to the 1980s and returned with fervor with electric bicycles gaining popularity for their savings of time and effort [1]. Take Shanghai as an example: trips by electric bicycle accounted for 20.2% among all trips in 2014 while, in 2004, the proportion was only 5.4% [2]. Aside from this, other regions of world, such as western Europe and North America, also witnessed a significant rise in electric bicycle sales [3]. It illustrates that the electric bicycle has become an indispensable means of transportation in city transportation systems.

Despite the growing deployment of electric bicycles, studies of the electric bicycle are still limited. Some studies look into the travel pattern and mobility of electric bicycles. Aguilera-García et al. [4] conducted a web-based survey in Spain to explore the usage of and citizens’ opinions towards the electric-bicycle-sharing system. Kazemzadeh et al. [5] modeled the navigation comfort of the electric bicycle in pedestrian crowds. Mohamed et al. [6] examined systematic differences in speed and road grade dynamics between electric and conventional bicycle trips. Moreover, several works discussed the equivalent unit estimation of electric bicycles. Chen et al. [7] used a concept method to calculate the bicycle equivalent of electric bicycles in terms of their time-space resource occupancy on traffic facilities. Jin et al. [8] proposed an innovative bicycle equivalent unit model for electric bicycles. Electric bicycle comfort and safety analysis is also a research direction. Chen et al. [9] investigated the overtaking disturbance caused by electric bicycles on non-motorized paths. Yu et al. [10] compared the frequency of electric bicycle riders’ risk-taking behaviors at signalized intersections with countdown signal devices (PCSDs) and intersections without PCSDs. However, most of the aforementioned literature treated electric bicycle users as a whole, thus failing to make a discrimination among users with different socioeconomic characteristics.

On the other hand, previously, studies on travel mode choice behavior were mainly based on the framework of expected utility theory (EUT), which adopts disaggregate logit models, including the multinomial logit model [11,12], rank-ordered logit model [13,14] and mixed logit model [15,16]. However, EUT assumes that decision-makers are completely rational under uncertain environments, which ignores many significant traveler characteristics such as risk preference and perception biases. Hence, conventional models using EUT cannot describe travelers’ behaviors accurately in reality. Kehneman and Tversky proposed the cumulative prospect theory (CPT) in 1992 [17]. This theory was initially applied to the economics field, aiming to describe investment behavior under uncertain conditions. Different from EUT, which deduces implication from normative preference axioms with the assumption of economic rational behavior of travelers, CPT considers individuals’ value of feelings in psychology and their attitude towards risk. In CPT, people tend to be risk-evasive towards gains and risk-seeking towards losses. Their evaluation is conservative for opportunity to increase gains, which proves to be more in line with reality [18]. Thus, CPT can be regarded as an amendment of EUT which is more descriptive.

According to CPT, the individuals’ decision-making process can be divided into two stages: editing and evaluation [18,19]. In the editing phase, outcomes of decision choices are calculated with a set of reference points, and then less outcomes are considered as losses and greater ones as gains. In the evaluation phase, the prospect value of each option, namely, the cumulative prospect value (CPV), is computed by value function and weighted function, and then decision-makers select the one with the greatest CPV among all the options.

In the transportation system, the decision-making process can be viewed as a selection process which is not rational and stable because travelers make judgments on many kinds of traffic situation. Thus, it is reasonable to apply CPT in modeling travelers’ decision-making behavior because this theory has successfully described the whole decision-making process under uncertain conditions. For instance, it has been widely applied in routing choice and departure time choice modeling [20,21,22,23]. However, few studies focused on the CPT-based travel mode choice behavior. Meanwhile, the body of the limited relevant literature paid more attention to public transport and cars. Lam et al. [24] proposed a convex prospect theory-based model to investigate the effects of population density on travelers’ mode choice behavior under an advanced transportation information system (ATIS) in a multimodal transportation corridor. An et al. [25] analyzed two kinds of car owner’s generalized subjective perception costs on four different transportation modes, including bus, subway, taxi and private car, and calculated the mode choice’s prospect value before and after the implementation of congestion pricing. However, the mode of electric bicycle was not considered in the above literature.

On these grounds, this paper attempts to quantify the influence of electric bicycle users’ psychological characteristics, as well as risk attitudes, on their mode choice behavior. Based on survey results, electric bicycle users are classified according to their socioeconomic characteristics. In total, six socioeconomic characteristics are involved in the paper. Prospect theory is applied to describe electric bicycle users’ choice tendency for four different transportation modes, including bus, subway, private car and electric bicycle. The comprehensive CPV model takes both travel time and travel cost into consideration. By comparing the CPVs of four travel mode choices, the paper aims to predict electric bicycle users’ future mode choice behavior and find out whether they will change and shift to public transport.

2. Methodology

Figure 1 shows the framework of the proposed CPV-based mode choice model. Because travel time and travel cost are two vital factors with respect to travel mode choice, this paper builds a CPT model including these two factors together. The distribution of travel time and travel cost (per kilometer) were acquired from the citywide resident trip survey. The reference points were derived from the stated preference (SP) survey, which represents people’s expected travel time and cost per kilometer. Then, we calculated CPVs for all the travel modes investigated based on the corresponding value function and weighting function. The mode with the largest CPV was the one chosen.

2.1. Value and Weighting Functions

The value functions for travel cost and travel time are constructed separately: travel distance is $l$; electric bicycle users’ expected travel speed is $v_0$; expected travel time is $T_0$. Given a certain travel distance $l$, the travel speed of travel mode $i$ may have multiple outcomes with different probabilities. Denote that the $k^{th}$ outcome of the speed of travel mode $i$ is $v_{i,k}$ and the corresponding travel time is $T_{i,k}$ and speed of travel mode $i$. Then, the difference value between $T_{i,k}$ and $T_0$ is:

$$\Delta T_{i,k}=T_0-T_{i,k}=\frac{l}{v_0}-\frac{l}{v_{i,k}}$$

(1)

When $\Delta T_{i,k}\ge 0$ ($v_{i,k}\ge v_0$), the psychological perception of electric bicycle users in terms of travel time appears to be profitable. When $\Delta T_{i,k}<0$ ($v_{i,k}<v_0$), they feel the loss. Value function of $T_{i,k}$ can be described as (14):

$$V(\Delta T_{i,k})=\{\begin{array}{ll}{(\Delta T_{i,k})}^{\alpha }& \Delta T_{i,k}\ge 0\\ -\lambda {(-\Delta T_{i,k})}^{\beta }& \Delta T_{i,k}<0\end{array}$$

(2)

where $\lambda$ is loss aversion parameter ($\lambda >1$). A larger $\lambda$ value means people are more sensitive to losses; $\alpha$ and $\beta$ are diminishing sensitivity parameters ($0<\alpha ,\beta <1$). Small $\alpha$ and $\beta$ values mean people are risk-seeking, while larger values indicate that they are risk-averse.

Supposing the probability that $T_{i,k}$ occurs is $P_{i,k}$, then the weighting function of $T_{i,k}$ can be obtained:

$$\pi (P_{i,k})=\frac{P_{i,k}{}^{\gamma }}{P_{i,k}{}^{\gamma }+{(1-P_{i,k}{}^{\gamma })}^{\frac{1}{\gamma }}}$$

(3)

where $\gamma$ ($0<\gamma <1$) is the curvature parameter for the weighting function.

Similarly, assume that travel distance is $l$; electric bicycle users’ expected travel cost per kilometer is $c_0$; expected total travel cost is $C_0$; and travel cost per kilometer of travel mode $i$ is $c_i$. Note that, for a certain travel mode, the unit travel cost is relatively fixed. Hence, the probability that $c_i$ occurs is 1. Then the difference value between $C_i$ and $C_0$ is:

$$\Delta C_i=C_0-C_i=(c_0-c_i)\times l$$

(4)

When $\Delta C_i\ge 0$ ($c_i\le c_0$), the psychological perception of electric bicycle users in terms of travel cost appears to be profitable. When $\Delta C_i<0$ ($c_i>c_0$), they feel the loss. Value function of $C_i$ can be described as:

$$V(\Delta C_i)=\{\begin{array}{ll}{(\Delta C_i)}^{\alpha }& \Delta C_i\ge 0\\ -\lambda {(-\Delta C_i)}^{\beta }& \Delta {\mathrm{C}}_i<0\end{array}$$

(5)

Supposing the probability that $C_i$ occurs is ${\tilde{P}}_i$, then the weighting function of $C_i$ can be obtained:

$$\pi ({\tilde{P}}_i)=\frac{{\tilde{P}}_i^{\gamma }}{{\tilde{P}}_i^{\gamma }+{(1-{\tilde{P}}_i^{\gamma })}^{\frac{1}{\gamma }}}$$

(6)

The discussion on the model parameters will be presented in the following section.

2.2. Determination of Parameter Values

Kahneman and Tversky employed a certainty equivalent to estimate the values of CPV parameters [17]. The result was: $\alpha =\beta =0.88$, $\lambda =2.25$, ${\gamma }_{gain}=0.61$, ${\gamma }_{loss}=0.66$. Some researchers adopted parameter values above [26]. However, other scholars argued that experimental results in a certain country may not be fully applicable worldwide. Mason claimed that people with different socioeconomic characteristics have various attitudes towards risk [27]. The disparate risk preferences may bring out typical values of reference point, which, in turn, have a great impact on parameter values in the model. In consequence, uniform parameter values cannot reflect the actual choice behavior. Additionally, Zeng et al. found that parameter values of Chinese people ($\alpha =1.21$, $\beta =1.02$, $\lambda =2.25$, ${\gamma }_{gain}=0.55$, ${\gamma }_{loss}=0.49$) differed from those of Americans [28].

Furthermore, some scholars set five pairs of $\alpha$ and $\beta$ and drew corresponding value function curves for each pair [29]. They compared value function curves with the utility curves of people with various risk preferences. It turned out that gain segments of value functions were in line with utility curves. To be specific, the utility curve for people with a risk proneness (RP) attitude was similar to the value function curve using Kahneman and Tversky’s parameter values. In addition, the utility curve of people with a risk aversion (RA) attitude was in accordance with the value function curve using parameter values from Zeng’s research [30]. With the above in mind, this study discriminated between conservative and adventurous electric bicycle users by using different parameter values. We adopted two sets of parameter values for users with different risk preference, namely, $\alpha =1.21$, $\beta =1.02$, $\lambda =2.25$, ${\gamma }_{gain}=0.55$, ${\gamma }_{loss}=0.49$ for conservative ones and $\alpha =\beta =0.88$, $\lambda =2.25$, ${\gamma }_{gain}=0.61$, ${\gamma }_{\mathrm{loss}}=0.66$ for adventurous ones.

2.3. CPV Calculation

Denote the speed vector for mode $i$ as $v_i=\{v_{i,-m},\dots ,v_{i,n}\}$, where $v_{i,-m}<\dots <v_0<\dots <v_{i,n}$. Then express corresponding probability as $P_i=\text{{}{P}_{i,-m},\dots ,P_{i,n}\}$. As a result, the travel time component of CPV for travel mode $i$ is:

$$CPV(T_i,P_i)={\displaystyle \sum _{k=0}^{n}V(\Delta T_{i,k})}{\pi }^{+}(P_{i,k})+{\displaystyle \sum _{k=-m}^{-1}V(\Delta T_{i,k}){\pi }^{-}(P_{i,k})}$$

(7)

where ${\pi }^{+}(P_{i,k})$ is the weighed function when $\Delta T_{i,k}$ is no less than 0 and ${\pi }^{-}(P_{i,k})$ is the weighed function when $\Delta T_{i,k}$ is smaller than 0.

Denote the travel cost vector for mode $i$ as $C_i=\{C_{i,-m},\dots ,C_{i.n}\}$, where $C_{i,-m}<\dots <C_0<\dots <C_{i,n}$, then the travel cost component of CPV for travel mode $i$ is:

$$CPV(C_i)={\displaystyle \sum _{k=0}^{n}V(\Delta C_i})+{\displaystyle \sum _{k=-m}^{-1}V(}\Delta C_i)$$

(8)

Considering travel time and travel cost together, obtain the comprehensive CPV of travel mode $i$:

$$CP{V}_i=w_1.CPV(T_i,P_i)+w_2.CPV(C_i)$$

(9)

where ${\omega }_1$and ${\omega }_2$ are weights of travel time and travel cost acquired from survey. They can be calculated as follows:

$$w_1=\frac{u_1}{N}$$

(10)

$$w_2=\frac{u_2}{N}$$

(11)

where $N$ is the total number of people surveyed and $u_1$ is the number of people who think time is more important than cost while $u_2$ is the opposite.

The probability that electric bicycle users choose travel mode $i$ is:

$${\stackrel{⌢}{P}}_i=\frac{e^{CP{V}_i}}{{\displaystyle \sum _{i=1}^M{e}^{CP{V}_i}}}$$

(12)

where $M$ means the number of travel mode options. The larger the CPV of the mode is, the larger the probability of it being selected by the decision-makers the mode has.

3. Case Study Set-Up

3.1. Data Collection

A questionnaire survey was conducted in 2017. Travelers who used electric bicycles frequently were defined as respondents and asked to fill in questionnaires. Note that free-floating bike-sharing systems were not implemented at that time, thus not making an impact. There were 14 survey sites in total. Seven sites were in an urban district while others were in a suburban district. We sent 1200 questionnaires in total and received 1068 back. The final available sample size was 1023. The questionnaire contained four parts: (a) individual’s socioeconomic characteristics: region, gender, age, educational level, income per month and private car ownership; (b) respondent’s emphasis on travel cost and travel time; (c) respondent’s risk attitude; and (d) respondent’s expected commuting travel cost and travel time.

In order to facilitate model construction and result comparison in the following part, we classified electric bicycle users into two groups in accordance with the six socioeconomic characteristics presented in

Table 1. Socioeconomic characteristics of survey samples.

Attribute	Range	Proportion
Region	Urban district	54%
	Suburban district	46%
Gender	Male	51%
	Female	49%
Age	20–30	13%
	30–40	37%
	40–50	33%
	50–60	17%
Education level	Junior high school or below	11%
	Senior high school	26%
	Undergraduate	54%
	Postgraduate or above	10%
Income per month	<2000 RMB	13%
	2000–4000 RMB	44%
	4000–6000 RMB	30%
	>6000 RMB	13%
Private car ownership	Own one or more cars	42%
	Do not own a car	58%

: (1) region: urban district users or suburban district users, (2) gender: male users or female users, (3) age: young users (20–40 years old) and old users (40–60 years old), (4) education level: highly educated users (undergraduate or above) and low-educated users (senior high school or below), (5) income per month: low-income users (<4000 RMB per month) and high-income users (≥4000 RMB per month) and (6) private car ownership: users who do not own a car and users who own one or more cars. The average monthly income for electric bicycle users of Shanghai was RMB 3878. Meanwhile, the percentage of people who earned more than RMB 4000 per month was 42% [1]. Taking these factors together, RMB 4000 was chosen as the standard to classify high-income users and low-income users.

3.2. Risk Attitudes and Weights

In the survey, we classified the respondents into two groups each time according to their socioeconomic characteristics. Then, we asked them whether they liked to take risks during their trips. As shown in

Table 2. Risk attitudes of different groups of users.

Attribute	Group	Risk Preference
		RP (%)	RA (%)
Region	Urban district	62.0	38.0
	Suburban district	48.8	51.2
Gender	Male	67.5	32.5
	Female	51.3	48.7
Age	Young	68.0	32.0
	Old	48.9	51.1
Education level	Low educated	55.9	44.1
	Highly educated	62.7	37.3
Income per month	Low income	45.3	54.7
	High income	67.5	32.5
Private car ownership	Own one or more cars	61	39
	Do not own a car	47.3	52.7

, for example, RP users occupied the larger percentage in urban districts (62% vs. 48.8%). Hence, we treated the electric bicycle users in urban districts as adventurous users and those in suburban districts as conservative ones. In the same way, users who were young, well-educated, with a high income and that owned private cars were regarded as adventurous travelers. The surprising result that highly educated people and high-income people were more aggressive in travel behavior can be explained as follows: highly educated ones have more curiosity and creativity to try a different travel mode to get their destination. As for high-income people, compared with low-income ones, they dared to choose an unfamiliar mode even if it costed more. The weights of the time and cost components of the CPV are also shown in

Table 3. Weights of time and cost components.

Attribute	Group	Weight (%)
		${\mathit{w}}_1$	${\mathit{w}}_2$
Region	Urban district	0.608	0.392
	Suburban district	0.576	0.424
Gender	Male	0.612	0.388
	Female	0.585	0.415
Age	Young	0.760	0.240
	Old	0.528	0.472
Education level	Low educated	0.433	0.567
	Highly educated	0.702	0.298
Income per month	Low income	0.551	0.449
	High income	0.803	0.197
Private car ownership	Own one or more cars	0.688	0.312
	Do not own a car	0.570	0.430

.

3.3. Reference Point Selection

Given a certain travel distance, the values of reference points can be calculated, respectively, through expected travel speed and expected travel cost per kilometer. Considering that travel distance may influence travelers’ perception of reference points, we initially investigated expected travel speed and expected travel cost per kilometer separately according to the range of travel distance (presented in

Table 4. Uniform speed and cost of different groups of users.

Group	Distance (km)		Uniform Speed	Distance (km)		Uniform Cost
	≤10	10–20		≤10	10–20
Urban district	17.32	17.85	17.50	0.28	0.21	0.25
Suburban district	18.35	19.43	19.00	0.28	0.16	0.22
Male	18.05	19.52	18.80	0.27	0.24	0.25
Female	16.80	18.57	17.70	0.26	0.22	0.24
Young	17.48	18.93	18.20	0.27	0.23	0.25
Old	15.48	17.52	16.50	0.25	0.20	0.22
Low educated	15.81	16.25	16.00	0.21	0.17	0.19
Highly educated	18.28	18.89	18.50	0.35	0.25	0.30
Low income	16.09	16.28	16.20	0.18	0.14	0.16
High income	17.50	18.67	18.00	0.33	0.29	0.31
Own one or more cars	18.93	20.02	19.50	0.34	0.28	0.31
Do not own a car	16.87	17.23	17.00	0.26	0.19	0.23

The unit of speed is km/h and unit of cost is RMB/km.

). From the table, we calculated that there was little difference between the short-distance trip (≤10 km) and the long-distance trip (10–20 km). As a result, the uniform speed and cost (average value of long- and short-distance trips) were chosen and used as the reference points.

3.4. Model Validation

In order to validate the accuracy of the proposed CPT-based model, we compared the probabilities of choosing four travel modes calculated by the proposed model with those acquired from survey. Possible speed values were set according to the Fifth Shanghai Traffic Survey [1] as follows: 15 km/h and 20 km/h for electric bicycle with a probability of 60% and 40%, 15 km/h and 20 km/h for bus with a probability of 30% and 70%, 25 km/h and 30 km/h for subway with a probability of 60% and 40% and 30 km/h and 35 km/h for private car with a probability of 70% and 30%. As shown in

Table 5. Validation of CPT model.

Mode	Urban District Users (%)			Suburban District Users (%)
	Survey Result	Model Result	Bias Value	Survey Result	Model Result	Bias Value
Electric bicycle	27.0	25.3	1.7	36.6	38.5	1.9
Bus	18.5	20.7	2.2	20.3	17.3	3.0
Subway	41.5	44.1	2.6	23.3	27.9	4.6
Private car	13.0	9.9	3.1	19.8	16.3	3.5
Mode	Male Users (%)			Female Users (%)
	Survey Result	Model Result	Bias Value	Survey Result	Model Result	Bias Value
Electric bicycle	33.0	30.2	2.8	30.0	25.8	4.2
Bus	17.0	20.0	3.0	21.7	26.0	4.3
Subway	36.2	34.7	1.5	40.0	37.5	2.5
Private car	13.8	15.1	1.3	8.3	10.7	2.4
Mode	Young Users (%)			Old Users (%)
	Survey Result	Model Result	Bias Value	Survey Result	Model Result	Bias Value
Electric bicycle	32.0	30.4	1.6	30.9	33.1	0.2
Bus	20.1	17.8	2.3	23.6	20.7	2.9
Subway	26.6	30.6	4.0	34.4	30.2	4.2
Private car	21.3	21.2	0.1	11.1	16.0	4.9
Mode	Low-Educated Users (%)			Highly-Educated Users (%)
	Survey Result	Model Result	Bias Value	Survey Result	Model Result	Bias Value
Electric bicycle	33.8	31.2	2.6	30.1	34.9	4.8
Bus	25.0	20.9	4.1	16.1	11.2	4.9
Subway	22.8	26.1	3.3	39.0	35.7	3.3
Private car	18.4	21.8	3.4	14.8	18.2	3.6
Mode	Low-Income Users (%)			High-Income Users (%)
	Survey Result	Model Result	Bias Value	Survey Result	Model Result	Bias Value
Electric bicycle	33.5	35.3	1.8	28.8	23.6	5.2
Bus	21.2	26.1	4.9	16.9	18.6	1.7
Subway	29.2	26.5	2.7	38.1	36.3	1.8
Private car	16.1	12.1	4.0	16.2	21.5	5.3
Mode	Users Own Private Cars (%)			Users Do Not Own Private Cars (%)
	Survey Result	Model Result	Bias Value	Survey Result	Model Result	Bias Value
Electric bicycle	26.3	23.0	3.3	35.4	40.2	4.8
Bus	16.9	14.1	4.8	21.2	19.8	1.4
Subway	21.9	25.3	3.4	41.5	36.5	4.0
Private car	34.9	37.6	2.7	1.9	3.5	1.6

, biases were all within 5.5%, so the proposed model was validated.

4. Results and Discussion

In this part, we calculated comprehensive CPVs of four travel modes by substituting various values of travel distance l (l = 5, 10, 15 and 20 km, respectively). On the basis of the six socioeconomic characteristics selected in this study, mode choice preference of electric bicycle users with different demographics were discussed and compared. Then, we realized to what extent electric bicycles replaced the share of other modes of transportation and what role electric bicycles played in a comprehensive transportation system.

To start with, the attributes of region and gender were considered. Just as in the results illustrated in Figure 2a,b, with travel distance increasing, the comprehensive CPV of bus for users in a suburban district declined while it increased for users in an urban district. The reason is that the density of the suburban bus network is relatively low, thus reducing accessibility. In addition, the layout of bus stops is too irrational to cover the main residential areas of suburban district. Aside from this, the vehicle condition is bad in outskirts. Compared with bus, the subway seemed to be more competitive for long-distance trips. In urban districts, the comprehensive CPV of subway exceeded 0 when travel distance reaches 10 km, while, in the outskirts, the corresponding distance was around 15 km. It states that the subway is highly accepted among users living in the downtown area. Additionally, there was a difference between male and female electric bicycle users concerning private car trips (shown in Figure 2c,d). Compared with males, the comprehensive CPV of the private car dropped faster according to females; that is to say, the longer the travel distance, the lower the probability for female users to choose a private car. Generally, females may not adapt to long-distance drives as well as males. From this perspective, the electric bicycle seems easier to master for females, seeing as its comprehensive CPV only dropped slightly with the increase of travel distance. Especially when the travel distance was within 15 km, the electric bicycle was superior to the other three travel modes. Only when the distance exceeded 15 km, the electric bicycle dropped to second place.

Age and private car ownership also influence travel mode choice significantly (labeled in Figure 3a,b). For old electric bicycle users, comprehensive CPV of electric bicycle remains stable with a slight increase when travel distance extends. The value is only second to subway when travel distance goes beyond 15 km. It proves that electric bicycle occupies an important role in old users’ daily trips. Meanwhile, in view of increasing proportion of old people among electric bicycle users currently, electric bicycle is likely to become the main means of transport for the old. This finding is similar to Lee’s survey [30] conducted in Netherland where bicycle market is prosperous. He suggested that the old may have difficulty riding a traditional bicycle because of physical problems, the electric bicycle offers an opportunity for them to stay on the bikeways instead of switching to automobiles. Also, considering their economic level and physical condition, old users prefer to choose travel mode that is cheap and door-to-door. Unlike old people, young users give priority to subway whose travel time is relatively stable because it is not affected by road conditions. Moreover, the comprehensive CPV of electric bicycle for electric bicycle users who do not own a car are greater (shown in Figure 3c,d). It means whether electric bicycle users own private cars has a great impact on their dependence on electric bicycles.

Finally, users with different education and income levels were compared (shown in Figure 4). In the light of the survey results, income has a positive relationship with education level. As a result, there were some similarities between high-income respondents and well-educated ones. Compared with the travel cost, high-income and well-educated users paid more attention to convenience and quickness. From the survey, we also find that high income and well-educated users are more likely to have private cars which may explain the car’s predominance for long-distance travel. For people who earn less money or are less educated, the comprehensive CPV of the electric bicycle kept rising, which goes against the common sense that this kind of vehicle is more suitable for short-medium travel. This result has much to do with the occupation characteristic of people surveyed because, among these two groups of people, the proportion of respondents who work for a delivery service accounted for nearly 20%. They usually use electric bicycles to deliver packages or food.

Based on the discussion and comparison above, some conclusions can be drawn as follows:

(1)

Old electric bicycle users may continue to choose the electric bicycle because of its advantage in saving physical strength. This kind of vehicle keeps them on the bike lanes for visiting relatives or recreation in the neighborhood;

(2)

The relatively high distance tolerance of the electric bicycle is found among low-income and low-educated groups. However, as travel distance extends, its comparative advantage gradually diminishes;

(3)

Electric bicycle users have realized that the electric bicycle is not suitable for long-distance trips and they do not repel choosing other means of transport according to specific needs;

(4)

When the travel distance is not quite so long (within 15 km), the electric bicycle is superior to the other three travel modes for suburban and female users;

(5)

Whether electric bicycle users have private cars has a great impact on their electric bicycle dependence.

5. Conclusions and Future Work

In many cities of China, the number of electric bicycles even surpasses traditional ones [31]. The role the electric bicycle plays in the urban transportation system remains controversial. On the one hand, it may become the barrier to the government generalizing mass transit. On the other hand, to some extent, the presence of the electric bicycle promotes motorization, thus appealing to people who used to adopt non-motorized modes. Compared with high-energy-consuming vehicles, such as cars or scooters, their environmental burden is relatively low. Moreover, the demographics of respondents denote that electric bicycle users in China are not only middle class or lower-income populations. This phenomenon coincides with Cherry’s study [4], which found electric bicycle users had significantly higher income and education attainment than traditional bike users. In the future, relevant policies can focus on specific strategies that guide the use of electric bicycles, making it the feeder to public transport or the substitution for cars.

CPT can describe an individual’s decision-making characteristics and an individual’s psychological status under uncertain situations. As a result, the application of CPT in modeling travelers’ mode choice behavior is more suitable for measuring a traveler’s psychological expectations, as well as their attitude towards risk.

The progress of the research can be concluded in the following aspects:

• Supposing that individual socioeconomic characteristic make a difference in electric bicycle users’ travel expectations, the study divided them into various groups and investigated expected travel time and travel cost separately. Thus, reference points in the model varied among different groups based on the risk attitudes;
• The study did not apply the traditional parameters used in Kahneman and Tversky’s experiment to all user groups. Instead, two sets of values were used, respectively, for conservative and adventurous users, which better explain their attitude under uncertain and risk conditions;
• The study adopted CPT to analyze electric bicycle users’ mode choice preference of four different means of transportation. By comparing the comprehensive CPVs of each travel mode, we saw the role electric bicycles play in transportation systems.

This work still had some limitations which can be studied in the future. The proposed model only had two components: travel time and travel cost, but failed to consider the influence of travel purpose, departure time and safety on electric bicycle users’ travel mode choice. Furthermore, users’ perception of comfort could not be analyzed due to the mechanism of CPT. In addition, the typical values of speeds used to calculate the comprehensive CPVs of the four travel modes were from the Fifth Shanghai Traffic Survey [2] investigating all the travelers in Shanghai. They may lack pertinence for electric bicycle users.