Scholarly article on topic 'Dynamics in Mode Choice Decisions: A Case Study in Nanjing, China'

Dynamics in Mode Choice Decisions: A Case Study in Nanjing, China Academic research paper on "Civil engineering"

CC BY-NC-ND
0
0
Share paper
Academic journal
Procedia Engineering
OECD Field of science
Keywords
{"Travel Mode Choice" / Dynamics / "Traffic State" / "Congestion Level" / "Discrete Choice Model"}

Abstract of research paper on Civil engineering, author of scientific article — Ling Ding, Ning Zhang

Abstract This study aims to estimate dynamics in mode choice decisions at different traffic states and transit congestion levels. Specifically, dynamics were examined over time of explanatory variables such as in-vehicle time and comfort of transit. The trips to the central business district (CBD) in Nanjing City of China were taken as a case study. Three travel modes were investigated: bus, metro, and car. Travelers’ socioeconomic characteristics and alternative specific attributes were collected through a reveal preference (RP) survey. A multinomial logit (MNL) model was proposed using RP Data1 from the questionnaire survey. It was found that in-vehicle time of cars and buses varied with traffic states. In addition, congestion level was divided by passengers per carriage to obtain the comfort of transit. Then, the mode choice decisions at different traffic states or congestion levels were estimated using a MNL model to analyze the dynamics. MNL analysis on the mode choice decisions revealed that those who own cars prefer auto trips. The income influence was also confirmed that individuals with high income prefer driving. The predicted mode choice decisions were compared with the actual choices to evaluate the model. Some possible reasons were explored to examine the mispredictions. Last, a comparison among different departure time with reference to their utilities of choosing modes revealed that traffic state and congestion level of transit took a significant effect on mode choice decisions. The proposed model had important implications to study travel behavior to improve the service of transit although some limitations in the model, such as only one mode determining rule, one transportation environment. The result of this prediction, however, can be viewed along with the results of other studies to obtain a development of dynamic travel behavior.

Academic research paper on topic "Dynamics in Mode Choice Decisions: A Case Study in Nanjing, China"

CrossMark

Available online at www.sciencedirect.com

ScienceDirect

Procedía Engineering 137 (2016) 31 - 40

Procedía Engineering

www.elsevier.com/locate/procedia

GITSS2015

Dynamics in Mode Choice Decisions: A Case Study in Nanjing,

Ling Dinga,%, Ning Zhanga

intelligent Transportation System Research Center, Southeast University, 35 Jinxianghe Road, Nanjing, 210008, China

Abstract

This study aims to estimate dynamics in mode choice decisions at different traffic states and transit congestion levels. Specifically, dynamics were examined over time of explanatory variables such as in-vehicle time and comfort of transit. The trips to the central business district (CBD) in Nanjing City of China were taken as a case study. Three travel modes were investigated: bus, metro, and car. Travelers' socioeconomic characteristics and alternative specific attributes were collected through a reveal preference (RP) survey. A multinomial logit (MNL) model was proposed using RP Datal from the questionnaire survey. It was found that in-vehicle time of cars and buses varied with traffic states. In addition, congestion level was divided by passengers per carriage to obtain the comfort of transit. Then, the mode choice decisions at different traffic states or congestion levels were estimated using a MNL model to analyze the dynamics. MNL analysis on the mode choice decisions revealed that those who own cars prefer auto trips. The income influence was also confirmed that individuals with high income prefer driving. The predicted mode choice decisions were compared with the actual choices to evaluate the model. Some possible reasons were explored to examine the mispredictions. Last, a comparison among different departure time with reference to their utilities of choosing modes revealed that traffic state and congestion level of transit took a significant effect on mode choice decisions. The proposed model had important implications to study travel behavior to improve the service of transit although some limitations in the model, such as only one mode determining rule, one transportation environment. The result of this prediction, however, can be viewed along with the results of other studies to obtain a development of dynamic travel behavior. © 2016PublishedbyElsevierLtd. Thisisanopenaccess article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the Department of Transportation Engineering, Beijing Institute of Technology Keywords: Travel Mode Choice, Dynamics, Traffic State, Congestion Level, Discrete Choice Model

* Ling Ding. Tel.: +86-13805159632; E-mail address:dinglingseu@gmail.com

1877-7058 © 2016 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the Department of Transportation Engineering, Beijing Institute of Technology doi:10.1016/j.proeng.2016.01.231

1. Introduction

Mode choice estimation and analysis play a crucial role in the decisions of transportation strategies, such as the implementation of transit priority and congestion pricing. The determination of strategies should take account into the real-time traffic states. For instance, managed bus lanes are often set in peak hours in many cities. Travelers' mode choice decisions can be affected by some dynamic factors such as travel time, cost and comfort. Previous studies on mode choice decisions have successfully investigated the decision variations of different individual travelers. However, the dynamics due to the variations of different traffic states within a day attracted little attention.

Analysis on mode choice decisions have been widely investigated in previous literatures using logit models [1-3]. The relationship between travelers' mode choice decisions and the contributing factors has been explored in many studies [4-7]. The contributing factors include individuals' socioeconomic characteristics and the alternative specific attributes. The socioeconomic characteristics mainly refer to travelers' personal characteristics such as gender, age, income, and car ownership. The alternative specific attributes mainly involve with the travel cost, travel time, travel comfort, etc. The limitation of these studies with cross-sectional mode choice is that they did not take account into the dynamic and longitudinal properties of the contributing factors at different time intervals within a day. For instance, the travel time at peak and regular hours can be very different. This can also greatly affect travelers' mode choice decisions. The literatures about dynamic and longitudinal mode choice decisions have been sparsely reported due to the challenges of data collection and modelling [8]. Pas and Koppelman [9] developed and examined the day-to-day variability of urban travel behaviors, whereas Burnett [10] and Golledge [11] explained the variability in shopping behavior and location choice. Kitamura [12] examined behavioral dynamics using panel data with long intervals. The benefits of panel analysis compared to traditional cross-sectional approaches were well claimed in this study. While these studies on mode choice decisions mostly focused on the long-term dynamics, such as the dynamic travel behaviors from week to week but not the variability within a day. It is widely accepted that individuals' travel behaviors vary within a day because of the effects of previous activities. The dynamics among multiple mode choice decisions by individuals on a given day were studied [8, 13]. Ramadurai [8] and Wan [14] investigated the within-day dynamics and variations in mode choice decisions using the models based on activity-travel. It showed that current travel mode choice was affected by past choice decisions. While this paper will examine the dynamics in mode choice decisions from another perspective. It focuses on the dynamics due to the variations of contributing factors, such as travel time, cost, and comfort, during different traffic states within a day.

2. Motivations and objectives

This study is motivated by the following aspects. First, mode choice decisions are made by respondents on the same day. Second, mode choice decisions are influenced by socioeconomic characteristics and alternative specific factors. A third, the alternative specific factors are dynamic within a day, for example, in-vehicle time and comfort vary along with traffic state or transit congestion level. In addition, the variation of in-vehicle time and comfort are reflected in travel utilities so that individual's mode choice decisions changed.

Given these motivating considerations, the objectives of this study are: 1) to identify traffic state and congestion level of transit during different periods within a day; and 2) to propose a within-day dynamic mode choice model. The variations of in-vehicle time and comfort along with traffic states and congestion level will be investigated. The variations in mode choice decisions within a day with different states will also be examined in this study.

To achieve these objectives, a multinomial logit (MNL) model is utilized to analyze mode choice decisions. The model is estimated using respondents' travel data in Nanjing, Jiangsu. To capture dynamics, the mode choice decisions at different traffic states and transit congestion levels for the same individual are predicted.

This study is distinct from the literatures on mode choice decisions in following respects: It examines within-day dynamics in mode choice decisions considering dynamic contributing factors, whereas most previous studies analyzed cross-sectional or long-term dynamic, and few studies focused on the within-day dynamic based on activity;

Traffic states and transit congestion levels are identified in this study; and the influences of state-varying variables, such as in-vehicle time and comfort, on mode choice decisions are examined.

3. Research plan

This study aims to investigate the dynamics in mode choice decisions taking account into the influences of different traffic states and transit congestion levels. To achieve this goal, travel behaviors to the central business district (CBD) of Nanjing City in China are investigated as a case study in this paper. First, a revealed preference (RP) survey on travelers' last trip to CBD is taken to obtain travelers' socioeconomic characteristics and their mode choice decisions. Then, traffic states and transit congestion levels during various time intervals are identified, and variations of variables are also investigated. Afterward, the mode choices of all the travelers under different traffic states are predicted based on the MNL models.

3.1. Data collection

The data collection aims to obtain travelers' socioeconomic characteristics and their travel mode decisions. Xinjiekou is the CBD region of Nanjing City, attracting large amounts of travelers in Nanjing. There are mainly three travel modes for the trip to CBD: car, bus and urban rail transit (metro). There is an urban freeway named Zhongshan road passing through the CBD region. Meanwhile, there is a metro line passing through this region. The urban freeway provides the on-ground travel by either bus or car, while metro provides the third travel mode.

An RP survey was taken to collect travelers' socioeconomic characteristics and their travel mode choices. Travelers' socioeconomic characteristics include gender, income and car ownership. There are also some alternative specific attributes that affect travelers' choice behavior, including walking time, waiting time, in-vehicle time, fare, comfort, etc. The survey was taken through the internet and interview on the travelers who had been to CBD at least once before. Finally 272 valid responses with 1632 data were collected. The information of 150 respondents (RP data1), which was the useful sample size suggested by Orme [15] in a previous study, were randomly sampled from the collected data to estimate model coefficients. The whole date (RP data2) was used for mode choice prediction and model evaluation.

The traffic volume, speed, and occupancy of the urban freeway were collected to identify traffic states. The congestion levels in metro carriages during different time intervals were also collected. The number of metro passengers was obtained from the collection system of metro to identify congestion levels.

3.2. Traffic state discrimination and transit congestion level identification

Traffic state identification by point parameters, such as volume, speed, and occupancy, has been discussed in many studies [16-18]. Fuzzy C-means (FCM) clustering, which is a clustering method based on optimum function, was proposed in traffic state identification [19, 20]. FCM clustering was conducted with the aid of matlab software. The main steps were shown as below [21].

Transit congestion levels were divided by the standing density in the metro carriage. The relationship between standing density and congestion perception is provided in Chinese urban rail transit engineering technical design specification [22].

When the standing density is low, the service is not sensitive to standing density changes because passengers always feel comfortable under this situation. However, when the standing density is high, passengers would feel uncomfortable if more people get into the carriage. Based on this concept, the relationship between the standing density and the service quality can be expressed using an exponential equation, as shown in equation 1 [23].

where, w, 0 are calibration coefficients; F0 is the passenger number per carriage when the grade is 0; qk is the number of passengers which can be calculated by the standing density. The passengers corresponding to different standing density can be calculated by Equation 2.

fk = w(1 - e F )

qk - F

qk ={L*(B - 2b)* y+S}*S

where, L is the length of carriages; B is the width of carriages; b is the width of seat; y is the standing density; S is

the number of seats; 8 is the convert coefficient.

The congestion level is identified by comparing the actual amount of passengers in carriages and the calculated amount of passengers by standing density. Congestion was divided into 5 levels according to standing density, as seen in Table 1.

Table 1. The evaluation criterion of congestion level in metro carriages Congestion level Passengers' perception Standing density (people/m2) Passengers in a carriage (people)

1 Comfortable <3 <140

2 Some comfortable 3-5 140-200

3 Mild crowded 5-6 200-220

4 Crowed 6-9 220-330

5 Very crowed >9 >330

3.3. Mode Choice decisions

A simple multinomial logit model is applied in this study for the mode choice analysis. An individual traveler is assumed to choose the mode if the corresponding utility is larger than all other modes. The utility of a mode is determined by travelers' socioeconomic characteristics and alternative specific attributes. In addition to deterministic terms, the utilities are also affected by random errors. Thus, a utility of a mode (for corresponding individual n) can be expressed as ^ (j = 1,L, J) , as shown in Equation 8.

where, V . is a function of the measured attributes which is also called Representative Utility; e . is the error.

' n nJ

The alternative specific attributes of the rail transit users includes waking time, waiting time, in-vehicle time, fare, and comfort; while the alternative specific attributes of the car users include travel time and cost. Normally, the utility function is linearly correlated with its variables. The utility function for traveler n, alternative mode j is expressed as Equation 9 [24, 25].

v, (X,) = aWn = b,xIn] + b2x2ni +l+bKxKRi = bxR] (4)

where, X =(X X L X ), are variables for alternative j, and b = (b b Lb )', are estimated coefficients.

. . ] 1 1] 2]' ' K] / J ] \ 1]' 2]' ' K] /

Coefficients of MNL models are estimated through the regression by Biogeme using RP datal. The estimated mode choice decisions can be obtained by Equation 5. In addition, models were evaluated by comparing the

predicted with actual mode choice decisions.

p. (5)

4. Results and discussion

4.1. Data Collection

272 respondents and 1632 data were collected from the survey. The respondents' socioeconomic characteristics were obtained. The percentages of the male and female travellers are 54.4% and 45.6%, respectively, indicating the gender distribution was overall balanced. In terms of the income, 17.6% of the respondents' annual income is higher than ¥100,000, which was regarded as the high income category in this study. Car ownership was an important factor determining whether travellers choose cars or not. The survey revealed that more than 40% of the

respondents' families have at least 1 car. The car alternative was considered unavailable to a person if no cars were available to his/her household.

In addition to travellers' socioeconomic characteristics, travellers' mode choice decisions were also obtained from the RP survey. It was found that traveller numbers of the bus, metro and car were 112, 100 and 60, respectively. In another word, the mode shares of the bus, metro, and car travel were 41.2%, 36.8%, and 22.0%, respectively, as shown in Table 2.

Table 2. Respondents' Last Trip to the CBD

NUMBER percent

Total surveys 272 %

Travel Mode Bus 112 41.2

Metro 100 36.8

Car 60 22.0

Travellers' alternative specific attributes, including walking time, waiting time, seat, and in-vehicle time were also obtained from the survey. Only 11.43% of the respondents claimed their walking time to a metro station is higher than 20 minutes, indicating that the acceptable walking time for most travellers to take metro was within 20 minutes. The walking time to a bus stations of all the respondents were less than 20 minutes, indicating a good accessibility of bus stations. Similar, most of the waiting time was within 10 minutes, indicating the acceptable waiting time for most travellers was about 10 minutes. It is noticed that the waiting time of metro travellers are lower than 5 minutes. This is because the intervals between two metros in Nanjing are 4 minutes. A reasonable in-vehicle time distribution was also observed with a mix of short and long time. However, average metro in-vehicle time (23 minutes) was much higher than bus in-vehicle time (16 minutes), indicating a longer trip by metro. The average trip length by metro, which was calculated by average speed (60km/h), was longer than Nanjing' average trip length [26]. About 50% of transit travellers have no seat.

The alternative specific attributes of car travellers include travel time, fuel cost and parking fee were compared with that of transit users. A wide range of travel time was observed varying from within 5 minutes to more than 30 minutes. It was noticed that more than 90% of the travel time was less than 30 minutes corresponding to the distance between origin and destination. About half of the fuel cost was between ¥5 and ¥10. This was also consistent with the relative travel distance. All of the parking fees were lower than ¥30. The possible reason of this is that the data was collected shortly after the increase of the parking price, while the respondents' last trip to the CBD area may was before that.

4.2. Traffic state and congestion level of transit identification

The fuzzy cluster analysis was conducted for the traffic state identification of Zhongshan road. Fig. 1 gave the results for traffic state identification. It clearly showed that traffic states of Zhongshan road were divided into three levels, green (state 1), red (state 2), and blue (state 3), and thereby provided support for the traffic state estimations within a day.

Transit congestion levels were determined by the comparison between the actual passengers in carriages and passengers calculated by standing density. It can be seen from Fig. 2 that transit congestion level was not consistent with the traffic state at the same moment.

The in-vehicle time and comfort of metro vary significantly under different traffic states and congestion levels. Fig. 3 shows the in-vehicle time of cars and buses under different traffic states. Comfort of metro was calculated by Equation 1 according to the amount of passengers per carriage. The coefficients can be fitted by grades on the service of metro from passengers. Fig. 4 provided the relationship between comfort and amount of passengers per carriage. The comfort was positive when the number of passengers per carriage was less than 140. The comfort was between -1 to 0 when the number of passengers per carriage was less than 220 but more than 140. In addition, the comfort drops slowly when passengers were between 140 and 220, while it decreased quickly when passengers were more than 220 per carriage.

State 2* "

...... |

• T'-J I . :

Fig. 1. The clustering results of Zhongshan road's traffic states

a) Traffic states b) Congestion levels

Fig. 2. The results of traffic states and congestion levels identification

Table 3 provided the range and mean value of comfort in different congestion levels. Based on the varying attributes above, the service of modes was got. Utilities were calculated using varying in-vehicle time and comfort with states, so the mode choice decisions under different traffic states and congestion levels were obtained „ • In terms of the values of the coefficient, it was as expected that all the time coefficients, including in-vehicle travel time (IVTT), waking time, and waiting time, were negative. It indicated that when the walking time, waiting time, or in-vehicle time for a particular mode is increased, the utility of that mode would be decreased, all other variables held constant.

4.3. Model Variables and Coefficients

150 respondents, which was the smallest sample size proposed by Orme [15], were randomly sampled from the collected data to obtain RP Data1. The coefficients in the utility equations were estimated based on RP Data1. The MNL regression was used to identify the influences of various factors on mode choice decisions. Table 4 listed the estimates and t-statistics. Estimation was performed by maximum likelihood method described by McFadden [27]. Based on estimation through the MNL regression, the following observations were obtained.

a) car b) bus

Fig. 3. The in-vehicle time with varying states

Passengers

Fig. 4. The fitting curve of Nanjing metro's comfort

Table 3. The comfort with different congestion levels

Congestion level Passengers' perception Passengers in a carriage (people) Comfort Range Mean

1 Comfortable <140 0-0.28 0.14

2 Some comfortable 140-200 (-0.6) -0 -0.3

3 Mild crowded 200-220 (-1) - (-0.6) -0.8

4 Crowed 220-330 (-10) - (-1) -5.5

5 Very crowed >330 <-10 --

• Since the coefficient of metro comfort was positive, individuals were found to show more preference to metro travel with the increase of metro comfort levels.

• Income variable revealed that travellers with annual income more than ¥ 100,000 showed more preference to car. This was consistent with the results in previous study [28]. It also showed that males have a positive effect on metro choice which was different from previous study by [29] that males tend to drive. The possible reason was that males in the study by Cutler [29] mainly were elder males, and rail transit was not considered in the study. Taking account into variables' significance, most of the variables' t-statistics absolute values were more than 1.96. Therefore, it can be concluded that most of variables, except car fuel cost and gender, used in the model had significant contribution toward predicting the mode choice decisions.

Table 4. The Model Estimation Results

VARIABLES UTILITY COEFFICIENT Std.Err (T-statistics)

IVTT_BUS BUS -0.369 1.69 (-2.02)

IVTT_METRO METRO -0.283 1.58 (-3.37)

IVTT_CAR CAR -0.012 3.01 (-2.07)

Walktime_BUS BUS -0.423 0.622 (-2.32)

Walktime_METRO METRO -0.446 0.768 (-2.23)

Waittime_BUS BUS -0.495 0.780 (-2.11)

Comfort_METRO METRO 0.101 0.37 (2.30)

Fuel cost CAR -0.035 1.04 (-0.34) *

Park fee CAR -0.327 1.04 (-3.31)

Genderl METRO 0.365 0.13 (0.12) *

Income4 CAR 0.0306 0.34 (2.80)

Carowner CAR 0.0815 0.095 (1.98)

Constant_BUS BUS -0.133 0.48 (-3.01)

Constant_METRO METRO 0.000 fixed

Constant_CAR CAR 0.0806 0.301 (14.8)

Number of observations 272 Log likelihood -298.412

Rho-square 0.490

Adjusted rho-square 0.395

*note: G1, G2, and G3: Group 1, Group 2, and Group 3; IVTT_A: in vehicle travel time of auto; IVTT_T: in vehicle travel time of transit

4.4. Mode Choice Decisions

The predicted share is defined as the share of sample which the model of Table 4 predicted. The predicted share was defined as:

S = N Z* (6)

where, pt was the probability of each sample who chose alternative i; N was the sample size.

The mode shares were as follows: car, 17.6%; bus, 35.3%; and metro, 47.1%, respectively based on the model of Table 4. An important model evaluation method was to compare the predicted choices with actual mode choices. It revealed that bus and car travel probabilities were underestimated, while metro travel was overestimated. Table 5 presented the actual and predicted mode choices, with predictions based on the model of Table 4. The actual mode choice for a particular alternative was the choice of people in the survey who actually chose the alternative. It can be observed that 88 out of the 112 individuals who were found to choose buses were correctly predicted. Therefor the accuracy of prediction for bus travels is 78.6%. Similar analysis on metro and car travellers showed that their prediction accuracies were 92.0%, and 73.3%, respectively. The model has an overall accuracy of 83.8%. A comparison of the actual and predicted mode choice decisions in Table 5 indicated that the MNL model in this paper (i) underpredicted the choice of cars; (ii) underpredicted use of bus; (iii) greatly overpredicted use of metro.

Several reasons may cause the mispredictions. The actual walk time to metro was perhaps considered more onerous than the time in memory. Another potential source of error was the rule which determined what modes were available to a person. Lastly, the sample size in this paper was not large enough although it reached the smallest sample size.

Table 5. Prediction accuracies

Observed Predicted

Bus Metro Car Percent correct

Bus (112) 88 20 4 78.6%

Metro (100) 8 96 0 92.0%

Car (60) 0 16 44 73.3%

Total 96 132 48 83.8%

The mode choice decisions above were made at a static state. However, variables in utilities varied with traffic states or transit congestion levels during different time intervals. So individuals changed their mode choices to maximize their utilities. A respondent was taken as an example to describe this. Fig. 5 showed that utilities of car, bus, and metro trips for the respondent during different time intervals within a day. It was found that the utilities of taking the three modes were largely dependent upon the time intervals. For example, the utilities of car, metro and bus trips were -2.1, -2.4, and -2.0 duding 6:00 to 7:30, respectively, while the utilities were -2.6, -2.4, and -2.7 respectively during 15:10 to 17:45. In other words, the bus trip had the highest utility during 6:00 to 7:30 while the metro trip had the highest utility during 15:00 to 17:45. It showed that the mode choice decisions of this individual also varied during different time intervals to maximize his (her) utility.

Fig. 5. The utilities of choosing different modes for individual

5. Summaries and conclusions

The paper investigated the travel mode choice decisions using a multinomial logit model considering the within-day dynamic traffic states and transit congestion levels. The data from the travel survey revealed the presence of within-day dynamics: the effect of varying explanatory variables (in-vehicle time and comfort) along with departure time on mode choices.

MNL analysis on the influence of mode choice decisions revealed that those who own cars prefer auto trips. The income influence also confirmed that individuals with high income prefer to drive. A comparison among different departure time with reference to their utilities of choosing modes revealed that traffic state and transit congestion level have a significant effect on mode choice decisions.

The main contribution of this study lies in that it examined the within-day dynamics in mode choice decisions. The results highlighted the influences of traffic states and transit congestion levels within a day on the variables of mode choice utilities. The MNL model utilized in this paper is helpful to policymakers to study travel behavior and improve transit service.

This paper also presented some possible reasons caused mode choice mispredictions. However, this study concerned only one model, one transportation environment, and one mode determining rule. Models need to be examined in a variety of transportation environments and rules to test whether the predictions in the model were good in future studies.

References

[1] Anas, A., 1983. Discrete choice theory, information theory and the multinomial logit and gravity models. Transportation Research Part B,

Methodological, 17(1), 13-23.

[2] Ben-Akiva, M. and Lerman, S. R., 1999. Discrete Choice Analysis, The Theory and Application to Travel Demand. Doctor, MIT.

[3] Walker, J. L., 2001. Extended Discrete Choice Models, Integrated Framework, Flexible Error Structure, and Latent Variables. Doctor, MIT.

[4] Train, K. and McFadden, D., 1976. An Application of Diagnostic Tests for the Independence from Irrelevant Alternatives. Transportation Research Record, Journal of the Transportation Research Board, 637, 39-46.

[5] Train, K. and McFadden, D., 1978. The Goods/leisure Tradeoff and Disaggregate Work Trip Mode Choice Models. Transportation Research, 12, 349-353.

[6] Dehghani, Y. and T. A, 1980. Model specification, modal aggregation, and market segmentation in mode-choice models, some empirical evidence. Transportation Research Record, Journal of the Transportation Research Board, 775, 28-34.

[7] Bordagaray, M., olio, L. d., Ibeas, A. and Cecin, P., 2014. Modelling user perception of bus transit quality considering user and service heterogeneity. Transportmetrica A, Transport Science, 10(8), 705-721.

[8] Ramadurai, G. and Srinivasan, K. K., 1977. Dynamics and Variability in Within-Day Mode Choice Decisions, Role of State Dependence, Habit Persistence, and Unobserved Heterogeneity. Transportation Research Record, Journal of the Transportation Research Board, 1977, 4352.

[9] Pas, E. I. and Koppelman, D. S., 1986. An examination of the determinants of day-to-day variability in individuals' urban travel behavior. Transportation, 13, 183-200.

[10] Burnett, K. P., 1977. Tests of a linear learing model of destination choice, application to shopping travel by heterogeneous population groups. Geographiska Annaler, Series B, 59, 95-108.

[11] Golledge, R., 1970. Some equilibrium model of consumer behavior. Economic Geography, 46, 417-424.

[12] Srinivasan, K. K. and Mahmassani, H. S., 2005. A Dynamic Kernel Logit Model for the Analysis of Longitudinal Discrete Choice Data, Properties and Computational Assessment. Transportation Science, 39(2), 160-181.

[13] Cascetta, E., Papola, A. and Pagliara, F., 2008. Analysis of mobility and location impacts of the HS-rail link Roma-Napoliusing within-day dynamic mode service choice. EuropeanTransport conference, 1-14.

[14]Wan, X., Wang, W. and Chen, J., 2012. Research on dynamic model of within-day travel mode choice of residents. Zhongguo Gonglu Xuebao (China Journal of Highway and Transport), 25(2), 121-126, 141.

[15] Orme, 2010. Sample Size Issues for Conjoint Analysis Studies. Sequim.

[16] Fengbo, Z. and Min, S. 2006. The identification method of traffic state based on fuzzy clustering. Microcomputer Applications, 22(2), 9-13.

[17] Pi, X. and Wang, Z., 2006. Application Research of Traffic State Classification Method Based on Collected Information from Loop Detector. Journal of Highway and Transportation Research and Development, 23(4), 115-119.

[18] Zhuang, b. and Yang, X., 2006. Criterion and Detection Algorithm for Road Traffic Congestion Incidents. Journal of Highway and Transport, 19(3), 82-86.

[19] Jiang, L., 2007. Study on the method of traffic state identification and congestion propagation estimation for urban road. Doctor, Jilin University.

[20] Huang, Y., Xu, L. and Luo, Q., 2013. The method of traffic state identification and travel time prediction on urban expressway. Research Journal of Applied Sciences, Engineering and Technology, 5(4), 1271-1277.

[21] An, L., Hu, Y. and Xie, W., 2003. A modified fuzzy C-means (MFCM) clustering algorithm. Journal of Hefei University of Technology, 26(3), 354-358.

[22] Construction, P. M. o., 2008. Urban rail transit project construction standards. Beijing.

[23] Dai, J., 2013. Research of distribution problem for urban metro traffic based on game theory and logit model. Master, Southeast University.

[24] Train, E. K., 2003. Discrete Choice Methods with Simulation[M]. Cambridge, Cambridge University Press.

[25] Cascetta, E., 2009. Transportation Systems Analysis, Models and Applications (Springer Optimization and Its Applications), Springer.

[26] Zhang, F., Liu, B., Bi, J. and Chen, J., 2012. Traffic Alternatives for Urban Residents and Affecting Factors. Sichuan Environment, 31(3), 132-137.

[27] McFadden, D., 1973. Conditional logit analysis of qualitative choice behavior. New York, Acidemic Press.

[28] Administration, F. T., 2014. Characteristics of Premium Transit Services that Affect Choice of Mode. WASHINGTON, D.C.

[29] Cutler, S. and Coward, R., 1992. A vailabily of personal transportation in households of elders, age, gender, and residence differences. The Gerontolist, 32, 77-81.