Methodology Summary of Zhang et al.

(2013)

1. Data Collection
They used both RP (Revealed Preference) and SP (Stated Preference) surveys in Jinan,
China:

• RP survey:

• Collected actual travel behavior: weekly trips, payment mode, bus travel time, bus
satisfaction, vehicle ownership, car purchase plan, etc.

• SP survey:

• Designed hypothetical scenarios by varying bus ticket price, bus travel time, parking
fee, and fuel cost.
• Ticket prices: different levels for air-conditioned/non-air-conditioned buses (e.g., 1
yuan, 1.5 yuan, 3 yuan, etc.)
• Travel time: 2 levels
• Parking/fuel cost: 2 levels (unchanged, increased)
• They used an orthogonal design to generate efficient combinations.

Survey was household-based, conducted in June 2012, with 1223 valid questionnaires.

2. Model Framework
They built Multinomial Logit (MNL) models:

• First with only SP data.

• Then with combined SP + RP data.

Problem: You can’t directly merge RP and SP, because their random error variances differ
(SP data usually noisier).

Solution: They introduced a balance coefficient (λ) to correct this.

Formula: Var(εRP) = λ² Var(εSP)

They used a phase estimation method (two-step):

1. 1. Estimate SP model → get preliminary parameters.

2. 2. Estimate RP model with substitution.
3. 3. Revise SP data using λ, then combine with RP and re-estimate jointly.

3. Variables in the Utility Functions

Different alternatives (Car, Bus, Motorcycle, Others) had different systematic utility
components:
• Car utility: parking fee, fuel cost, weekly trips, vehicle ownership, car purchase plan,
driving experience, driver’s license.
• Bus utility: bus fare, travel time, income, occupation, age, gender, payment mode (e.g.,
reimbursement), bus satisfaction, factors affecting travel.
• Motorcycle/Other utility: ownership and usage variables.

4. Use of Dummy Variables

They created categorical dummies for socio-economic and attitudinal variables to fit into
the MNL model:

• Monthly income (4 levels → 3 dummies; reference = ≥8000 yuan).

• Occupation (4 groups → 3 dummies; reference = soldier/others).
• Factors affecting travel choice (ticket price = 1, reference = time/comfort/safety).
• Car purchase plan (within 1–2 years, 3 years, ≥4 years/no plan → 2 dummies, last is
reference).

Example:

• Monthly income ≤999 yuan → (1,0,0)

• 1000–3999 → (0,1,0)
• 4000–7999 → (0,0,1)
• ≥8000 (reference) → (0,0,0)

This dummy coding allowed categorical socio-economic factors to be estimated in the MNL
alongside continuous factors (e.g., cost, time).

5. Estimation Results (Key Findings)

The balance coefficient λ = 0.135 (<1) → SP data had more noise than RP data.

Signs of coefficients were logical:

• Higher bus fare (−3.40) and travel time (−0.019) decreased bus choice probability.
• Higher bus satisfaction (+0.6) increased bus choice probability.
• Owning a driver’s license (−0.703) increased car choice likelihood.
• Income, occupation, and reimbursement policy significantly affected bus choice.

Model fit:

• SP-only: ρ² = 0.438
• Combined RP+SP: ρ² = 0.654 → much better fit.
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/262488284

A Traffic Mode Choice Model for the Bus User Groups Based on SP and RP Data

Conference Paper · January 2013

CITATION READS

1 77

4 authors, including:

Yunqiang Xue
Beijing University of Technology
19 PUBLICATIONS 149 CITATIONS

SEE PROFILE

All content following this page was uploaded by Yunqiang Xue on 19 May 2016.

The user has requested enhancement of the downloaded file.

 Available online at www.sciencedirect.com

ScienceDirect
Procedia - Social and Behavioral Sciences 96 (2013) 382 – 389

13th COTA International Conference of Transportation Professionals (CICTP 2013)

A Traffic Mode Choice Model for the Bus User Groups Based on
SP and RP Data
Zhihu Zhanga,*, Hongzhi Guana, Huanmei Qina, and Yunqiang Xueb
a
Beijing Key Lab of Traffic Engineering, Beijing University of Technology, Beijing, 100124, China
b
Jinan Urban Transport Research Center, Jinan, 250031, China

Abstract

Enhancing the bus share rate is a major measure to relieve the traffic congestion. To analyze the effect of
public transit policy, this paper establishes MNL models based on both SP data and combining SP and RP data,
which was collected in Jinan city. Then the paper analyzes how the influencing factors affect the choice
proportion of bus travel mode for the bus user groups. In the end, the paper obtains some significant conclusions
and proposes measures which would enhance the bus attraction.
©
© 2013 TheAuthors.
2013 The Authors. Published
Published by Elsevier
by Elsevier B.V.access under CC BY-NC-ND license.
Ltd. Open
Selection and/or
Selection and peer-review
peer-review under responsibility
under responsibility of ChineseofOverseas
ChineseTransportation
Overseas Transportation Association (COTA).
Association (COTA).

Keywords:SP and RP Survey; MNL model; Combining data; Sensitivity analysis;

1.Introduction

With the rapid development of economy, the total number of motor vehicles ownership in large and medium-
sized cities in China grows rapidly. Traffic demand grows continuously, while the supply of urban land resources
strains increasingly. The contradiction between traffic supply and demand is obvious. It is unable to meet the
growing travel demand by relying solely on expanding and increasing the road construction. Bus priority policy
can improve the utilization rate of road resources, which is the effective way of solving road congestion problems
in city. It is the focus of study for many scholars to increase the public transit share rate.
Several studies have been conducted to increase the public transit share rate. Li studied the bus priority policy
affected the development of urban traffic. Wang studied the method of making the subway ticket price. Litman
studied how to predict the travel impacts of specific price reforms and management strategies. Paulley et al

* Corresponding author. Tel.:+86-10-67391870

E-mail address: q322101@126.com

1877-0428 © 2013 The Authors. Published by Elsevier Ltd. Open access under CC BY-NC-ND license.
Selection and peer-review under responsibility of Chinese Overseas Transportation Association (COTA).
doi:10.1016/j.sbspro.2013.08.045
 Zhihu Zhang et al. / Procedia - Social and Behavioral Sciences 96 (2013) 382 – 389 383

studied bus travel demand was affected by the fare, service level, individual income and the number of car
ownership.
It is essential to discuss the bus travel demand under different service level and technology before studying the
bus share rate. RP data(revealed preference data, RP data for short) can t describe the nonexistent traffic mode, on
the contrary, SP data(stated preference data, SP data for short) can design future traffic scene and analysis the
traffic demand under different conditions. However, revealed preference choice may be in contradiction with
stated preference choice, in others ways, the SP data has the biases. Many scholars made studies to solve problem
of SP data biases. Guan et al established the combining model by combining the traffic experiment data and SP
data and solved the SP data deviation. Ben.A et al combining RP data and SP data which is revised by RP data
established model to solve the SP data biases.
This paper uses above method and establishes MNL models using the SP and RP data of the bus user groups in
Jinan city. Then the paper makes the sensitivity analysis of some main factors in order to analyze their effects on
enhancing the bus attraction. In the end, some significant recommendations are concluded.

2. Travel Behavior Survey of the Bus User Groups

The method of Revealed Preference (RP) survey and Stated Preference (SP) survey was used to analyze the
user behavior in this paper. The survey items include three parts.
Personal information including gender, age, occupation, car purchase plan, and monthly household
income
Bus travel behavior including weekly trip times of used traffic mode, payment mode, bus travel time and
bus satisfaction degree
Stated Preference survey
Bus ticket price, bus travel time, parking fee and fuel cost are four important influencing factors and used to
design the questionnaire survey.
Air-conditioned bus and non air-conditioned bus price are set three levels. Bus travel time is set two levels.
Parking fee and fuel cost are set two levels. Orthogonal design method is used to obtain the most suitable factor
combination as shown in table 1.
Table 1. Factors Combination of SP

Air-conditioned and non air- Parking fee and fuel Parking fee and fuel
Travel time
conditioned bus ticket price cost unchanged cost increased
1yuan /0.5yuan unchanged  
1yuan /0.5yuan decreased by 20 %   Available traffic modes
including
1.5yuan /0.8yuan unchanged  
Car, bus, motorcycle, bike,
5yuan/0.8yuan decreased by 20 %   walking others
3yuan /2yuan unchanged  
3yuan /2yuan decreased by 20 %  

Household survey method was used in this paper. The interviewees of the bus user groups choose one travel
mode under different travel conditions. The survey was conducted from June 16 to June 24 in 2012. 1359
questionnaires are retrieved and the effective sample is 1223.
384 Zhihu Zhang et al. / Procedia - Social and Behavioral Sciences 96 (2013) 382 – 389

3. Model and Combining SP Data and RP Data

3.1.MNL Model
MNL(Multinominal Logit)Model is the basic type of logit models. The random utility in is mutually
independent and obeys the same Gumble extreme value distribution. Based on probability theory, MNL model
with J options can be expressed in the following formula.

exp( Vin )
pin
J i 1, 2, ,J (1)
exp( V jn )
j 1

Where, pin is probability of any alternative i being selected by person n from choice set J , is unknown
coefficient and Vin is called the systematic components of the utility of alternative i .

3.2.Combining SP data and RP data

SP data and RP data can t be combined simply because their random parts are different. Therefore, the balance
coefficient is introduced to estimate the parameters of random parts. The equation is shown below.
RP 2 SP
Var ( ) Var ( ) (2)
RP SP
Where, and are respectively the random parts of RP data utility function and SP data utility function.
It can establish the disaggregate model with combining the SP data which is revised by balance coefficient and
RP data. Simultaneous estimation method and phase estimation method are two ways to estimate parameters. This
paper uses phase estimation method.
The following equations are respectively utility function of SP data model and RP data model.
u inRP RP
in w inRP RP
in (3)
SP SP SP SP
u in in z in in (4)

Where, uin is the utility function of alternative i being selected by person n, in is the common variable of
RP data utility function and SP data utility function, win and zin are respectively variables of RP data utility
function and SP data utility function, , , are the unknown parameters.
The specific steps of parameters estimation are shown below.
The first step is to obtain the value of parameter and parameter by model with SP data and replace
V RP with X RP .
RP
The second step is to suppose the utility function of RP data model is u in VinRP w inRP RP
in and
obtain the value of parameter , parameter and parameter which is equal to 1 / .
The third step is to obtain the revised SP data using SP and zSP multiplying by . Then combining revised
SP data and RP data establish model and regain the value of parameter , parameter and parameter .
 Zhihu Zhang et al. / Procedia - Social and Behavioral Sciences 96 (2013) 382 – 389 385

4.MNL Model Estimation and Analysis

In order to avoid model misconvergence and estimated errors caused by the empty cells, the monthly income,
occupation, factors affecting travel and car purchase plan are reclassified and combined based on correlation
analysis. Table 2 shows the classification setting of these factors.
Bicycle travel mode and walking travel mode are merged into other travel modes because the total choice
proportion of bicycle travel mode, walking travel mode and other travel modes are less than 4%. Different traffic
modes utility functions include different variables of affecting factors. Parking fee, fuel cost, weekly trip times,
the number of car ownership, car purchase plan, driving experience and driver s license constitute the systematic
components of the utilities for car travel mode. Bus ticket price, travel time, monthly income, occupation, age,
gender, payment mode, factors affecting travel and bus satisfaction degree constitute the systematic components
of the utilities for bus travel mode. Weekly trip times and the number of motorcycle ownership constitute the
systematic components of the utilities for motorcycle travel mode. Weekly trip times and the number of other
vehicles ownership constitute the systematic components of the utilities for other traffic travel modes.
Table 2. The Classification Setting

Variable Classification Dummy variable

Monthly income 1 Not more than 999 yuan; 1 0 0
Monthly income 2 Between 1000 yuan and 3999 yuan; 0 1 0
Monthly income
Monthly income 3 Between 4000 yuan and 7999 yuan; 0 0 1
Monthly income 4 reference type Not less than 8000yuan; 0 0 0
Public utilities, non-public utilities ,education
Occupation 1 ,health protection ,scientific research, individual 1 0 0
household and freelancer;
Occupation 2 Primary school, middle school or college student; 0 1 0
Occupation
Organization unit, non- organization unit,
Occupation 3 agriculture, forestry, animal husbandry, jobless or 0 0 1
retirees;
Occupation 4 reference type Soldier and others; 0 0 0
Factors affecting travel 1 Ticket price; 1 0 0
Factors affecting Factors affecting travel 2 reference Travel time, travel comfort ,safety or
travel 0 0 0
type convenience;
car purchase plan 1 Having car, one year or two year; 1 0 0

car purchase plan 2 Three year; 0 1 0

Car purchase plan
car purchase plan 3 reference type
Not less than four year or No consideration 0 0 0

The result of model calibration is shown in table 3.Asymptotic rho squared 2 and adjusted rho squared 2 are
2
two important indexes of the model evaluation. Generally, the model precision is high when and 2 are both
between 0.2 and 0.4.
The coefficient of parking fee and fuel cost is -0.866, which indicates the choice proportion of car travel mode
for the bus user groups will decrease when the parking fee and fuel cost is increasing. The coefficient of bus
travel time is 0.004 and is not practical, which may derive from the irrationality of bus user groups. The
coefficient of bus ticket price is -0.46, which indicates the choice proportion of bus travel mode for the bus user
groups will decrease when the bus ticket price is increasing. The coefficient of bus satisfaction degree is 0.466,
which indicates the choice proportion of bus travel mode will increase when the bus service level is improved.
For SP and RP model, 2 is 0.654 and 2 is 0.653. This indicates the precision of SP and RP model is better
than that of SP model. The balance coefficient is 0.135 and the t test is 20.45, which indicates the establishment
of SP and SP model is significant. The balance coefficient is less than 1, which indicates the random noise
386 Zhihu Zhang et al. / Procedia - Social and Behavioral Sciences 96 (2013) 382 – 389

interference of SP data is higher than that of RP data. The variable coefficient signs of SP and RP model are
coincident with the fact. The coefficient of travel time is -0.019 and the coefficient of bus ticket price is -
3.401,which indicates the choice proportion of bus travel mode for the bus user groups will decrease when the bus
travel time and bus ticket price are increasing. The coefficient of weekly trip times is 0.003, which indicates the
choice proportion of car travel mode for the bus user groups will increase when the weekly trip times of car
increase. The coefficient of payment mode is 0.225, which indicates the choice proportion of bus travel mode for
the bus user groups will increase when the reimbursement of bus travel fare is increasing. The coefficient of
driver s license is -0.703, which indicates that the choice proportion of car travel mode for the bus user groups
will increase if the bus user has had the driver s license.
Table 3. The Result of Model Calibration

SP model SP and RP model

Variable name Parameter value t test Parameter value t test
Constant dummy 1 3.189 17.332 1.916 20.128
Constant dummy 2 2.501 11.384 3.12 49.879
Constant dummy 3 0.654 13.741 0.365 8.729
Parking fee and fuel cost -0.866 -15.294 -6.005 -14.977
Weekly trip times 0.03 10.909 0.003 0.405
The number of vehicles ownership 0.172 8.733 0.297 7.908
Car purchase plan 1 0.159 2.657 0.417 3.153
Car purchase plan 2 0.477 4.353 -0.512 -1.364
Driving experience -0.031 -1.563 -0.028 -0.915
Driver s license -0.645 -8.489 -0.703 -8.498
Bus travel time 0.004 6.061 -0.019 -17.648
Bus ticket price -0.46 -20.86 -3.401 -21.29
Bus satisfaction degree 0.446 12.744 0.6 9.619
Factors affecting travel 1 -0.154 -2.797 -0.157 -1.247
Monthly income 1 -0.16 -1.174 -0.856 -3.079
Monthly income 2 -0.41 -4.746 -0.927 -5.071
Monthly income 3 -0.525 -6.206 -0.425 -2.326
Occupation 1 0.371 5.647 0.148 1.074
Occupation 2 0.839 7.785 0.539 2.51
Occupation 3 0.287 3.711 0.049 0.271
Age 0.02 1.303 0.321 8.833
Gender -0.027 -0.689 -0.012 -0.136
Payment mode   0.225 3.424
Balance Coefficient   0.135 -20.45
L(0) -21443.201 -42886.402
-12051.573 -14847.901
2( L(0) L( )) 18783.256 56077.002
2
0.438 0.654
2
0.437 0.653
 Zhihu Zhang et al. / Procedia - Social and Behavioral Sciences 96 (2013) 382 – 389 387

5. Model Sensibility Analysis

Parking fee, fuel cost, bus ticket price and bus travel time are four important influencing factors which are
choosen to make sensibility analysis. The paper respectively discusses the sensibility analysis when the parking
fee and fuel cost are unchanged or increased.

5.1.Parking fee and fuel cost unchanged

The choice proportion of bus travel mode for the bus users groups is shown in figure 1a when the parking fee
and fuel cost are unchanged. The bus travel time varies between 0 minutes and 100 minutes and the bus ticket
price varies between 0 yuan and 5 yuan.

Figure 1a The choice proportion of bus travel mode  Figure 1b the shadow of figure 1a

It can draw some conclusions from the above figure 1a. When the parking fee and fuel cost are unchanged, the
choice proportion for the bus user groups who continue choosing bus travel mode gradually decreases with the
increasing bus travel time and ticket price. The choice proportion for the bus user groups who continue choosing
bus travel mode varies little when the bus travel time ranges between 60 minutes and 100 minutes or between 0
minutes and 20 minutes. When the bus travel time ranges between 20 minutes and 60 minutes, the choice
proportion who continue choosing bus travel mode exhibits considerable variation.
Figure 1b shows the choice proportion for the bus groups who continue choosing bus travel mode on two-
dimensional plane of figure 1a under different bus ticket price and bus travel time.
Figure 1b indicates that the bus user groups are more sensitive to the change of bus travel time than that of bus
ticket price. When the air-conditioned bus ticket price varies between 0yuan and 5yuan and bus travel time varies
between 0 minutes and 20 minutes or between 40 minutes and 100 minutes, the choice proportion for the bus
groups who continue choosing bus travel mode varies little. When the air-conditioned bus ticket price varies
between 0 yuan and 5 yuan and bus travel time varies between 20 minutes and 40 minutes or between 40 minutes
and 100 minutes the choice proportion for the bus groups who continue choosing bus travel mode exhibits
considerable variation, which is regarded as the sensitive area.

5.2.Parking fee and fuel cost increased

The choice proportion of bus travel mode for the bus users groups is shown in figure 2a when the parking fee
and fuel cost are increased. The bus travel time varies between 0 minutes and 100 minutes and the air conditioned
bus ticket price varies between 0 yuan and 5 yuan. The figure 2b is the shadow of figure 2a.
388 Zhihu Zhang et al. / Procedia - Social and Behavioral Sciences 96 (2013) 382 – 389

Figure 2a The choice proportion of bus travel mode  Figure 2b The shadow of figure 2a

The changing trend of choice proportion of bus travel mode for the bus user groups with bus travel time and
bus ticket price changing is similar to that in figures 1a and 1b when the parking fee and fuel cost are unchanged.
Therefore, the parking fee and fuel cost has little effect on the choice of bus travel mode for the bus user
groups.

6.Conclusions

At first, the paper makes a survey of the bus user groups in Jinan city and establishes the MNL model as well
as SP and RP model. Next, the paper made model sensibility analysis and evaluation. In the end, it discusses the
relationship between the choice proportions for the bus user groups who continue choosing bus travel mode and
parking fee, fuel cost, bus ticket price and bus travel time.
The conclusions are as follows.
The higher the public transit groups are satisfied with bus service, the easier the users choose the bus
mode as the traffic mode.
The choice proportion of the bus user groups who continue to choose bus mode will decrease when the
bus travel time and bus ticket price are increasing.
The bus user groups are more sensitive to the change of bus travel time than that of bus ticket price.
The parking fee and fuel cost has little effect on the choice proportion of bus travel mode for the bus user
groups.
The choice proportion of the bus user groups who continue choosing bus travel mode exhibits
considerable variation when the air-conditioned bus ticket price varies between 0 yuan and 5 yuan and
the bus travel time varies between 20 minutes and 40 minutes.
Therefore, there are two aspects to enhance the bus attraction and increase the choice proportion of the bus
user groups who continue to choose bus travel mode. One is offering higher bus service level and the other is
decreasing the bus travel time by designing the bus lane, bus signal and bus entrance lane, and so on.
This paper merely analyzes the travel mode choice for the bus user groups. However, the car user groups,
motorcycle user groups, bicycle user groups and other user groups also may choose bus travel mode as the traffic
mode. Some of them were the potential bus user and not included in this paper. It deserves analyzing for further
research.
 Zhihu Zhang et al. / Procedia - Social and Behavioral Sciences 96 (2013) 382 – 389 389

Acknowledgements

This research was supported by National Basic Research Program of China (No. 2012CB725403). The authors
are very grateful for the comments from the anonymous reviewers.

References

Ben-Akiva, M. and Morikawa, T. (1990) Estimation of switching models from revealed references and stated intentions. Transportation
Research A 24(6), 485 495.
BEN A, MORIKAWA T. Estimation of travel demand models from multiple data sources. Transportation and Traffic theory,1990, 36(6),461-
476.
Cummings, R.G, Brookshire, D.S. and Schulze, W.D. (eds) (1986) Valuing Environmental Goods: An Assessment of the Contingent
Valuation Method. Totowa, NJ: Rowman and Allanheld.
Dissanayake,D.,Morikawa,T. A Combined RP/SP Nested Logit Model to investigate Household Decisions on Vehicle Usage, Mode Choice
and Trip Chaining. Journal of Eastern Asia Society for Transportation Studies, 2001, 4(2), 235-244.
Elisabetta Cherchi, Juan de Dios Ortúzar. On Fitting Mode Specific Constants in the Presence of New Options in RP/SP Models.
Transportation Research Part A: Policy and Practice. 2006, 40(1):1~18.
Gardner, Transit, 2005. 31(1),24-28.
Hongzhi Guan, Kazuon, Takeshi M. A model of car and P&BR choice in commuting trips combining the experiment day data with the stated
preference data. infrastructure Planning Review,1999,I6:955-961.(in Japanese)
Hongzhi Guan, Shanchuan Wang, Li-ya Yao, Yi hong. Urban Railway Choice Behavior Model Based on RP Data and SP Data. the academic
journal of Beijing university of technology,2007.33(2),203~207.
Huanmei Qin, hongzhi Guan, Huanhuan Yin. A Study of the effect of parking price on the mode of inhabitant trip behavior with the cars,
public transit and taxi in Beijing as an example. China Civil Engineering Journal, 2008, 41(8), 93-98.
Hongzhi Guan. Disaggregate model-A tool of traffic behavior analysis. beijing:china communications press,2004.
John C. Whitehead, Subhrendu K. Pattanayak, George L. Van Houtven, Brett R. Gelso. Combing Revealed and Stated Preference Data to
Estimate The Nonmarket Value of Ecological Services: an Assessment of The State of The Science. Journal of Economic Surveys (2008) Vol.
22, No. 5, pp. 872 908.
Hensher, D.A. and Bradley, M. (1993) Using stated choice data to enrich revealed preference discrete choice models. Marketing Letters 4(2),
139 152.
Liya yao, hongzhi Guan, Yan hai. The Effects of Fare on Traffic Structure and Mode Split Model. academic journal of Beijing university of
journey,2007,33(8),834-837.
Litman, T. Transportation Elasticities: How prices and other factors affect travel behavior. Victoria Transport Policy Institute. Victoria,
British Columbia, 2005.
Morikawa, T. Correcting Sate Dependence and Serial Correlation in RP/SP. Combined Estimated Method. Transportation. 1994, 21(2),
153~166.
Morikawa, T., Ben-Akiva, M. and Yamada, K. (1991) Forecasting intercity rail ridershipusing revealed preference and stated preference data.
Transportation Research Record.
Maddala, G.S. (1983) Limited-Dependent and Qualitative Variables in Econometrics,Econometric Society Monographs No. 3. Cambridge:
Cambridge University Press.
Neil Paulley, Richard Balcombe, Roger Mackett, Helena Titheridge, John Preston, Mark Wardman, Jeremy Shires, Peter White. The demand
for public transport: The effects of fares, quality of service, income and car ownership. Transport Policy, 2006(13), 295 306.
Ricardo García, Angel Marín. Network Equilibrium with Combined Modes: Models and Solution Algorithms. Transportation Research Part
B: Methodological. 005, 39(3), 223~254.
Swait, J. and Louviere, J. (1993) The role of the scale parameter in the estimation and comparison of multinomial logit models. Journal of
Marketing Research 30, 305 314.

View publication stats