Estimating Modal Shift of Car Travelers to Bus on Introduction of Bus Priority System

JOURNAL OF TRANSPORTATION SYSTEMS ENGINEERING AND INFORMATION TECHNOLOGY Volume 9, Issue 6, December 2009 Online English edition of the Chinese language journal Cite this article as: J Transpn Sys Eng & IT, 2009, 9(6), 120í129.

RESEARCH PAPER

Estimating Modal Shift of Car Travelers to Bus on Introduction of Bus Priority System VEDAGIRI P.1,*, ARASAN V. T2 1 Department of Civil Engineering, Indian Institute of Technology Bombay, Powai, Mumbai 400076, India 2 Transportation Engineering Division, Department of Civil Engineering, Indian Institute of Technology Madras, Chennai 600036, India

Abstract:

This study is concerned with estimation of the probable shift of car users to bus due to increase in its level of service

after providing exclusive bus lanes on Indian city roads carrying heterogeneous traffic. The quantum of increase in level of service of bus due to introduction of exclusive bus lane was determined using a recently developed simulation model of heterogeneous traffic flow. The data of the other factors (variables) that might cause modal shift from car to bus were collected through home-interview survey based on the stated preference approach. A binary Logit model of mode-choice was then calibrated using the collected data and the model was also validated using holdout sample. Through this study, a set of causal factors, with reliable and predictable data base, to explain the variation in shift behaviour of personal vehicle users to buses consequent on the increase in the level of service of buses, has been identified. A mode-choice probability curve to depict the possible modal shift of car users to bus is developed, taking the difference in travel times of the two-modes as the basis, to serve as a user friendly tool to analyze the possible modal shift for a wide range of values of the involved variable. Key Words: public transportation; heterogeneous traffic; travel time; exclusive bus lane; mode choice models

1

Introduction

Road-based passenger mobility in India has increased tremendously over the years. The rapid increase in motorized mobility during the last two decades or so is primarily due to increase in urban population as a result of both internal growth and migration from rural areas and small towns. The high rate of growth of Indian economy has also contributed significantly to the increase in demand. Bus is the main urban transit system used in most Indian cities and gradually, its level of service is declining due to inadequate capacity and managerial and financial problems. In the absence of an adequate and efficient bus transit system, the potential bus users currently use private transport modes—mainly motorized two—wheelers and, to some extent, cars. Also, some of them resort to use of another para transit mode called auto-rickshaws (it is a motorized version of the traditional rickshaw, a small three-wheeled cart driven by a person). Thus, a large number of private and para-transit vehicles have entered into the market to meet the travel demand. As the available road space is limited, the proliferation of these

vehicles results in severe congestion, inordinate delay, high-energy consumption (particularly of fossil fuels), and intense pollution of the environment. The traffic on the roads of Indian cities is highly heterogeneous comprising vehicles of wide ranging static and dynamic characteristics. The vehicles occupy any lateral position on the road, depending on the availability of road space, at a given instant of time without any lane discipline and it is nearly impossible to impose lane discipline under such conditions. Under the said heterogeneous traffic flow conditions, the buses, being relatively larger vehicles, find it difficult to maneuver through the mixed traffic and are subjected to frequent acceleration and deceleration leading to lower speed and discomfort to both the driver and passengers. This also results in enormous delay and uncertainty to bus passengers and consequently, the level of service of buses gets reduced considerably making the bus, a less attractive mode of transport. Indian cities desperately need improved and expanded public transport service and not personal vehicles. This requires both an increase in quantity as well as quality of

Received date: Jul 6, 2009; Revised date: Oct 19, 2009; Accepted date: Oct 28, 2009 *Corresponding author. E-mail: [email protected] Copyright © 2009, China Association for Science and Technology. Electronic version published by Elsevier Limited. All rights reserved. DOI: 10.1016/S1570-6672(08)60092-6

VEDAGIRI P. et al. / J Transpn Sys Eng & IT, 2009, 9(6), 120í129

bus transport service and effective application of demand as well as supply-side management measures. This goal can be attained by encouraging bus transport by assigning priority to it. One of the common bus preferential treatments is provision of reserved bus lanes on major urban roads to facilitate faster movement of buses, which will make the mode more attractive. Provision of exclusive road space, thus, will enhance the level of service of buses and this may also result in shift of some of the personal vehicle users to buses. The motorized personal vehicles available for travel, in Indian cities are, Car, Auto-rickshaw—three-wheeled motorized transit vehicle and motorized two-wheelers. As the possible shift of travelers from these personal modes to bus is independent of one another, the probability of shift has been studied through three separate binary choice models. This paper is concerned with the study of the possible shift of car users to bus because of increased level of service of bus, after provision of exclusive bus lanes.

2

Review of earlier studies

A modal shift occurs when one mode gains a comparative advantage in a travel market over another. The comparative advantage can take various forms, such as costs, capacity, time, flexibility, or reliability. Depending on the kind of passengers traveling and their circumstances (socio-economic characteristics, purpose of trip, etc.), the relative importance of each of these factors vary. Mode choice models effectively could be divided into two classes: (i) models that cover all available modes and (ii) models focusing on selected modes only[1]. The former type of model is necessary, where the demand forecast is required for all modes, as would be the case if a broader social cost-benefit appraisal of a transport scheme is conducted. If one is solely interested only in two modes, a binary choice model[2–5] is more appropriate. For example, to study how many travelers may be attracted from car to bus, only car and bus need to be considered. As this study pertains to the analysis of modal shift from car to bus, the review of literature presented here is confined to the research works related to the choice behavior of travelers, under conditions wherein one of the involved modes is bus and the analysis are related to modal shift. Tischer and Dobson[6] studied the factors, which influence the intentions of the single-occupant commuters to switch to buses and carpools and suggested operating policies consistent with the intent to encourage the use of high-occupancy vehicles. They found that in buses, convenience is the most important variable associated with the shift intention. They also found that perceptions of carpool comfort do not appear to be important, rather, perceptions of carpool schedule flexibility, cost, safety, and a short wait in traffic were found to be the prime factors associated with potential shift to carpool.

Alvinsyah et al.[7] developed a binomial Logit model based on stated preference (SP) data to study the response of the travelers in using the proposed Jakarta bus-way system. Travel time and travel cost were considered as the main variables to develop utility functions. On the basis of these modal characteristics and the different service strategies offered, peoples’ perception and their probability of selecting the proposed system are predicted. The results show a wide range of peoples’ perception and their probability of choosing the better service. Nurdden[8] identified the factors that prevent personal transport users from utilizing public transport, so that rational polices could be formulated to encourage greater utilization of public transport. Binary Logit models were developed involving car and bus and car and train. The most important variables, found likely to encourage the use of public transport, were as follows: reduced travel time, walking distance to public transport stations, and subsidized fare. Gebeyehu and Takano[9] studied the citizens’ perceptions of the bus condition, as a determining factor for their choice of bus transportation, and developed a binary Logit model to analyze traveler choice behavior. The result of this study shows that citizens’ perceptions of the three chosen bus-transit condition aspects (fare, convenience, and frequency) have a significant influence on public-transport-mode choice. Fillone et al.[10] developed nested Logit models, which divided the available modes into private and public, and one of the public transport modes, namely, bus was further sub divided into air-conditioned and non air-conditioned buses. One interesting output of this study is with regard to the utility ranking of modes, arrived at based on the derived utility equations of each mode. With this utility-ranking result, the extent of actual share of bus in the travel market was quantified. The results show that, there is a need to redefine the role of buses in Metro Manila. Hensher and Reyes[11] studied the reasons for individuals’ choice to undertake trip chains and the extent to which trip complexity is a barrier to the propensity to use bus transport. Discrete choice models are developed to identify the role that socioeconomic and demographic characteristics of households have on the propensity to the use of car and public transport. This study suggests that, as the number of vehicles per household increases, the relative utility yielded from bus transport use, for work, decreases. Mackett[12] identified different policy actions to reduce car use for different types of trips and the actions that are required to meet the travel needs that the car currently fulfils. Then, the evidence on why people used their cars for a set of real short trips is considered in terms of a number of dimensions including age, sex, and trip purpose. This is followed by a discussion of the alternative modes to the car that drivers say that they might adopt and the factors, which they say, would

VEDAGIRI P. et al. / J Transpn Sys Eng & IT, 2009, 9(6), 120í129

Fig. 1 Modeling approach for modal shift analysis

make them consider switching to these alternatives. The analysis of results from the surveys shows that, “improving public transport” is the specific action, which drivers say, is most likely to attract them out of their cars. All these studies, in summary, are motivating, and shed some light on the factors that influence the way individuals decide to choose their modes of travel and the contrasting roles of perceptions and satisfactions in their ability to respond to switching intentions to buses. At present, there is no research material available in behavioral study of switching intention of personal vehicle users to buses, under Indian traffic conditions, and this study has been done by taking the methodologies and results of the reported earlier studies as general guidelines.

3

Methodology

As per the available literature, two different approaches are used in mode choice analysis: (1) revealed preference (RP) approach and (2) stated preference (SP) approach. The RP approach has been used to model mode-choice when data on actual choice of mode by travelers are available. Whereas, the SP approach has been used to analyze the response of people to hypothetical choice situations, these, of course, can cover a wider range of attributes and conditions than the real system. In this study, SP approach has been adopted for the model development. Because the modes considered were only bus and car, a binary choice model has been used. As the interpretation and specification is straight forward in the Logit model than the Probit model, the Logit model was adopted for the study. The different elements related to the general methodology of approach to the modeling process are depicted in the form of a flow chart in Fig. 1. The shaded portions of the flow-chart show the sequence of conceptual steps related to the model development in this study.

4

Model specification

The model specification here is based on utility theory, which is based on the assumption that individuals select that mode which maximizes their utility (U). Utility theory enables

prediction of changes in choices that occur when an attribute of one of the alternatives changes. Moreover, the utility-based model is able to capture differences in the responses of different individuals to the same attribute change. The utility of an alternative ‘i’ is assumed to be made up of two terms: a deterministic term (V) representing systematic and observed effects, and a random term (H) representing unobserved factors affecting the choice. The random-error term, H, is assumed to be independently and identically distributed as per the Gumbel distribution. As per this specification, an individual is likely to shift from car to bus if the utility of bus mode is more than the utility of car. Without loss of generality, the utility of shift is given as the difference between utilities of bus and car. Therefore, a traveler is likely to shift from car to bus if the utility difference, Udiff.•0. The utility of shifting is also assumed to be made up of two terms: a deterministic term (Vdiff=VBus–VCar) representing systematic and observed effects and a random term (Hdiff=HBus–HCar) representing unobserved factors affecting choice. On the basis of the aforementioned assumptions, it can be shown that Hdiff. is distributed as per the logistic distribution. Therefore, the probability of shift can be obtained: Pshift Pr (U diff t 0) Pr (Vdiff  H diff t 0) (1) V e diff t0 1  e Vdiff

The deterministic term Vdiff is assumed to be given by a linear-in-parameters specification. Accordingly, Vdiff=A0+A1X1+···+AnXn (2) Therefore, the probability of shift can be given as, eVdif e A0  A1 X 1  A2 X 2 " An X n (3) ȇshift Vdiff . 1 e 1  e A0  A1 X1  A2 X 2 " An X n where Pshift is the probability of shift from car to bus mode; Vshift is the deterministic utility function of difference in utilities of bus and car; A0, A1, A2··· are the model parameters to be estimated; X1, X2··· are variables influencing modal shift.

5

Data base for SP survey

As the objective here is to predict the aggregate shift of car users to bus, consequent on reduction in travel time of buses due to provision of exclusive bus lanes, it is necessary to estimate the possible reduction in travel time of buses due to provision of exclusive bus lanes under the prevailing roadway and traffic conditions. For this purpose, the Chennai city, in the south eastern part of India, was considered as an example. The major roads in Chennai city, which carry a significant amount of bus traffic, are either six-lane divided or eight-lane divided roads. Thus, the width of road space available for one-way movement on these roads, is equivalent to either three or four lanes, which are sufficient to allocate one lane exclusively for buses. To conduct the SP survey, it is necessary to know the speeds of buses and cars, on these two types of roads for a wide range of traffic-volume conditions,

VEDAGIRI P. et al. / J Transpn Sys Eng & IT, 2009, 9(6), 120í129

M.T.W, 55%

M.Th.W, 12%

Bicycle, 6%

Bus, 5%

Car, 18%

LCV, 3% Truck, 1%

LCV—light commercial vehicles; M.Th.W—Motorized Three-Wheelers; M.T.W—Motorized Two-Wheelers Fig. 2 Representative traffic composition R2=0.9992

R2=0.9972

Fig. 3 Journey speed of buses and cars

and this can be achieved through appropriate traffic simulation experiments. As the available simulation models are based on fairly homogeneous traffic conditions, where, strict lane discipline exists, these models are not suitable for simulating Indian traffic conditions. Hence, a recently developed and validated model of the highly heterogeneous traffic flow prevailing on Indian roads[13] was used for this study. The simulation modeling framework is explained briefly here to provide the background for the study. As per the simulation framework, the entire road space is considered as - single unit, and the vehicles are represented as rectangular blocks on the road space, the length and breadth of the blocks representing respectively, the overall length and the overall breadth of the vehicles. The front left corner of the rectangular block is taken as the reference point, and the position of vehicles on the road space is identified based on the coordinates of the reference point with respect to an origin chosen at a convenient location on the space. The simulation model uses the interval scanning technique with fixed increment of time. For the purpose of this study, a traffic composition representing the mean composition of traffic on the major roads of Chennai city- was considered (Fig. 2). The roadway width for the simulation of traffic flow in one direction was fixed as 11.0 m (three lanes) and 14.5 m (four lanes),

respectively. The traffic flow on the assumed arterials was simulated for road conditions with and without bus lane and for a wide range of traffic volume (from near free flow condition to capacity level). In actual practice, the stoppage of buses at bus stops increases the journey time for buses. Hence, it is better to account for the stoppage of buses while justifying provision of exclusive bus lanes. To incorporate the speed reduction suffered by buses due to stops on exclusive bus lanes, a survey was conducted on typical urban arterial roads in Chennai City to measure the dwell time of buses. The average dwell time for buses, obtained through the survey, is 17 s, and the average distance between bus stops is 1.02 km. Then, using the basic equations of motion, the time and distance required for acceleration and deceleration of buses at a bus stop is calculated. The journey speed of buses and cars, for the roadway and traffic conditions considered, is depicted in Fig. 3. As per the recommendations of Indian Roads Congress[14], the desirable level of service for urban roads is ‘C’ and the corresponding traffic volume level is equal to 0.7 times the capacity. Hence, it would be appropriate to consider this value (volume-to-capacity ratio=0.7) as the base for determining the speed difference between bus and car. Accordingly, it can be found that the difference in journey speed (alternatively the difference in in-vehicle travel time), between buses (on exclusive lane) and cars, expressed as the percentage of bus speed, falls in the range of about 16% to 48 %, when both the roadway conditions (three and four lane) are considered together. This result was used as the base to prepare the questionnaire for the SP survey.

6

Study area and survey design

To be able to use a model in a practical situation, it is necessary to estimate the model parameters using survey data. To study the effect of reduction in travel time on the demand for bus travel, a SP questionnaire was prepared. The data of the factors (variables) that might cause modal shift from car to bus were collected through home-interview survey conducted in a residential area, named, Todhunter Nagar, in the southern part of Chennai city, India, which has reasonable accessibility to bus service (walking time to bus stop varies from 3–15 min). The home-interview survey was carried out in households owning cars. During the survey, the respondents were asked to base their response on their previous day trips. The questionnaire had provision to collect data on the following attributes: (a) gender, (b) age, (c) walking time to bus stop, (d) trip purpose, and (e) willingness or otherwise to shift to bus for various in-vehicle travel time differences (bus travel time being 0, 10%, 20%, 30%, and 40% less than the travel time by car). As the travelers state their preference after perceiving the cost implications of the alternative modes, the cost of travel was not included as the variable in mode-choice model.

VEDAGIRI P. et al. / J Transpn Sys Eng & IT, 2009, 9(6), 120í129

Table 1 Description of the variables considered for modeling Variable name Gender Age 1 Age 2 Age 3 Age 4 Walking time 1

Walking time 2

Walking time 3

Trip-W Trip-E Trip-O SP (dependent variable)

Description and coding details Male/female. The variable will be assigned value 0, if male and 1, if female People in the age group, 10-20 years. The variable will be assigned the value 1, if the respondent falls in the age group and 0, otherwise People in the age group, 21-40 years. The variable will be assigned the value 1, if the respondent falls in the age group and 0, otherwise People in the age group, 41-60 years. The variable will be assigned the value 1, if the respondent falls in the age group and 0, otherwise People in the age group greater than 60 years. The variable will be assigned the value 1, if the respondent falls in the age group and 0, otherwise Walking time to bus stop is ” 5 minutes. The variable will be assigned the value 1, if the walking time of the respondent falls in this range and 0, otherwise Walking time to bus stop is 6-10 minutes. The variable will be assigned the value 1, if the walking time of the respondent falls in this range and 0, otherwise. Walking time to bus stop is greater than 10 minutes. The variable will be assigned the value 1, if the walking time of the respondent falls in this range and 0, otherwise Trip made for work. The variable will be assigned the value 1, if the trip under consideration is made for work and 0, otherwise Trip made for education. The variable will be assigned the value 1, if the trip under consideration is made for education and 0, otherwise Trip made for other purposes. The variable will be assigned the value 1, if the trip under consideration is made for other purpose and 0, otherwise Will be assigned the value 1, if the trip maker prefers to shift to bus and 0, otherwise

The list of variables considered for modeling and their descriptions are given in Table 1.

7

The home-interview survey was conducted with 100 car users. The data set pertaining to the 100 car users, with their responses for shifting to bus (for five travel-time-difference scenarios), was processed into 500 (5×100) data points for modeling. For the purpose of model calibration, a set of 400 data points (80% of the total) was used, while setting aside the rest of the observations (20%) for the purpose of validation. For the model-calibration analysis, a software tool, named, Statistical Software Tools (SST) was used. The goodness-of-fit for the calibrated model can be assessed by likelihood ratio index (U2), which is given as, (4) LLȇ  LL0 2 U

Table 2 Results of model calibration

Model development

7.1 Model calibration Model calibration or estimation involves finding the values of the parameters, which make the observed data more likely under the model specification; in this case, one or more parameters can be judged non-significant and left out of the model. The estimation also considers the possibility of examining empirically certain specification issues, for example, structural and/or functional form of parameters may be estimated. In this study, the SP (willingness or otherwise to shift) of the respondent is the dependent variable and gender, age, walking time to bus stop, trip purpose, and in-vehicle travel time difference are the independent variables considered for model estimation. As the dependent variable is discrete in nature, the model was calibrated by maximum-likelihood estimation using Newton Raphson method[15]. For a fixed set of data and underlying probability model, the maximum-likelihood picks the values of model parameters that make the data “more likely” than any other values of the parameters would make them.

LL0

where LL(P) is the log-likelihood of the estimated model and LL(0) is the log-likelihood when the coefficients are assumed to be 0. The model calibration results are shown in Table 2. It can be seen that the signs of the parameters of the variables are logical. The value of the t-statistic for different variables, when compared with the corresponding table value, indicate that all the parameter estimates are significant at 1% level. 7.2 Model validation For the purpose of model validation, the holdout data set, with 100 data points, was used as follows: first, a separate model of modal shift, using the data of the hold out sample, was calibrated and the Log-likelihood (LL) was estimated. Next, the model initially calibrated using the 400 data points, was applied to the hold out sample with 100 data points, to predict the modal shift, and the value of LL was calculated. Then, the two values of LL were compared for their closeness. The relevant details are given in Table 3.

Variable Constant Age 3

Parameter estimate

t-statistic

–2.28 0.86

–6.75 3.28

Trip-W

0.63

2.35

Walking time -3

–1.07

–2.48

Percentage time difference

6.34

6.15

Likelihood ratio index (ȡ2) = 0.25 Table value of t, 1% level of significance=2.33.

Table 3 Results of model validation Value of model statistics Description

Model initially Model calibrated using calibrated with 400 data hold out sample with points 100 data points

Initial LL

–221.81

–55.45

Final LL

–165.59

–41.32

0.25

0.25

-

–41.32

-

–44.06

ȡ2 Estimated LL for the model calibrated using hold out sample Calculated LL by applying model initially calibrated based on 400 data points, in hold out sample

VEDAGIRI P. et al. / J Transpn Sys Eng & IT, 2009, 9(6), 120í129

Table 4 Calibration and validation results of the models based on trip purpose Values of parameter and t-statistic Influencing variable

Model based on trips for work

Model based on trips for other purposes

Parameter estimate

t-statistic

Parameter estimate

t-statistic

Constant

–0.93

–2.58

–1.95

–3.72

Age 3

1.08

2.76

1.67

3.78

Walking time 3

–1.04

-2.42

–1.10

–2.32

6.01

4.83

5.91

3.62

Time difference* 2

Likelihood ratio index (ȡ )

0.27

0.25

Model validation ȡ2 for model calibrated using hold out sample Estimated LL for the model calibrated using hold out sample Calculated LL by applying the model initially calibrated, on the hold out sample

0.28

0.24

–23.34

–20.54

–24.42

–20.81

*Percentage by which Bus journey time is less than car journey time.

Table 5 Probability of shift of car users Percentage difference between bus and car journey time 0

Probability of shift (considering trips for all purposes) 0.13

10

0.22

20

Probability of shift (based on trip-purpose) Work

Others

0.29

0.20

0.42

0.31

0.35

0.57

0.45

30

0.51

0.71

0.59

40

0.66

0.81

0.72

Fig. 4 Probability of shift of car users to bus

It can be seen from Table 3 that the two LL values are close to each other, thus proving the validity of the model. The acceptable ȡ2 value ranges from 0.2 to 0.4[7] and ȡ2 values of around 0.4 may give excellent fits[16]. Hence, the validation result may be considered to be satisfactory.

8

Sub-models based on trip purpose

Two different sub-models for trips made for work, and all other purposes, were calibrated and validated by following the same procedure explained in the previous section, to study the shift behavior of car users. The number of trips made for education, using car, were found to be too small to build a model. The model-calibration and validation results are given in Table 4. It can be seen that the variables used in the models

are statistically significant and the two models are also valid. The probability of shift (for various travel time differences), from car to bus, estimated using the models, are presented in Table 5. It can be seen that, as expected, the higher the time difference (bus travel time less than car travel time), higher is the probability of shift in all the cases. It can also be seen that for a given time difference, the probability of shift is maximum in the case of trips made for work purpose.

9

Sub-models considering only travel-time difference as the influencing variable

For policy decisions on urban-transport-system management, it would be appropriate to consider the impact of changes in the operating characteristics of travel mode(s). Accordingly, in this case, it would be appropriate to estimate the probability of shift to bus, considering only travel-time as the influencing variable. Accordingly, the models were calibrated and validated by considering travel time difference as the only basis for modal shift and following the same procedure explained in the previous section. Then, the calibrated models were used to predict the aggregate shift of car users to bus, the only criterion considered being reduction in travel time, and the results are given in Table 6. The probability of shift for different travel-time differences between car and bus, obtained using the calibrated models, are presented in Table 7.

10

Modal shift probability curve

To illustrate the usefulness of the modal shift modeling exercise in urban road transport management, the developed aggregate modal shift model (involving all the trips made for the different purposes), considering travel-time difference as the influencing variable, was used to develop a modal shift probability curve (Fig. 4). In which, the solid line represents 3-line, and the dotted line represents 4-lane. As per the Indian Roads Congress (IRC) guidelines (IRC, 106-1990), the

VEDAGIRI P. et al. / J Transpn Sys Eng & IT, 2009, 9(6), 120í129

Table 6 Calibration and validation results of the models with travel-time difference as the only causal variable Values of parameter and t-statistic Model based on trips for other Model based on trips for work purposes Parameter Parameter t-statistic t-statistic estimate estimate -0.88 –2.95 –2.06 –4.81

Model based on trips for all purposes

Influencing variable

Parameter estimate –1.90

Constant Time difference*

5.72

Likelihood Ratio Index (ȡ2 )

0.22

t-statistic –7.59 6.00

5.62

4.39

5.99

0.24

0.25

0.24

0.26

-24.83

–20.05

-23.48

–23.44

3.94

Model Validation ȡ2 for data set pertaining to holdout sample 0.23 Estimated LL for the model calibrated using –42.539 hold out sample Calculated LL by applying model, initially –44.60 calibrated, on the hold out sample *Percentage by which bus journey time is less than car journey time.

Table 7 Probability of shift of car users to bus estimated by taking travel-time difference as the only causal variable Percentage difference between bus and car journey

Shift probability (considering trips for all purposes)

Work

0

0.13

0.28

0.11

10

0.20

0.41

0.18

20

0.31

0.55

0.29

30

0.45

0.68

0.42

40

0.59

0.79

0.57

Probability of shift (based on trip purpose) Others

recommended level of service for urban roads is ‘C’ and the volume of traffic corresponding to this level of service can be taken as 0.7 times the capacity. Hence, the percentage travel-time difference between bus (on exclusive lane) and car at volume corresponding to V/C ratio value of 0.7 was determined using the simulation model. The percentage travel-time differences were 16 and 48, respectively, for the roads with 14.5 and 11.0 m widths for movement of traffic in one direction. The probability of shifts then can be obtained from the curve, as 0.28 and 0.70 for 14.5 m and 11.0 m wide road spaces.

11

Conclusions

The following are the important findings of the study: (1) Through this study, a set of causal factors, with reliable and predictable data base, that explain the variation in shift behavior of car users to buses has been identified. The identified factors are: gender, age, walking time to bus stop, trip purpose, and travel time difference. (2) The calibrated general Logit model of modal shift (involving all trips and all variables) is found to be statistically significant with a satisfactory rho-square value. The model, when validated using hold-out sample, was found to be valid based on the comparison of the predicted LL value against the originally estimated LL value. (3) The different models developed to facilitate understanding of the shift-behaviour changes with respect to

trip purpose, indicate that the probability of shift is maximum in the case of trips made for work purpose. (4) The modal-shift model, developed considering the travel-time difference alone as the causal variable, is also found to be statistically valid indicating the relatively high significance of the variable in explaining the modal shift. (5) The modal-shift-probability curve, drawn based on the modal shift model (involving all the trips made for different purposes), considering travel-time difference as the influencing variable, can serve as a user-friendly simple tool to estimate modal shift probabilities. For example, it can be inferred from the curve, that on city roads, at traffic flow corresponding to level of service C, (V/C=0.7), the probability of shift of car users to bus, after the implementation of exclusive bus-lane scheme, is 0.70 for 11 m wide road and 0.28 for 14.5 m wide road, respectively.

References [1]

Gebeyehu M, Takano S. Modelling the relationship between seasonal constraints and modal choice characteristics of high school students in snowy region. Journal of the Eastern Asia Society for Transportation Studies, 2005, 6: 844–1857.

[2]

Rao S B S. Sequential binary mode choice models for urban travel to work. Highway Research Bulletin, Indian Road Congress, Highway Research Board, 1993, 48: 91–107.

[3]

Arasan V T, Rengaraju V R, Rao K V K. Trip characteristics of travelers without vehicles. Journal of Transportation Engineering, 1996, 122(1): 76–81.

[4]

Ghareib A H. Evaluation of Logit and Probit models in mode-choice situation. Journal of Transportation Engineering, 1996, 127(4): 282–290.

[5]

Hamed M, Easa S M. Integrated modeling of urban shopping activities. Journal of Urban Planning and Development, 1998, 124(3): 115–131.

[6]

Tischer M L, Dobson R. An empirical analysis of behavioral intentions of single-occupant auto drivers to shift to high occupancy vehicles. Transportation Research A, 1979, 13:

VEDAGIRI P. et al. / J Transpn Sys Eng & IT, 2009, 9(6), 120í129

[11]

43–158. [7]

propensity to use public transport. Transportation, 2000, 27(4):

attitude based on choice model parameter characteristics (case

341–361.

study: Jakarta bus way system). Journal of Eastern Asia

[12]

Mackett R L. Why do people use their cars for short trips?

[13]

Arasan V T, Koshy R Z. Methodology for modeling highly

Transportation, 2003, 30(3): 329–349.

Society for Transportation Studies, 2005, 6: 480–491. [8]

Nurdden A, Rahmat R A, Ismail A. Effect of transportation

heterogeneous

polices on modal shift from private car to public transport in Gebeyehu M, Takano S. Diagnostic evaluation of public

[10]

of

Eastern

Transportation Studies, 2007, 6: 480–491.

Asia

Society

for

of

Transportation

[15]

Ypma T J. Historical development of the Newton-Raphson

Roads in Plain Areas, IRC: 106, New Delhi, India, 1990. method. Society for Industrial and Applied Mathematics

Fillone A M, Montalbo C M, Tiglao N C. Transport mode Journal

Journal

Indian Road Congress (IRC). Guidelines for Capacity of Urban

(SIAM) Review, 1995, 37(4): 531–551.

choice models for metro manila and urban transport policy applications.

flow.

[14]

transportation mode choice in Addis Ababa. Journal of Public Transportation, 2007, 10(4): 27–50.

traffic

Engineering, 2005, 131(7): 544–551.

Malaysia. Journal of Applied Sciences, 2007, 7(7): 1013–1018. [9]

Hensher D A, Reyes A J. Trip chaining as a barrier to the

Alvinsyah, Soehodho S, Nainggolan P J. Public transport user

[16]

Ortuzar J D, Willumsen L G. Modelling Transport. 3rd Edn, John Wiley& Sons Ltd., England, 2001.