The Third International Conference on Traffic and Transportation Studies (ICTTS'2002), ASCE, pp 588-595, 23-25 July, 2002, Guilin, China.

A Combined RP/SP Nested Logit Model of Vehicle Ownership, Mode Choice and Trip Chaining in Developing Countries Dilum Dissanayake and Takayuki Morikawa2

Abstract Developing countries are recently suffering from severe transport problems due to intense vehicle ownership and usage. Therefore, investigation of travel behavior is found as an important issue. In this study, a Nested Logit (NL) model is developed considering vehicle ownership, mode choice and trip chaining aspects for household traveling. Developed NL model has two levels where the upper level consists of vehicle ownership choices and the lower level represents the mode choice combinations for two-traveler households. Model is analyzed using Revealed Preference (RP) and Stated Preference (SP) data since combined RP/SP is an effective method to express complex travel behavior and to forecast travel demand for new transport services. Proposed model is verified using the data from Bangkok Metropolitan Region (BMR). Introduction Increasing trend of vehicle ownership and usage in developing countries creates variety of transport problems such as traffic congestion and environmental pollution. Most of the researchers and practitioners in the transportation field are searching for reasonable answers to solve the problems. One of the reasons for the excessive vehicle usage can be due to inefficient services of public transportation. In addition, advantages of vehicle usage, for instance, comfort, convenience, safety and privacy, encourage travelers to own 

Research Fellow, Graduate School of Environmental Studies, Nagoya University, Chikusa-ku, Furo-cho, Nagoya 464-8603, Japan; phone 81-52-789-3730; [email protected]

2

Professor, Graduate School of Environmental Studies, Nagoya University, Chikusa-ku, Furo-cho, Nagoya 464-8603, Japan; phone 81-52-789-3564; [email protected]

1

The Third International Conference on Traffic and Transportation Studies (ICTTS'2002), ASCE, pp 588-595, 23-25 July, 2002, Guilin, China.

vehicles. Considering the situation in developing countries, travel decisions of household members are interrelated, and investigation of individual travel behavior is not a suitable method to understand the actual travel pattern. Therefore, emphasis is made on household travel behavior in this study investigating the household decisions of vehicle ownership, mode choice and trip chaining. The vehicle ownership has a long-term conceptual validity and therefore, it is a dominant factor in travel demand analysis. In particular, when a new transport alternative is in proposing stage, increasing trend of vehicle ownership has to be considered as an important element in the analysis. Among the methodological approaches for modeling travel demand, discrete choice models are with great popularity, especially to investigate complex travel behavior and to forecast the travel demand for new transport services. Considering the data requirement for travel demand investigations for existing modes, revealed preferences of transport users’ (RP data) are found as important. In addition, stated preferences (SP data) that are from hypothetical travel scenarios are required to forecast the travel demand for new modes since the actual travel data for future modes are not available. The proposed Nested Logit (NL) model uses RP and SP data that were obtained from Bangkok Metropolitan Region (BMR) and it mainly investigates the transport user behavior on proposed Mass Rapid Transit (MRT) system. Most of the fast growing cities in developing countries are facilitated with MRT systems and those are identified as the main support for the sustainable transport. BMR doesn’t have mass transit facilities and therefore, it always causes for severe traffic congestion and environmental pollution. Since Bangkok travelers do not have any experience of using such MRT systems, SP data may include different biases. In such a situation, RP/SP combined data analysis is advantageous to overcome the critical weaknesses of each data sources (Morikawa, 1989; Dissanayake and Morikawa, 2000; Dissanayake and Morikawa, 2001). The main features of the proposed model are:  Analyze travel demand for existing modes with explicit coverage of household decisions on vehicle ownership, mode choice and trip chaining.  Forecast travel demand for proposed MRT system in BMR. Study Area and Data Description The study area, BMR consists of Bangkok Metropolitan Area (BMA) and five adjacent provinces of Samut Prakan, Nonthaburi, Pathum Thani, Nakorn Pathom and Samut Sakorn. It was basically divided into 505 internal traffic zones covering 7758km2. According to the recent estimations, the population in BMR is 13 million in year 2001. RP Data. The RP data that are used in this study were obtained from a household travel survey in BMR during 1995/96. The survey was considered as a part of a major transport project in the BMR named UTDM (Urban Transport Database and Model Development) project. The RP survey collected the data for household traveling and the database consists of all attributes of the trips that were made on the date of the survey and the information of household members.

2

The Third International Conference on Traffic and Transportation Studies (ICTTS'2002), ASCE, pp 588-595, 23-25 July, 2002, Guilin, China.

SP Data. The research group of Infrastructure and Transportation Planning laboratory at Nagoya University conducted a SP survey in 1996 to obtain information on user preference regarding the future MRT project in Bangkok. The SP questionnaire has been prepared to achieve the requirements relating to the commuter travel and the data was collected by either direct interviews or mailed questionnaires from randomly selected individuals. More specifically, the transport users were asked to select the choices that correspond to their personal constraints, among the hypothetically created travel scenarios. In the SP questionnaire, attributes that relate to the SP choices have been explicitly incorporated and those are in the forms of travel time, travel cost, travel speed, reliability (minimum delays), safety, comfort, service frequency, accessibility, intra modal transfers and access/egress time. The Proposed Model To improve the reliability of parameter estimates, the model is developed as combined RP/SP. At first, RP and SP models are developed separately. Then, the two models are combined, and in the combined model, the portions of SP model estimate using SP data and the portions of RP model estimate using RP data while sharing the coefficients that are common for both RP and SP data. RP Model. The NL structure presented in Figure 1 consists of two levels where the upper level shows Car-owning/Motorcycle-owing/No Vehicle-owning choices and the lower level represents the corresponding mode choice combinations for twotraveler households. The RP model estimates with 1205 household trips. One of the travelers in the household makes a commuting trip and trip purpose of the second traveler in the same household can be any type such as work, school, shopping or recreation. Commuter (main traveler) trip can be either home-to-work or work-tohome. There are 17 mode choice combinations that represent the household travel pattern. In Figure 1, C, MC, CH, B, H and T indicate car, motorcycle, chain, bus, hired motorcycle and taxi respectively. Furthermore, alternatives 1~7 are the mode choices for car owning households either using car (alt. 1~4) or traveling by other modes (alt. 5~7) that are available in the system. More specifically, alternatives 1~4 are the variety of car using patterns, where the commuter travels by a car and the second traveler of the same household selects one from the available options of car chain (CH), bus (B), hired motorcycle (H) or taxi (T). In alternatives 5~7, both travelers who belong to car owning group travel by B, H or T. Similarly, alternatives 8~14 and 15~17 are for MC-owning and No Vehicle-owing households respectively. SP Model. The SP model basically analyzes individual travel behavior, especially for commuters, and it is developed as a Multinomial Logit Model (MNL) considering three SP travel alternatives of MRT, bus and car. The attributes such as travel time, travel cost, car ownership and the first factor (the most important factor for the SP choice) are appropriately included in the SP model. The SP database consists of 1240 individual commuter trips.

3

4

C B

2

C CH

1

3

C H

Commuter mode Second traveler mode

4

C T 6

H H 7

T T 8

MC CH 9

MC B 10

MC H

MC-owning

11

MC T

12

B B

13

H H

14

T T

15

B B

No Vehicle-owning

1 Figure 1. Nested logit (NL) model for two-traveler households

5

B B

Car-owning

Mode choice representation:

16

H H

17

T T

The Third International Conference on Traffic and Transportation Studies (ICTTS'2002), ASCE, pp 588-595, 23-25 July, 2002, Guilin, China.

The Third International Conference on Traffic and Transportation Studies (ICTTS'2002), ASCE, pp 588-595, 23-25 July, 2002, Guilin, China.

Combined RP/SP NL Model. The combined model shares some of the coefficients, especially for common attributes belonging to RP and SP models. In particular, following changes have been considered in this combined model: 

Level of service variables such as travel time and travel cost share for all RP and SP based utility functions in the combined model.



Mode specific dummies for all modes are specified separately for the commuter and for the second traveler in the RP utility functions, and those commuters’ mode specific dummies are appropriately made to share in SP utility functions.



A scale parameter is included in the SP utility functions to observe the relative level of randomness in the RP and SP data sources.

Simultaneous estimation (full information maximum likelihood) method is used to estimate the combined RP/SP NL model. Basically, it is assumed that the scale parameter for the bottom level (level of mode choices) is equal to one and then, the scale parameter for the upper level is estimated. Attributes, which are obtained from the RP and SP databases, are explicitly incorporated for the analysis. Results and Discussion Table 1 illustrates the parameter estimation results for the combined RP/SP NL model. Most of the parameters are statistically significant with expected signs by explaining the travel behavior for existing and new transport modes in developing countries. Mode specific dummies are introduced separately for each traveler to observe their individual intention for travel alternatives in the system. By considering the results for mode specific dummies, it is clear that the usage of car as well as motorcycle is attractive for both travelers since those dummies are positively significant. By observing the commuter’s mode choice preferences, they have an intention to use the proposed MRT. Although commuters’ preferences for bus and hired motorcycle are negative, second travelers have great intention to use bus as well as hired motorcycle options. The alternative specific constants for MC-owning and No Vehicle-owning are significantly positive. Coefficients for travel time and travel cost/income are significantly negative as expected. Initially, scale parameter for the bottom level (mode choice level) is assumed to be 1. According to the estimation results, scale parameter for the upper level (carowning/mc-owning/no vehicle-owning) is 0.72, and it is in the specified limits between 0 and 1. The scale parameter (RP:SP) significantly estimates as 0.62, implying that the SP data contain more random noise than RP data. A variety of alternative specific dummies are included in the developed NL model to observe the household travel behavior in developing countries. Dummy variable for male commuters, which is included in the car chain (alt. 1) and the MC chain (alt. 8) alternatives, significantly yields with positive sign expressing their contribution for household travel responsibilities by making trip chains.

5

The Third International Conference on Traffic and Transportation Studies (ICTTS'2002), ASCE, pp 588-595, 23-25 July, 2002, Guilin, China.

Table 1. Estimation results for the SP/RP combined NL model Variable description

parameters t-statistics

Mode specific dummies Commuter/ 2nd traveler: Car (RP/SP) Commuter/ 2nd traveler: Motorcycle (RP) Commuter: Bus (RP) Commuter: Hired motorcycle (RP) Commuter: MRT (SP) 2nd traveler: Bus (RP) 2nd traveler: Hired motorcycle (RP)

1.38 1.42 -0.54 -1.17 3.42 2.85 1.80

4.7 4.6 -1.7 -3.1 5.9 9.9 6.0

Alternative specific constants MC-owing : RP No Vehicle-owning : RP

0.84 1.94

2.1 4.0

Level-of-service variables travel time (hrs): RP/SP travel cost/income/102: RP/SP

-0.57 -2.90

-4.6 -6.8

0.72 0.62

5.6 6.5

1.67 1.99 1.12 0.73 -2.06 0.58 -1.07 1.60 0.61 0.72 1.15 1.99 -1.11 0.92 -1.05 -1.82 -0.93 2.23 -0.78 0.42

6.8 3.6 5.4 2.2 -2.8 2.6 -4.2 4.2 2.6 2.0 4.3 3.8 -1.6 4.7 -4.7 -3.4 -3.4 6.0 -2.9 2.0

Scale parameters  car owning / mc owning / no vehicle owning

scale (RP:SP) Alternative specific dummies male commuter, alt. 1, 8: RP travel distance for both travelers > 30km, alt. 2: RP distance between destinations  10km, alt. 9: RP distance between destinations  15km, alt. 1, 8: RP second travelers travel distance >5km, alt. 3: RP distance share of both travelers > 75%, alt. 8: RP commuter’s job (executive), alt. 8: RP commuter’s job (executive or business),Car-owning: RP travelers jobs are not executive, No Vehicle-owning: RP commuter’s age >50 yrs, No Vehicle-owning: RP school children in the household  1, Car-owning: RP household income  25000, No Vehicle-owning: RP household travel purpose, alt. 8: RP trips touching CBD, alt. 2: RP trips within CBD, alt. 8: RP trips within CBD, alt. 16: RP trips within CBD, Car-owning: RP present mode, Bus, Car: SP car ownership, MRT: SP first factor: travel time and service reliability, MRT: SP Number of observations L( βˆ ) L( c )

2445 -2990.8 -4640.3 0.36 62

2

VOT (Baht/hr)

6

The Third International Conference on Traffic and Transportation Studies (ICTTS'2002), ASCE, pp 588-595, 23-25 July, 2002, Guilin, China.

When the travel distances for both travelers in the household are more than 30 km, related dummy variable is with positive significance that indicates the household tendency to use car & bus (alt. 2) rather than making trip chains for longer trip lengths. When the distance between destinations is greater than or equal to 10 km, the corresponding dummy variable estimates with positive significance indicating the household attraction on motorcycle & bus (alt. 9), where the commuter uses motor cycle allowing second traveler to use bus. When the distance between destinations is less than or equal to 15km, related dummy variable is positive significant highlighting the household tendency to make car chains (alt. 1) or motorcycle chains (alt. 8). If the second travelers travel distance is more than 5 km, car & hired motorcycle (alt. 3) is not a household selection indicating that hired motorcycle is not suitable for distance traveling. When the travel distance of both travelers shares more than 75%, mc chain is found as a suitable option. Commuter’s job condition is also analyzed by introducing dummy variables into the alternatives of motorcycle chain (alt. 8), car-owning alternatives. For the executive job oriented commuters have negative intention to form motorcycle chains. When the commuter’s job is executive or business oriented, the related dummy variable in car-owning alternative is significantly positive indicating their preference to own a car. Similarly, when the travelers’ jobs are not in executive groups, they do not prefer to own vehicles. Also, aged commuters (>50yrs) prefer not to own vehicles. Households with school children have positive intention to own a car. When the household income belongs to the lower income group, no vehicle-owning is found to be attractive for them. Traveling in CBD is tested with several dummies. For the trips that touch the CBD zone, car & bus (alt. 2) is found as an attractive mode selection for developing country households. In other words, traveling through CBD zone, especially in BMR, is extremely difficult during peak congestion hours, and therefore, the commuter drives alone by allowing the second traveler to use a bus rather than attempting to form a trip chain in such an area. If household trips are in CBD zone, mc chain (alt. 8), hmc & hmc (alt. 16) and car-owning alternatives are not suitable choices for them. RP mode is also tested in the SP utility functions of car and bus. Those dummies are significantly positive indicating the users’ hidden preference to continue their RP mode. Car ownership is tested as a dummy in MRT utility function and the related parameter is negatively significant indicating their preference of using cars. First factor, in other words, the most dominant factor for the mode choice, is tested as a dummy in MRT utility function. For the travelers, whose dominant factors for the mode choice are either travel time or service reliability (no delays), have an intention to select MRT as their future mode since the estimated parameter is significantly positive. Using those log likelihood values, goodness of fit of the estimated model is obtained as a very high value of 0.36. VOT is calculated using the estimated coefficients of the travel time and the travel cost/income, and it is obtained as 62 Baht/hr.

7

The Third International Conference on Traffic and Transportation Studies (ICTTS'2002), ASCE, pp 588-595, 23-25 July, 2002, Guilin, China.

Concluding Remarks A RP/SP combined NL model is developed in this study to investigate the household travel behavior in developing countries with explicit consideration of recent behavioral variations relating to car ownership, mode choice and trip chaining. The proposed model analyses the travel behavior for existing modes and also forecasts the user intention on proposed MRT system in BMR. The findings of this study provides further evidence that transport users’ stated preferences can be profoundly used as an accurate guide to represent the actual underlining preferences. It has been clearly observed that the combined estimation of RP and SP data in travel demand modeling is an effective technique for expressing complex travel behavior and forecasting the travel demand for new transport services. Male commuters attract highly to form trip chains indicating their contribution for household responsibilities. CBD travel, long distance travel and distance between the destinations are also found to be important for households to get their decisions on mode selection. The results obtained from this study is realistic and it can be effectively used for decision-making activities related to the transportation sector in developing countries. Further investigations on the proposed model will be focused on policy related analysis. References Dissanayake, D., and Morikawa, T. (2001). “A Combined RP/SP Nested Logit Model to Investigate Household Decisions on Vehicle Usage, Mode Choice and Trip Chaining, J. of Eastern Asia Society for Transportation Studies, Vol. 4(2), 235-244. Dissanayake, D. (2001).“ Urban Transport Policy Analysis for Developing Countries Considering Household Choices of Modes, Trip-chaining and Vehicle Ownership”, Doctoral Dissertation, Department of Civil Engineering, Nagoya University, Japan. Dissanayake, D., and Morikawa, T. (2000). “Travel Demand Models with the RP/SP Combining Technique for the Developing Countries”, The International Conference CODATU IX, Mexico, 103-107. Morikawa, T. (1989).“Incorporating Stated Preference Data in Travel Demand Analysis”, Doctoral Dissertation, Department of Civil Engineering, MIT. Urban Transport Database and Model Development Project (UTDM). (1998) Office of the Commission for the Management of Land Traffic (OCMLT), Final Report Vol. 1 & 2.

8