Development of a mode choice model considering heterogeneous traffic of Dhaka city

Abstract

Mode choice analysis is one of the most important and challenging components of the conventional four step travel demand modeling process. However, analysis of mode choice behavior seems to have been considered as a neglected component of the planning of cities in the developing world. Although several transport related studies have been conducted for Dhaka city, none has achieved the desired success in the planning of the transportation system; which is clearly reflected by the inadequately planned and developed transportation system of the city. The presence of motorized and non-motorized vehicles in the same right of way, slow moving and fast moving vehicles in non-lane based road network makes the transportation system of Dhaka city heterogeneous in nature. Moreover, recently the Government of Bangladesh has taken initiatives to introduce metro, or bus rapid transit, or elevated express way which requires huge investment and infrastructure development. Therefore, a proper understanding of the factors acting behind the choice of travel mode of individuals is imperative for the decision-makers. In this regard, the study has two major objectives: the identification of the factors that influence mode choice behavior of the city dwellers and the formulation of a representative mode choice model. The study area covers Dhaka City Corporation (DCC) and its surrounding areas comprising of 168 sq km. This area is divided into 94 Traffic Analysis Zones (TAZs) and the road network is comprised of 534.72 km with 1,068 nodes and 1,565, links. The data set used for this study is extracted from the Household Interview Survey (HIS) carried out for Strategic Transport Plan (STP) in 2004. The mode choice model is estimated using JICA STRADA 3.0 software package developed as a tool for the technical assistance program of JICA in the transport sector for developing countries. Personal attributes such as sex, age, household size and income, etc. and travel attributes such as mode selection, trip purpose, zone type, etc. of 4,825 relevant household records representing 19,792 people making 40,138 trips are analyzed to reveal the mode choice behavior in the study area. Rickshaw is found to dominate with 36% share followed by public bus with 32% share for all trips. Major portion of Home Based Education (HBE) and Home Based Other (HBO) trips share rickshaw; while major portion of Home Based Work (HBW) and Non Home Based (NHB) trips share public bus. Multinomial logistic regression using SPSS 17.0 is performed to analytically determine the effects of the attributes or factors on choice of travel mode. Household income is found to be the most important factor in determining choice of travel mode; gender being the next significant factor. Disaggregate approach using 32,422 individual level inter zonal trips are used to develop the mode choice model. With consideration of random utility maximization principle, multinomial logit (MNL) model is estimated for HBW, HBE, HBO and NHB trips. Alternative specific constant for each mode with variables travel time and travel cost/income are found statistically significant for HBW and HBO trips. For HBE trips, alternative specific constant for each mode with variables travel time, travel cost/Ln(income) and income specific to alternative modes are found statistically significant. For NHB trips, alternative specific constant for each mode with variables travel time, travel cost/ income and income specific to alternative modes are found statistically significant. A market segmentation test on gender is executed and found significant for HBW, HBE and HBO trips. In order to overcome the restrictive independence from irrelevance property of MNL model, nested logit (NL) approach is tested with different tree structures. Nested structure consisting Private transport (with walk, private bus, motor cycle, car/jeep in the first level), Para transit (with rickshaw, taxi, CNG in the first level), Public transit (with public bus only in the first level) in the second level is more explicable over MNL model in determining utility of modes for HBW, HBE and HBO trips. However, the NL model is not significant for NHB trips. Finally, an additional analysis for model estimation with BIOGEME is carried out to check the reasonableness of the model output in JICA STRADA.