Abstract:
Dhaka, the capital city of Bangladesh and the home of 15 million people, is subjected to acute traffic
congestion on a regular basis resulting in lost productivity, fuel wastage, commuter frustration and
environmental degradation. The city is perhaps the only megacity with no well organized public
transport system and one of the very few ones without Mass Rapid Transit (MRT). In Strategic Transport
Plan for Dhaka (STP, 2005) recommendations have been made to launch new MRT systems like Bus
Rapid Transit (BRT) and Metro in order to strengthen the public transport system of the city. Planning of
these new systems warrants comprehensive mode choice models that can help in quantifying the
relative importance of attributes, determining the Value of Time (VOT) for cost-benefit analysis,
predicting ridership, etc. The existing models however do not account for the deficiencies of existing
data like missing choice sets, measurement errors in the level of service (LOS) data, lack of information
regarding the new modes etc. and can lead to incorrect travel demand predictions.
This has prompted the current research where Stated Preference (SP) data has been collected to
capture the preference for proposed new alternatives (MRT), methodologies have been developed to
address the other limitations of the existing data and a comprehensive mode choice model has been
developed combining Revealed Preference (RP) and SP data.
In the SP survey conducted in the research, respondents have been presented with choice scenarios
that included BRT and Metro alongside their current modes. Different levels of three attributes (travel
time, travel cost and waiting time or frequency) were used to describe the new alternatives. The
attributes and associated levels were selected as the most important attributes as perceived by the
respondents on the basis of the findings of an initial survey.
To address the unobserved choice sets of the respondents in the available RP data a choice set
generation model has been developed using SP data. The estimated parameters of the developed
model have been used to predict the choice sets of the respondents in the RP data probabilistically.
Regression analysis has been done to address the measurement errors of the travel time derived from
network analysis
Discrete choice models have been developed using the corrected RP data and the collected SP data and
the coefficients of the utility parameters have been estimated using a maximum likelihood approach.
The observed taste heterogeneity of the respondents have been taken into account by the introduction
of socio-economic variables like income, age, gender, occupation, employment, etc into the model and
market segmentation tests have also been performed. The VOT values from the combined model are
plausible compared to the values obtained from previous choice models as well as the disjoint RP
and SP models. Further, the methodologies proposed in the current research can be useful tool for
transport related analysis in other developing countries facing similar data issues.
