Impact of Congestion, Pricing, and Travel Time Reliability on Travel Demand: Summary of Applied Models Estimated in US



Impact of Congestion, Pricing, and Travel Time Reliability on Travel Demand: Summary of Applied Models Estimated in US

Authors

P Vovsha, Parsons Brinckerhoff, US; M Bradley, Mark Bradley Research & Consulting, US; T Adler, Resource Systems Group, US

Description

The paper synthesizes various travel choice models estimated under conditions of congestion and pricing with either RP or SP data for such regions in US as New York, San Francisco, Los-Angeles, and Seattle.

Abstract

The paper synthesizes various travel choice models estimated under conditions of congestion and pricing with either RP or SP data for such regions in US as New York, San Francisco, Los-Angeles, and Seattle. Choice dimensions include route, mode, auto occupancy, and time-of-day, as well as joint formulations at a trip and tour level. The paper is focused on the following advanced features and recommends specifications for:
? Non-linear effects of travel time, cost, and distance. These can take the form of a linear effect plus some additional non-linear effect, general non-linear form by itself, or interaction of trip length with travel time and cost. Marginal sensitivities to travel time and cost have been consistently found to decrease with increasing trip length.
? Perceived travel time by congestion levels. Travelers have different travel perceptions of time spent in different conditions. These differences reflect the different levels of effort required for driving and/or different levels of uncertainty attached to travel time in different conditions.
? Perceived travel time by facility type. Some of studies included segmentation of travel time by facility type and reported large negative coefficients for travel time spent on local roads vs. freeways. Some other models addressed this by inclusion of a bias for freeway routes. In general, however, these results are less conclusive.
? Impact of income. Higher income results in a higher VOT. However, there is a great deal of variation in model specifications. The current research included such specifications as segmentation of time and/or cost (as well as reliability) coefficients by income, income-specific biases, and various forms of scaling travel cost by income.
? Impact of car occupancy. In many studies, a full cost sharing between the travelers in a carpool is assumed that is modeled by dividing the cost variable by occupancy. In the current study, several alternative specifications were tried including occupancy-specific time and cost coefficients. For most travel segments, a functional form where the cost variable was scaled by a non-linear function of occupancy proved to be the best. This implies that VOT of a carpool grows with occupancy slower than linearly. Additionally, inter-household carpools and carpools of adults exhibited a higher VOT comparing to intra-household carpools and carpools with children.
? Impact of gender, age, and other person characteristics. Quite contrary to many previously reported sources, females proved to have a large negative toll bias compared to males. Additional effects included gender impact on VOT through either time or cost coefficient and interactions of gender with other variables like presence of children. Age did not produce insignificant estimates in most cases.
? Travel time reliability and estimation of Value of Reliability (VOR). A method for generating travel time distributions was developed to support RP estimations. Various reliability measures were tested including travel time standard deviation, standard deviation per unit of distance, difference between the 80th or 90th percentile of travel time distribution and mean, etc. For most segments, the best results were achieved with a standard deviation per distance unit. This avoids multi-collinearity between average time, cost, and standard deviation of travel time that is pertinent to practically all RP data. In addition to RP studies, various reliability measures embedded in SP studies like frequency and length of delays were explored. There is compelling evidence that VOR lies in the range of 0.6-1.5 of VOT.
? Toll-averse bias. If toll and non-toll alternatives are separated either in a route choice or mode choice context, a significant negative bias for toll alternatives is found on the top of all other variables. This finding proved to be persistent even in regions like New York where toll facilities have been in place and publicly accepted for many years.
? Situational variables and time pressure. In general, people with busier daily patterns have a higher VOT/VOR. These effects were captured by introducing various time-pressure measures in the framework of time-of-day choice and mode choice.
? Unobserved heterogeneity. After accounting for all of the systematic effects above, some differences in preferences remain to be captured as random heterogeneity. And, it is often the case that that random heterogeneity is observed in both the travel time and travel cost coefficients as well as in the reliability coefficient. Different approaches for randomization are discussed and successful examples with a reasonable improvement of the likelihood function are reported.

Publisher

Association for European Transport