Two-Stage Vehicle Availability Model

Abstract

It is well accepted that travel forecasting models benefit from the stratification of travel markets by socioeconomic levels. The number of vehicles available is a key indicator of that level. Using this variable requires that the proportion of households by vehicles available be forecast for each zone. An improved submodel for forecasting vehicle availability by incorporating transit accessibility and land use indicators along with the usual demographic variables is described. This model uses a two-stage approach. The first stage is similar to many other models in current use. In the first step, a lookup table is used to identify an initial estimate of the proportion of 0-vehicle, 1-vehicle, 2-vehicle, and 3+-vehicle households on the basis of the household's size (1–4 +), number of workers (0–3 +), and income quartile (1–4). This lookup table has 52 cells, with each cell containing the four proportions by vehicles available. Census Public Use Microdata Sample data were used to create this lookup table. The second stage applies an incremental logit model to the initial proportions. In this step, the effects of transit accessibility and land use form on vehicle availability are modeled. Accessibility and density measures are used to calculate a “disutility” measure, which is then used to modify the initial percentages. Good transit service and high development density are associated with lower vehicle ownership. Vehicle availability models of this type recently have been successfully calibrated for the Washington, D.C., and Seattle, Washington, areas.