In a National Transit Institute course on “Coordinating Land Use and Transportation,” co-taught by Robert Cervero, Uri Avin, and the PI on this project, the analytic tools session began with a hypothetical: assume that all households, jobs, and other trip generators are concentrated in a walkable village rather than segregated by use and spread across a traffic analysis zone in the standard suburban fashion. The instructor then asks: How would the outputs of conventional four-step travel demand models differ between these two future land use scenarios? The answer, to most participants’ surprise, was “Not at all.”
Conventional four-step travel demand models are used by nearly all metropolitan planning organizations (MPOs), state departments of transportation, and local planning agencies, as the basis for long-range transportation planning in the United States. In the simplest terms, the four-step model proceeds from trip generation, to trip distribution, to mode choice, and finally to route assignment. Trip generation tells us the number of trips generated (produced or attracted) in each traffic analysis zone (TAZ), usually based on some prediction of vehicle ownership. Trip distribution tells us where the trips go, matching trip productions to trip attractions by considering the spatial distribution of productions and attractions as well as the impedance (time or cost) of connections. Particularly tricky are predictions of trips that remain within the same zone. Mode choice tells us which mode of travel is used for these trips, factoring trip tables to reflect the relative shares of different modes. Route assignment tells us what routes are taken, assigning trips to networks that are specific to each mode. A flaw of the four-step model is its relative insensitivity to the so-called D variables. The D variables are characteristics of the built environment that are known to affect travel behavior. The Ds are development density, land use diversity, street network design, destination accessibility, and distance to transit.
This report develops a vehicle ownership model (car shedding model), an intrazonal travel model (internal capture model), and mode choice model that consider all of the D variables based on household travel surveys and built environmental data for 32, 31, and 29 regions, respectively, validates the models, and demonstrates that the models have far better predictive accuracy than Wasatch Front Regional Council (WFRC) and Mountainland Association of Governments’ (MAG) current models.