Development of a neighborhood commute mode share model using nationally-available data

Practitioners often use demand models to predict how neighborhood-level land use, infrastructure, demographic, and other changes may impact transportation systems. Few of the models available to predict automobile, transit, bicycle, and pedestrian travel are based on easily-accessible data, which creates barriers for transportation agencies with limited data or modeling resources. We help fill this gap by developing a fractional multinomial logit model that estimates United States neighborhood work commute mode shares using existing, nationally-available data from 5000 randomly-selected, non-adjacent census tracts. After controlling for socioeconomic characteristics, the model shows that public transit, walk, and bicycle commuting are associated with higher population and employment density, more housing constructed prior to 1940, and more rental housing. Public transit and walk commuting are associated with being in a Northeastern state, automobile commuting is associated with being in a Southern state, and bicycle commuting is associated with being in a West Coast state. Validation of the model using a separate set of 1000 census tracts and application of the model in the Milwaukee metropolitan region show promise but also highlight several limitations of this approach. This sketch-planning model is a building block for future practically-oriented neighborhood commute mode share models.