Exploring the importance of detailed environment variables in neighborhood commute mode share models

Abstract

This paper analyzes the relationship between detailed neighborhood environment variables and commute mode share using a dataset drawn from across the United States and includes model validation results. Representing one of the first studies of its kind, we use United States journey-to-work data to explore the following questions: 1) Which detailed environment variables have significant associations with the proportion of people in a neighborhood who take public transit, walk, or bicycle to work? 2) Does adding detailed environment variables to existing, nationally available neighborhood variables improve the predictive accuracy of work commute mode share models? We use a set of 120 randomly selected census tracts to estimate fractional multinomial logit models that predict walk, bicycle, transit, and automobile commute mode shares. The Base Model includes a set of nine significant, nationally available variables identified from a previous analysis of 5,000 tracts. We test 18 additional detailed neighborhood environment variables and identify five variables that have significant associations with commute mode share: sidewalk coverage (positive association with transit and walk), proximity to a rail station (positive association with transit), bicycle facility density (positive association with bicycle), freeway presence (negative association with walk), and mixed land use (positive association with transit, walk, and bicycle). While these detailed environment variables add clarity to our understanding of factors that influence travel behavior, our validation analysis using 50 separate census tracts does not provide conclusive evidence that these variables improve model accuracy. Further studies with larger sample sizes are needed to determine the optimal set of variables to include to predict automobile, transit, pedestrian, and bicycle commute mode shares.