Estimating Mixed-Mode Urban Trail Traffic Using Negative Binomial Regression Models

Abstract

Data and models of nonmotorized traffic on multiuse urban trails are needed to improve planning and management of urban transportation systems. Negative binomial regression models are appropriate and useful when dependent variables are nonnegative integers with overdispersion like traffic counts. This paper presents eight negative binomial models for estimating urban trail traffic using 1,898 daily mixed-mode traffic counts from active infrared monitors at six locations in Minneapolis, Minnesota. These models include up to 10 independent variables that represent sociodemographic, built environment, weather, and temporal characteristics. A general model can be used to estimate traffic at locations where traffic has not been monitored. A six-location model with dummy variables for each monitoring site rather than neighborhood-specific variables can be used to estimate traffic at existing locations when counts from monitors are not available. Six trail-specific models are appropriate for estimating variation in traffic in response to variations in weather and day of week. Validation results indicate that negative binomial models outperform models estimated by ordinary least squares regression. These new models estimate traffic within approximately 16.3% error, on average, which is reasonable for planning and management purposes.