Land use regression models for active travel: Spatial estimates of walking and biking for exposure assessment

Abstract

Introduction: Physical inactivity is a significant global health burden. Transportation planners and public health officials need better spatial estimates of active travel to assess exposure to hazards and deploy health-promoting infrastructure. Applying Land Use Regression (LUR) to bicycle and pedestrian traffic counts is a potentially useful tool for generating spatial estimates of active travel. Methods: We present LUR models based on peak-period (4-6pm) counts of pedestrian and bicycle traffic in Minneapolis, MN. Our count database (n=954 observations; 471 locations) has sufficient spatial density (~3 locations km-2) to develop spatially resolved models (~100 m resolution). We apply an LUR approach originally developed for air quality (i.e., stepwise linear regression) to vary the spatial scale of independent variables (e.g., land use). Using LUR we compare fully-specified (i.e., statistically optimal) models to reduced-form models based on fewer variables with theoretical validity. Results: Our fully-specified (i.e., statistically optimal) models have reasonable goodness-of-fit (pedestrian [bicycle] adjusted R2: 0.53 [0.58]), but included many variables that violated a priori assumptions about direction of effect. We developed reduced-form models using a supervised variable selection process that resulted in 4-9 predictor variables per model. The reduced-form models have similar goodness-of-fit (pedestrian [bicycle] adjusted R2: 0.50 [0.46]) as the statistically optimal models. Our models include variables with both large (e.g., >1 km; industrial area and population density) and small (100-400 meters; bicycle facilities, retail area, open space) spatial scales. Conclusion: Our results suggest that (1) LUR can be used to estimate spatial patterns of active travel, (2) reduced-form models explain nearly as much of the variation in active travel as fully-specified models, and (3) using LUR to model active travel could be a useful tool in exposure assessment.