Modeling The Health Effects of Adding Bicycle & Pedestrian Paths At The Census Tract-Level (Preprint)

Abstract

Adding additional bicycle and pedestrian paths (BPPs) to an area should lead to improved health outcomes of residents over time. However, quantitatively determining which areas stand to benefit more from BPPs, how many miles of BPPs are needed, and the health outcomes that may be most improved remain open questions.

Our work provides and evaluates a methodology that offers actionable insight for city-level planners, public health officials, and decision makers tasked with the question: “To what extent will adding specified BPP milage to a census tract improve residents’ health outcomes over time?”

We conduct factor analysis on data from the American Community Survey (ACS), CDC 500 Cities Project, Strava and BPP location and usage data from two different cities (Norfolk, VA and San Francisco, CA). We construct two city-specific factor models and use an algorithm to predict the expected mean improvement that a specified amount of BPP miles contributes to identified health outcomes.

We show that given an amount of additional BPP miles in 2012 and a specific census tract, our models forecast 2017 health outcome improvements more accurately than two alternative approaches for both Norfolk, VA and San Francisco, CA. Furthermore, for each city we show that the additional accuracy is a statistically significant improvement (P < .005 in every case) compared to the alternate approaches. For 2017, in Norfolk, VA (n=29 census tracts) our approach estimates, on average, the: (a) % of individuals with high blood pressure in the census tract within 1.40%, (b) % of individuals with diabetes in the census tract within 1.63%, and (c) % of individuals who suffer more than two weeks worth of poor physical health days in the census tract within 1.82%. For San Francisco (n=59 census tracts), in 2017 our approach estimates, on average, the: (a) % on individuals who suffer a stroke in the census tract within 1.79%, and (b) rate of % of individuals with diabetes in the census tract within 1.27%. The data and source code from our methodology, evaluation and research artifacts are provided.

We propose and evaluate a methodology to enable decision-makers to: (1) weigh the extent to which two BPPs of equal cost proposed in different census tracts improve residents’ health outcomes, (2) identify areas where BPPs are unlikely to be effective interventions and other strategies should be employed and (3) quantify the minimum amount of additional bicycle path miles needed to maximize health outcome improvements. Our methodology shows statistically significant improvements, compared to alternative approaches, in historical accuracy for two large cities, in different geographic areas and with different demographics.