Smartphone Mobile Sensing and Geospatial Physical Activity Space Analytics in the PASTA-LA study

Abstract

Physical activity is intimately linked to the built environment and health. Major on-going efforts are being made to increase physical activity rates in urban settings. One such effort, quickly becoming prevalent across many global cities, is the implementation of bike sharing programs. In October 2017, UCLA launched the Bruin Bike Share program. It provided a unique opportunity to conduct a before-and-after natural intervention study, the Physical Activity through Sustainable Transport Approaches in Los Angeles (PASTA-LA) study, to assess the health-related outcomes of bike-sharing programs. During the 2016-2018 study period, the PASTA-LA study enrolled 440 physically-capable participants from the UCLA-Westwood area. Participants were administered online questionnaires and a smartphone app (MOVES) was deployed on their phones for location and activity tracking. A nested sample of 140 participants were also outfitted with research grade GPS and accelerometers for comparison. Across all participants, over 30 million observations of location and activity were recorded. Using these ‘big data’, we can address our specific research questions: (1) can smartphones be used instead of expensive research-grade devices without sacrificing measurement validity, and (2) can these ‘big data’ be used to model Physical Activity Spaces (PASs) to better quantify environmental context. We have addressed these research questions using advanced geospatial and geostatistical approaches. To evaluate context variable allocation for individuals, we have used location and activity levels (steps) to demonstrate the predictive accuracy of four approaches to defining physical activity space: (1) circular home buffers, (2) directional ellipses, (3) machine learning route allocation models, and (4) kernel density ranking. These models can be used to quantify variables such as exposure to green space, dose of air pollution, and other neighborhood-based statistics that are instrumental for spatial models of association (e.g. such as the impact of heat exposure on physical activity rates, as moderated by green space access.)