Accounting for Spatial Heterogeneity Using Crowdsourced Data

Given the numerous benefits of active travel (human-powered transportation), in this paper, we argue that using crowdsourced data and a spatial heterogeneity treatment enhances the predictive performance of data modelling. Using such an approach thus increases the amount of insight that can be obtained to improve active travel decision-making. In particular, we model cyclists’ route choices using data on cycling trips and street network centralities obtained from Strava and OSMnx, respectively. It was found that: i) the number of cyclist trips is spatially clustered; and ii) the spatial error model exhibits a better predictive performance than spatial lag and ordinary least squares models. The results demonstrate the ability of the fine-grained resolution of crowdsourced data to provide more insights on active travel compared to traditional data.