Comparing spatial patterns of crowdsourced and conventional bicycling datasets

摘要

Conventional bicycling data have critical limitations related to spatial and temporal scale when analyzing bicycling as a transport mode. Novel crowdsourced data from smartphone apps have the potential to overcome those limitations by providing more detailed data. Questions remain, however, about whether crowdsourced data are representative of general bicycling behavior rather than just those cyclists who use the apps. This paper aims to explore the gap in understanding of how conventional and crowdsourced data correspond in representing bicycle ridership. Specifically, we use local indicators of spatial association to generate locations of similarity and dissimilarity based on the difference in ridership proportions between a conventional manual count and crowdsourced data from the Strava app in the Greater Sydney Australia region. Results identify where the data correspond and where they differ significantly, which has implications for using crowdsourced data in planning and infrastructure decisions. Fourteen count locations had significant low-low spatial association; similarity was found more often in areas with lower population density, greater social disadvantage, and lower ridership overall. Five locations had high-high spatial association, or were locations of dissimilar rank values indicating that they did not have a strong spatial match. Higher coefficients of variation were associated with population density, the number of bicycle journey to work trips, and percentage of residential land use for the significant locations of dissimilarity. IRSD and bicycle infrastructure density were lower than the locations that were not significantly dissimilar. For the significant locations of similarity, all coefficient of variation measures were lower than the locations that were not significant. Areas where ridership show locations of similarity are those where it may be suitable to substitute conventional data for the more detailed crowdsourced data, given further investigation into potential bias related to rider demographics.