Assessing the accuracy of routing engines in replicating children's walking routes to school: a comparative study of...

Active school travel (AST) can significantly increase children's physical activity. AST research often utilises global positioning system (GPS) devices to objectively track the routes children take to school. However, GPS-based studies are costly and often subject to recruitment bias which limits population representativeness. Routing engines, such as Google Maps, designed to help people navigate might be low-cost alternatives to the use of GPS if the engines were able to accurately model the routes children took in AST. To date, no studies have systematically evaluated whether routing engines can serve as a viable alternative for modelling children's walking routes to school. We assessed whether widely available routing engines can approximate children's walked routes, providing an internationally transferable, low-cost alternative.

We compared three routing engines: Google, Mapbox, and Open Source Routing Machine (OSRM), in replicating the GPS measured walking trajectories of 233 school children (age 10-11) across urban and rural Scotland. The percentage of overlap (overlap accuracy, OA), between routing engine predictions and actual GPS tracks was assessed.

Mean OA was 67.6% for OSRM, 66.8% for Mapbox, and 62.6% for Google, indicating that approximately two-thirds of each predicted route overlapped with the GPS-measured track. These values were substantially higher than a shortest-path baseline computed on the Ordnance Survey (OS) road network (45.0%). No significant differences in OA were found by sex, deprivation, or urban/rural setting. However, route distance was negatively associated with OA (r = - 0.16 to - 0.23, p < 0.05). At an individual level, one-third of participants showed consistent results across all engines, whilst 49.4% had one discordant engine and 17.2% had mutually inconsistent engines.

Contemporary routing engines reproduce roughly two-thirds of children's walked routes to school, without systematic differences by common sociodemographic or contextual factors. Overall, contemporary routing engines can complement GPS in children's mobility research and practice, offering an empirical reference point for future AST studies.

Not applicable.