Towards confirmatory data analysis? Deriving and analysing routing information for an origin-destination bike share dataset

Data collected from urban bike share schemes allow observed travel behaviours to be analysed on a uniquely large scale. Exploring such timed origin-destination (OD) data from the London Cycle Hire Scheme (LCHS), we previously generated detailed insights into spatiotemporal patterns of travel and suggested new hypotheses for their motivation. A limitation was that with only the origins and destinations of cycle journeys, little was known about the nature and context of likely cycled routes. In this study, we use the CycleStreets routing engine to derive routing information for every OD pair made through the LCHS. From these suggested routes, we collect heuristics for the nature of each journey. Information on the number of signalled junctions encountered, on any bridges crossed, as well as a proxy for the busyness of suggested routes is recorded. We then analyse over 5 million journeys made by LCHS members during a 12-month period (September 2011 – September 2012). Focussing on LCHS journeys that involve crossing the River Thames, we observe differences in male and female cyclists’ apparent use of bridges, which appear to be strongly related to a commuting function. Studying heuristics of suggested routes over these bridges, we find some evidence to suggest that women may be underrepresented amongst commuting journeys that involve a river crossing because those very journeys are associated with relatively busy and demanding routes. We also find evidence that the nature of frequently cycled journeys involving a river crossing might explain imbalances in the direction of journeys made over the river when we select periods of more discretionary activity – when studying weekend journeys. These findings are nevertheless quite speculative. A number of confounders cannot be easily accounted for within this analysis: the economic geography of the city, spatial interactions between docking stations at particular space-times and the relative availability of transport alternatives. Perhaps most importantly, our analysis assumes that routes suggested by the routing algorithm closely reflect individuals’ actually cycled routes.