In the first effort on this scale, we sourced data from 11 types of surveys spanning a period of 23 years (1989 – 2012) and covering 150 countries.
Map showing number of survey sources for each country in bicycle ownership dataset
Given that the data were sparse, we applied unsupervised learning to discover any underlying patterns that might useful to researchers and other stakeholders in sustainable active transport. Using a suite of methods: dynamic time warping (data alignment), the gap test (optimal cluster numbers) and hiearchical agglomerative clustering, we found four prevailing groups of characteristic ownership levels.
To find the group trends, we used the weighted average rolling mean over a five-year window. Extrapolating from the 1.25 billion households represented in the surveys we mined and the common question of interest: “Does your household have at least one bicycle?” we estimate that 580 million or more bicycles were available to the world’s households as of 2012.
Finally, we shed light on aggregate global trends in ownership, with a spotlight on India and China, as they together account for a quarter of the world’s households. We conclude that globally, bicycle ownership has not been on the rise.