Testing a generalized model to map bicycle ridership using crowdsourced data in a racially diverse city

Bike counts are important for measuring safety and prioritizing infrastructure investment. Crowdsourced data, generated by the bicycle app Strava, provides bike counts at unprecedented resolutions and extents for the sample of bicycle counts where the riders are contributing data. Researchers have developed models to expand the sample of bicycle counts measured in crowdsourced data to all bicycle counts. However, these models were developed and tested in cities with predominantly white populations, and it is unknown how well these models apply to cities with racially diverse populations. In this work, we tested expansion methods in Oakland California, one of the most racially diverse cities in the United States. We built a bike count and geographic predictor data set, collected new count data to improve representativeness of training data, used lasso regression to select and build regression models, and compared the results to the cities previously published. We found that the variables selected and accuracy of estimation of average annual daily bicyclists (AADB) by road segment aligned with other cities. Our findings, that crowdsourced data can adequately represent bicycling in racially diverse cities, will allow crowdsourced data to be used in racially diverse cities, which are often underserved by transportation infrastructure. • We bias-corrected crowdsourced bike count data in a racially diverse city • Bias-correction models performed with similar accuracy as in less-diverse cities • Crowdsourced bike count data is valuable in racially diverse cities