Previous studies of the effect of ridesharing on traffic fatalities have yielded inconsistent, often contradictory conclusions. In this paper we revisit this question using proprietary data from Uber measuring monthly rideshare activity at the Census tract level. Virtually all existing studies are based on publicly-available information about Uber entry dates into U.S. cities, but our data reveal that an indicator variable for Uber availability is a poor proxy of rideshare activity. We show, for example, that Uber entry explains less than 2% of the tract-level variation in ridesharing, reflecting the enormous amount of variation in ridesharing activity both within and across cities. Estimating “naive” regressions of alcohol-related fatalities on Uber entry, we find inconsistent and often statistically insignificant estimates. However, when we use the detailed proprietary data, we find a robust negative impact of ride sharing on traffic fatalities. Impacts are larger during nights and weekends, and similar and statistically-significant across a range of alternative specifications. Overall, our results imply that ride sharing has decreased U.S. alcohol-related traffic fatalities by 3.6% and reduced total U.S. traffic fatalities by 2.2%. Based on conventional estimates of the value of statistical life, the annual life-saving benefits range from $1.4 to $2.9 billion and are of similar magnitude to estimates of producer surplus captured by Uber shareholders or non-safety-related consumer surplus captured by Uber customers.
