We employ a novel data set of 219,000 workers’ compensation claims and pay records to understand whether traffic officers account for injury risk when making daily labor supply decisions. We use variation in the leave of coworkers as an instrument inducing officers to work. We find that officer appear to mitigate their own injury risk: on average an officer at the 60th percentile of willingness to work is 2.6 times more likely to be injured as an officer at the 80th percentile. Using variation in hourly wages, we also calculate the marginal value of injury risk, defined as the value of a 1 percentage point reduction in injury probability for officers that are indifferent between working and staying home. A 1 percentage point decrease in injury risk is valued on average between 31 and 67 dollars in earnings. Using our model, we simulate the benefits of switching to a shift-auction mechanism of shift assignment. We find that shift auctions more fully leverage officer’s natural tendency to select against risk, and as a result reduce the injury rate by 38 percent compared to a random list mechanism.