Learning from Law Enforcement

This paper studies how punishment for past offenses affects future compliance behavior and isolates deterrence effects mediated by learning. Using administrative data from speed cameras that capture the full driving histories of more than a million cars over several years, we evaluate responses to punishment at the extensive (receiving a speeding ticket) and intensive (tickets with higher fines) margins. Two complementary empirical strategies - a regression discontinuity design and an event study - coherently document strong responses to receiving a ticket: the speeding rate drops by a third and re-offense rates fall by 70%. Higher fines produce only a limited additional effect. All responses occur immediately and are persistent over time, with no 'backsliding' towards speeding even two years after receiving a ticket. Our evidence rejects unlearning and temporary salience effects. Instead, it supports a learning model in which agents update their priors on the expected punishment in a 'coarse' manner. Additional results indicate that learning from law enforcement affects drivers' behavior more broadly, including spillovers on non-ticketed drivers.