Can targeting messages to high-risk drivers prevent crashes?

Why is this issue important in DC?
Every year, DC records dozens of driving-related fatalities, thousands of traffic crashes, and hundreds of thousands of driving violations.¹ Mayor Bowser's Vision Zero initiative takes a multifaceted approach, including enforcement and education, to eliminate traffic-related deaths. The District’s Automated Traffic Enforcement (ATE) system automatically takes photos and videos of a vehicle and license plate if drivers run red lights, roll through stop signs, or exceed speed limits. Evidence from other jurisdictions suggests that drivers with multiple violations are more likely to be involved in traffic crashes, and that a small number of drivers are responsible for a large number of crashes.²

What did we do?
We used data on vehilces with as few as two camera tickets to build a statistical model to predict their likelihood of being involved in a serious crash. Then, we randomly assigned the registered owners of approximately 100,000 high-risk vehicles to receive a customized letter, text message, both, or neither. The messages included information about their vehicle and its previous violations and communicates that their vehicle is at higher risk of a crash than others. Finally, we evaluated whether registered owners who received these messages had fewer red-light, stop-sign, and speeding violations, and ultimately, crashes, compared to the registered owners of high-risk vehicles who did not receive the messages.

What did we learn?
Our model was able to correctly identify 65% of vehicles involved in a crash in the following year. This was a seven percentage-point improvement over ranking vehicles by their total number of tickets and a 47 percentage-point improvement over a rule that identified any vehicle with over 10 tickets of any type or at least two tickets for red-light or severe speeding.

We found that the messages had no detectable effect on tickets—neither total tickets nor tickets for especially risky driving behavior—after three or 12 months. This was true whether we sent a letter, a text message or both. We also found no measurable impact on crashes after 12 months. These results suggest that it may be difficult to move the needle on driver behavior with a one-time message on its own.

What comes next?
What we've learned will inform how targeted messaging can complement other efforts to reduce serious crashes in the District. These results suggest that this message, sent once, did not reduce drivers’ future tickets or crash involvement after one year. Future efforts will likely still target messages to specific groups, which is low-cost and supported by the results of the predictive model and by other research.³ At the same time, the message content, frequency, or delivery mode may change. More research would be needed to understand whether these changes can impact driver behavior.

What happened behind the scenes?
Before finalizing the message design, we user-tested the messages at the DMV and online, and we updated the design with the feedback. We also interviewed owners of high-risk vehicles and heard that the letters were compelling. One reason was that the letters included ticket history, which implied they had spent more than they thought on tickets. Strong reactions that didn’t come up in user-testing or interviews surfaced in replies to the texts. Some replied that they felt harassed or judged. Others responded that they felt the message itself made a crash almost fated. Together, these findings should inform the design of future interventions.