Identifying and Intervening with High-Risk Drivers

Project Description

Every year there are dozens of recorded driving-related fatalities, thousands of traffic crashes, and hundreds of thousands of driving violations (speeding and dangerous driving) in the District. With 27 traffic fatalities in 2019 and 209 crashes resulting in injuries to date in 2020, we are far from the District’s goals of achieving zero traffic fatalities by 2024.

DDOT, the Department of Motor Vehicles (DMV), the Metropolitan Police Department (MPD), and The Lab @ DC in the Office of the City Administrator (OCA) are working together on this project to identify drivers at high risk of being involved in a serious crash and to test proactive interventions to these high-risk drivers.

Project Need

A joint study (not yet published) conducted by the New York City Department of Transportation (NYC DOT) and the University of Chicago (UChicago) analyzing traffic violations and crash data found that those with repeat speeding and red-light violations were more likely to have subsequent crashes. Researchers from this study found that repeat offenders are 30% more likely to be involved in a serious traffic crash.[1] Another (internal and not publicly available) driver behavior study conducted by the University of Chicago for the Chicago Department of Transportation found that previous citations and crashes can predict subsequent crash risk. Their preliminary results found that drivers with at least two tickets in the past three months and at least three tickets in the past 2.5 years are 3 times more likely to be involved in a crash in the following year. A different study found that a small share of drivers (5%) were responsible for 35% of crashes in Louisiana (Das et al, 2015).

Research suggests that many dangerous drivers are simply not aware of: (1) the fact that they are driving unsafely; (2) the risk associated with their dangerous driving; and (3) how far out of the norm their dangerous driving is.[2] Automated notifications have also been shown to increase desirable behavior and reduce undesirable behavior across many contexts, including driving. For example, in a study of teenage drivers, alerting both the teenagers and the parents of teenage drivers of risky behavior occurring in their cars can reduce risky driving.[3] The District does something similar, by using Automated Traffic Enforcement (ATE) systems to enforce traffic safety and regulations for red light and speeding violations. ATE systems do this by automatically taking photographs of the rear of the vehicle and its license plate if the driver violates regulations, then sends a citation and fine to the registered vehicle owner’s address. However, these are reactive measures towards reducing risky driving behavior. Our study proposes to build upon this system further by targeting proactive measures to risky drivers to reduce crashes.

[1] NYC DOT study has not been made public yet.
[2] Leonard Evans, Traffic Safety (2004).
[3] Simons-Morton et al, The Effect on Teenage Risky Driving of Feedback From a Safety Monitoring System: A Randomized Controlled Trial. Journal of Adolescent Health 53 (2013).

Desired Outcome & Expected Benefits

Outcomes: Our primary outcome of interest is the level of traffic violation levels, measured by the number of drivers with repeat red-light and speeding violations. We expect to see fewer red-light and speeding infractions for drivers in the treatment group, i.e. those receiving the intervention (notification that they are a risky driver) compared to their baseline level compared to those who do not receive the intervention.

How Results could be Implemented: If we see statistically significant results after evaluating the effectiveness of the intervention, The Lab will work with DDOT to deploy the model so that DDOT can use the predictions and send notifications to all drivers who are predicted to be at high risk.

Benefits to the District: In addition to helping the District reduce the level of traffic injuries and fatalities by encouraging safer driving behaviors, this project would benefit the District by moving us closer to our Vision Zero goals of reaching zero traffic fatalities by 2024.


DDOT and The Lab will collaborate to design the modeling and intervention for this project. There are two key components to the intervention:

  1. analysis of data from the District’s ATE systems and MPD crash data, to predict a driver’s likelihood of being involved in a crash
  2. proactive intervention(s) to reduce risky behavior for drivers likely to be involved in a crash

The Lab @ DC will complete the analysis (#1) with existing staff time and resources. DDOT is funding the costs associated with proactive interventions (#2).

Predictive Model

Our model will use regression and machine learning methods to predict the likelihood that a driver will be involved in a crash in the next year.  The goals of our model are to use the model predictors (e.g., features describing the people involved in a crash and/or receive a citation, features of the vehicles involved, locations, weather conditions, time of day, season) to  (1) develop risk levels of being involved in a crash and (2) to develop “profiles” of risky drivers to target our proactive interventions. We also plan to evaluate whether the impact of an intervention varies with a driver’s predicted probabilities of being involved in a crash.


In partnership with The Lab @ DC, DDOT will use the model’s predictions to target proactive interventions to risky drivers. While we do not know yet what intervention will be the most effective in changing drivers’ behaviors, we’ll bring a behavioral and evidence lens to the messaging, for instance, some examples could be:\

  • Loss-aversion. "Your household is at risk of losing your vehicle -- and very possibly a life -- due to risky driving. The [make and model] has been cited __ times for dangerous driving, putting you at risk of losing your car insurance and your car, and placing the driver at high risk of an accident. Don't lose your family member or your car -- determine who is driving dangerously and remind them to drive safely."
  • Social norms & Pluralistic ignorance. "The vast majority of drivers are safe, but someone in your household is driving very dangerously. Your vehicle is in the top __% for riskiest driving in the District. Save the lives of your family members and others on the road. " [some evidence-based advice]
  • Social Influence. “You can reduce the risk of [losing your vehicle / being in an accident] by making sure that everyone in your household knows the rules of the road, including speed limits and how to change lanes safely.”
  • Identity-affirmation. “Your vehicle is in the top __% for riskiest driving in the District. As the vehicle owner, you control what happens next. You are uniquely capable of reducing driving-related risks in your vehicle. Take action! ... ”

We will prioritize at least one such intervention for rigorous testing based on feedback from experts, user-testing, and feasibility of random assignment.

Experimental Design

The evaluation design for this project will depend, in part, on the intervention(s) designed. Broadly, we intend to use a randomized controlled trial to evaluate the impact of an intervention on the number of speeding and red-light violations in the District.

Project Oversight

DDOT Stakeholders

Vision Zero team: Linda Bailey (Unlicensed) Former user (Deleted) 

Automated Enforcement: Kelli Raboy (Deactivated) Former user (Deleted) 

Peer Reviewers

Quarterly updates

QuarterProgress this quarterIssues Encountered
FY21 Q4Scope for DDOT funding refined. Funds obligated.
FY22 Q1

Project Materials

Any relevant materials, including problem statement, scope of work, interim deliverables, reports, data can be uploaded below.

  File Modified
No files shared here yet.