Automate as much traffic enforcement as possible

Let's welcome our robot overlords

The New York Times did a great story this week about how frequently traffic stops turn deadly, along with a subsidiary piece looking at the financial motivations underlying some municipal traffic enforcement.

Traffic enforcement is so fraught in large part because of the way policing culture interacts with widespread firearms ownership in the United States. Unlike in most European countries, any given person in the U.S. could have a firearm, so American cops are taught to treat the basic “license and registration” routine as a potentially lethal confrontation. And while some officers are undoubtedly spared injury by this approach, it transforms routine traffic enforcement into a nerve-wracking experience for civilians. At worst, you get the case of Philando Castile, who was shot seven times by police officer Jeronimo Yanez during a traffic stop for no reason other than an inaccurate fear that he might be reaching for the firearm he was legally carrying.

And the reliance on fines for revenue is clearly a problem. If fining people becomes a profitable activity for a municipal government or local police department, it’ll naturally distort decision-making about how to allocate resources.

But these very real issues with American traffic enforcement are sometimes invoked in a problematic way, as if there’s no real reason to have rules about safe driving and all enforcement is just a big scam.

The best solution is to largely automate this function. We have the technology to detect vehicle speed, take images of cars breaking the rules, read license plate numbers from those photos, and fine the responsible drivers. When these systems are introduced, they tend to generate a lot of complaints. But they generate a lot of complaints because heavily discretionary enforcement by police officers is very ineffective, and the many people who frequently break the rules don’t like that automatic enforcement leads to them getting caught. Worse, the cameras even enforce the rules against middle-class white citizens who are more likely to catch a break from an officer exercising discretion than to get shot.

But this is all actually good. Traffic rules should be enforced rigorously, uniformly, and without people getting killed. This makes the roads safer for both drivers and pedestrians and frees up police officers to do something more useful with their time.

Discriminatory traffic enforcement is a problem

A classic study on racial discrimination in traffic enforcement is from Cheryl Phillips and Sharad Goel, who studied an extensive set of police stop data and found that the Black/white disparity in the odds of being pulled over vanishes at night when it’s hard to see the driver’s face.

Other studies indicate that officers use harsher tones when interacting with Black civilians during traffic stops. And while I’ve heard pushback that causation may go in the opposite direction (meaning the difference in tone is caused by Black civilians having a worse attitude about being pulled over), I would also have a worse attitude about being pulled over if I were aware that people with my skin color were likely to be pulled over on a discriminatory basis. The downstream impact of doing banal traffic enforcement in a discriminatory manner can have deleterious social consequences, even if you take a fairly rosy view of police behavior.

And as Dean Knox and Jonathan Mummolo argued in an important paper, the existence of this bias undermines studies that show no racial disparity in police shootings. Even if the odds of a stop ending in death are unbiased conditional on the stop taking place, the existence of bias in the decision to pull someone over in the first place introduces a huge disparity.

In general, everyone should be less racist. But biases and “statistical discrimination” are hard to expunge from human existence, and generally, if you reduce the scope of discretion, you get a more fair outcome. The potential for algorithmic bias is certainly a valid concern, but when Sabrina Howell, Theresa Kuchler, David Snitkof, Johannes Stroebel, and Jun Wong looked at Paycheck Protection Program lending, they found that Black-owned businesses did better with algorithmic lenders than with traditional banks and better with big banks than with small banks.

The more automated the process, the less discriminatory it is.

Speeding is bad

Washington, D.C. has a lot of automated traffic enforcement, which I think is appropriate.

This, in turn, has generated a critique from William Farrell at the DC Policy Center and John Harden at the Washington Post that the placement of the cameras themselves exhibits a discriminatory pattern because more fines are levied in Black neighborhoods.

I think this illustrates a problem with the simplistic gap-based analysis of social problems. You could imagine a scenario in which there was greater camera density in white neighborhoods, and people were complaining that white pedestrians receive a greater level of protection from reckless driving. Indeed, as another Washington Post article (this one by Ian Duncan) explained, the toll of America’s rising number of pedestrian traffic deaths during the pandemic fell predominantly on Black and Latino Americans. This pattern existed before the pandemic (CDC report; academic study by Cara Hamann, Corinne Peek-Asa, and Brandon Butcher) and shows no sign of letting up.

I think the most relevant asymmetry of this situation is between the people getting caught on speed cameras in Black neighborhoods, who are breaking the law and speeding, and the people getting killed by reckless drivers, who aren’t doing anything wrong. To the extent that we care about disparities, we should be focused on the carnage afflicting innocent people.

The risk of pedestrian death is highly sensitive to vehicle speed.1 The only way to make urban areas safe for pedestrians is to force drivers to drive at speeds that strike them as un-intuitively slow, and the only way to do that is with rigorous speeding enforcement. And cameras are a much better way of accomplishing that than flooding these zones with cops doing traffic stops.

The penalties are too harsh

What I said about murder on Monday is even more true of the lesser crime of speeding; one benefit of increased enforcement is that you can make the penalties less harsh.

When it’s easy to get away with driving too fast, the penalty for speeding needs to be quite harsh to generate deterrence. But high fines can interact with unrelated financial precarity to generate devastating consequences for relatively minor legal violations. If you catch a much higher share of speeders, you can generate equivalent deterrence with much lower fines and save harsher penalties for repeat violators.

And here’s where the fiscalization issues that the Times flagged are relevant. In a well-functioning enforcement system, fines should bring in very little revenue because the rules are both reasonable and well-enforced, so people don’t break them. You optimize for revenue by doing the reverse: lulling people into a false sense of security with spotty enforcement, then hammering them while selectively exempting people with political clout from meaningful penalties.

All that really is bad. But the solution isn’t to say that traffic enforcement is a sham, it’s to try to actually fix it.


On highways, I think the question of the appropriate speed is a little less clear, but a high share of highway fatalities involves someone breaking the speed limit.