Every year, about 40,000 Americans die in car crashes. As a friend of mine sometimes says, that's about as many people as are killed by guns, except that guns are designed to kill people.
The National Highway Transportation Safety Administration famously reported that 94% of crashes are due to human error: speeding, not paying attention, changing lanes without signaling, and the like. This is part of the reason that some advocates reject the use of the term "car accident" and prefer the term "traffic violence." If you're scrolling Instagram while doing 70 down I-405 (something I have actually seen!) then is it really an "accident" when you rear-end someone into a coma?
That 94% figure has come under scrutiny. Critics point out driver behavior is dictated, in part, by the structure and arrangement of the roadways, and that it's actually the job of traffic engineers to design roads that encourage safer behavior. There is some truth to this, too. Better engineering choices do, in fact, generate safer outcomes. Road damage causes car crashes, and better maintenance reduces them. Speed kills, but the presence of a median, an increased number of access points, and narrow shoulders are all associated with reduced speed on highways. Roundabouts are safer than traffic lights and stop signs.
It's true that drivers exist in the context of a broader system, but policymakers have few realistic levers with which to make meaningful changes to this system. America's infrastructure is already infamously decrepit. Is a wholesale rejiggering of our roadways really in the cards? Autonomous vehicles will probably revolutionize safety, but it's anyone's guess when that long-delayed future will arrive.
Should we punish bad drivers more severely?
Another context surrounding driver behavior is, of course, the law. It is illegal to drink and drive, illegal to text and drive, illegal to change lanes without signaling, illegal to speed, and, for the love of God, illegal to surf the internet and drive. Unfortunately, that hardly stops anyone from doing these things.
Why not? One theory says that people persist in illegal behaviors because the punishments for these behaviors are not sufficiently harsh. In the rational choice theory of crime formulated by the economist (and Nobel Prize winner) Gary Becker, increased punishment severity is one lever that society has to make crime less "worthwhile" for potential criminals. Even if capture is unlikely, the risk of extreme punishment should you actually be caught, says this theory, will deter you. If, as I sometimes fantasize about, we started guillotining people for taking selfies while driving, would such behaviors cease?
The evidence on punishment severity as an instrument of deterrence is mixed. A 2017 literature review found that punishment severity has a relatively small—though not totally inconsequential—effect on crime. In the traffic setting, the evidence is somewhat stronger. A series of randomized experiments in the US and Israel estimated an elasticity of approximately -0.2 between red light running and the fine levied for such an offense. In other words, a 10% increase in the fine results in a 2% decrease in the incidence of red light running.
In other settings, there's also evidence for what's known as specific deterrence — individuals who are caught and punished for speeding, for instance, may be less likely to do it again, even though there may be no effect on speeders who aren't caught. In a natural experiment analyzed in Florida, economists found that speeders who received a larger fine for speeding were about 20% less likely to receive another speeding ticket in the following year than speeders who received a more lenient one.
There are a few obvious problems with a model of enforcement that relies on punishment severity to deter crime. The most obvious is that flat fines — e.g. $500 for speeding — are regressive: their burden falls most severely on the poor. This also means that they have accordingly less deterrent capacity for richer drivers. So one failure mode of severity-based deterrence is that even if we succeed in raising speeding tickets to $10,000 a pop, we might end up in a world with fewer car crashes but where fry cooks spend years paying off traffic ticket debt but assholes in Lamborghinis still kill people at 100mph.
One way of getting around this issue is to implement income-graduated penalties or "day fines": richer offenders should pay more! In addition to not bankrupting poorer speeders, this approach also achieves similar levels of deterrence for drivers of different levels of wealth. Sadly, it is probably a nonstarter in the U.S.
Enforcement works
Luckily, there's another side to deterrence theory: offenders seem to respond robustly to an increase in the probability of being caught. This is true in general, but seems particularly true in the case of traffic crimes:
The same US-Israeli set of experiments that found the -0.2 elasticity for fine severity found a roughly similar result for number of tickets issued: a roughly 10% increase in the probability of being caught (as a result of the installation of red light cameras) caused between a 1.5% and a 2.2% reduction in the number of violations issued.
A 2015 study analyzing the effects of a temporary enforcement campaign on car crashes in Massachussetts found an elasticity of -0.28: a 10% increase in the number of tickets issued led to a 2.8% drop in car crashes.
A 2013 study analyzing the effects of mass police layoffs in Oregon found an elasticity of between -0.33 and -0.38 of police presence and traffic fatalities: a 10% decrease in the number of highway troopers per vehicle mile traveled caused a roughly 3.5% increase in the number of road deaths.
The results of a 2011 study suggest an elasticity of roughly -0.25: a 10% increase in the number of tickets issued led to a roughly 2.5% drop in the number of car crashes.1
The big takeaway from all this is, very generally, that more traffic enforcement—in the form of more cops doing more traffic stops—effectively reduces traffic accidents and traffic deaths. The Oregon study puts a finer point on this, estimating that a highway fatality can be avoided with an additional roughly $300k of expenditure on the police.
This also puts a finer point on one of the issues with this strategy. Police are expensive! Moreover, it's not at all clear that they are the right choice for traffic enforcement. To start with, there are lots of other things police could be doing: breaking up fights, solving murders, and simply being visible.
This last item is important, since the evidence about police presence is clear. Having more police results in less crime. The general consensus abouts these results is that the mechanism at work is not incapacitation, but deterrence: it is a visible police presence that stops potential criminals from doing crime. But officers sitting behind the wheels of their cruisers writing out a traffic ticket by hand (for some reason!) are not spending that time walking Main Street, or even just driving around. In other words, deterrence of traffic crime by police might come at the cost of deterrence of other crimes. Police spend more than half of their time doing traffic stops (thanks to Darrell Owens for this stat)!
It's not clear why police are doing this work in the first place — or even why traffic stops are necessary at all. After observing a driver running a red light, there is no particular reason a police officer needs to wail their siren, chase down that driver, and then ask them if "you know why I pulled you over." Citations generally rely on what police officers have observed; there's no reason they shouldn't be able to just write down the violation and have the ticket mailed.
And, if we're going to do that, there's really no reason that it has to be an armed officer of the law doing this work. Many cities already have trained traffic police; why not just hand them a radar gun? In fact, why have a person do it at all?
Now there's an idea. The use of red-light cameras and speed cameras is known as automated traffic enforcement, and the evidence around them is pretty good.
The installation of fixed speed cameras in the UK saved an estimated 190 lives per year, though the effects of the cameras were highly localized around their positions.
A similar program in Belgium reduced serious and fatal accidents by 29%.
A 10% increase in the coverage of Safety Tutor automatic enforcement program in Italy resulted in a 4% reduction in total accidents.
Systematic reviews of the evidence find the same thing: a Cochrane review concluded that "speed cameras are a worthwhile intervention for reducing the number of road traffic injuries and deaths," while another meta-analysis found that the installation of red light cameras resulted in a 12% reduction in crashes. The nonpartisan Congressional Research Service agrees.
The threat of government overreach
Civil libertarians are famously—and reasonably—worried about the implications of increased automated surveillance. The US Code is 22 million words, and state and local codes add millions more on top. There are so many laws and, as libertarians are very quick to point out, simply existing often means breaking them. Traffic laws range from the very important (don't drive 100mph!) to the reasonable but mostly supid (don't have something hanging from your windshield). They are largely mundane and also, one imagines, the most commonly broken.
So expanding surveillance of the roads could simply mean giving the government an excuse to manhandle or imprison you where they'd previously have none.
This is a reasonable concern, but in the case of automated enforcement, I think it has things exactly backwards. The status quo is one in which the government already has this power, but in which it is implemented capriciously and unpredictably. The power to stop or harass you for a traffic crime is vested in an armed agent of the state who can shoot you if you disobey them. Traffic stops are infamously a pretext for various other kinds of activity. Unlike a policeman, a camera cannot search your car or beat you with near impunity.
Moreover, automated enforcement allows the introduction of a system of checks and balances that simply don't exist with human enforcement agents. Police patrol routes aren't public data, but the locations of speed cameras are. If issuance of citations is automatic, then there's no fallible human deciding to let certain people off or invent reasons to arrest others. It also makes it much easier to challenge a citation: there's photographic evidence in every case.
Napkin calculations
So how worthwhile could this be? New York city currently has 2000 speed cameras that have just started operating 24/7 for the first time. Suppose we want to double this number — each camera costs about $100,000, so that would mean a 100% increase in the number of cameras at a cost of $200 million. That's a lot of money!
Still - there were about 37,000 crashes in NYC in 2021, so if we use the elasticity reported for the Italian Safety Tutor program reported above, we should expect a 40% decrease in the number of crashes — to save about 15,000 crashes. Since there's about 1 fatality for every 140 accidents, that suggests we can save about a hundred lives this way. The cost of a car crash is estimated—at the low end—at roughly $10,000 on average, suggesting a total benefit of about $150 million — in the first year alone. And the Italian estimate may be optimistic, but even if the true effect is only one-quarter the reported size, we'd still expect the increased number of speed cameras to recoup their cost after 5 years — and 500 lives saved.
This study didn’t report its results in the form of an elasticity; I backed this figure out from the provided data.
Fascinating blog post. I work in road safety and would love to get in touch. Got a twitter handle?
"This is a reasonable concern, but in the case of automated enforcement, I think it has things exactly backwards. The status quo is one in which the government already has this power, but in which it is implemented capriciously and unpredictably."
Still better than all the time, for everybody, and strickly.