Laura Marsh and Alex Pareene write:
In 2020, American auto traffic was down 20 percent . . . That significant reduction in driving, however, has had an odd side effect: It made an already dangerous activity even deadlier. The National Safety Council estimated that 2020 saw the biggest single-year spike in traffic deaths in a century. Thirty-six thousand, six hundred eighty people died on the roads in 2020. 2021 was somehow even worse.
That’s amazing. Not just more deaths per mile traveled, but more deaths in total. Why? Marsh and Pareene continue:
Journalists have sought to explain this by turning to cognitive scientists for neurological answers and blaming drugs and alcohol. The pandemic is said to have made us more reckless, more anxious, more stressed, and more likely to make bad decisions behind the wheel. But can pandemic stress explain why other countries haven’t seen the same skyrocketing numbers of traffic fatalities?
They interview Charles Marohn, who says:
This is a pathology exclusive to Americans. What we saw in European countries is that as driving went down, crashes went down; as driving went down, fatalities went down; there was not the correlation we saw in the U.S. By the way, those places have lower rates of death per mile traveled than we do. We have very dangerous roadways, we have very dangerous streets. . . .
Our roadways are designed based on the precepts that we developed and gave to engineers at the end of World War II, which are all around mobility. If we can move a high volume of traffic at a high rate of speed, we can have increased mobility, and increased mobility—particularly in the postwar era, the couple of decades right after World War II—was tied to things like economic growth. . . . When you focus on speed and volume, what you do, from an engineering standpoint, in order to attain that, is you create buffers. You get higher speeds by widening out lanes, adding shoulders to the sides of roads, by removing things from the edge of the roadway that would create visual friction and slow things down. You basically create a highway. In America we took the knowledge that we had gained from building safe highways and applied that to our local streets. On a local street you don’t have the simplicity that you have on a highway. You will have cars that randomly switch lanes, that turn in and out of traffic, that pull out of driveways—and that’s just the cars. You also have people walking, people in wheelchairs, kids that chase the soccer ball. The challenge with our streets is that when you combine speeds that are above 20 miles an hour with random complexity, you get a high degree of collisions, traumatic injury, and death.
OK, that’s the background. Pareene then serves up the question:
So the existing state of U.S. streets was already very dangerous prior to the pandemic, but then we took a bunch of cars off the road and they somehow got worse—and you have a theory as to why.
And Marohn replies:
Prior to the pandemic, the one ubiquitous condition that we experienced across North America is high levels of traffic congestion. What congestion does from a traffic standpoint is it slows everything down. It calms traffic. You can’t drive fast because there’s someone in front of you, and because you’re limited to how fast you can drive, all of that overengineering and overdesign to get you someplace quickly goes away. Well, now get to March 2020 and remove the congestion, and what happens? . . . They’re essentially driving in the most dangerous conditions, which is a medium volume of traffic, at high speeds, with lots of complexity added in. . . .
Marsh says:
This gets back to that red herring about people driving recklessly because they’re angry. When you have a lot of congestion people actually are angry, but they don’t have the ability to drive recklessly because they can’t really move. When you’re quite happily zipping along—that’s when you may be at the highest likelihood of causing a really bad crash.
Marohn:
Congestion is frustrating, but there’s no outlet for it—that’s why it becomes very frustrating. One of the things that’s been reported is that a lot of these crashes are people who have been smoking marijuana or drinking. The reality is that those people exist in the traffic stream today—they just can’t go very fast. There’s a limit to the mayhem they can create in a system that is really overcongested, but remove that congestion—give them lots of buffer room, lots of room to move—and they’re going to wreak havoc.
The real story and the journalistic narrative
One thing that Marsh and Pareene get into in their article is the difference between what is actually happening—a lack of congestion allowing people to drive too fast and then crashing their cars—and a much different narrative that’s been appearing in the news.
Consider this bit from the New York Times a couple days ago:
Social distancing leads to the isolation and disruption that have fed so many problems over the past two years — mental health troubles, elevated blood pressure, drug overdoses, violent crime, vehicle crashes and more.
In a separate post, Marsh takes apart the claim that “Social distancing leads to the isolation and disruption that have fed so many problems over the past two years” such as “vehicle crashes.”
It turns out that the link from above paragraph goes to another NYT article which attributed the higher death rate, not just to emptier roads allowing higher speeds, but to “two years of isolation and disruption.” For evidence on this claim, we’re referred to another NYT article, “Pedestrian Deaths Spike in U.S. as Reckless Driving Surges,” that is subtitled, “Fatalities are climbing to record levels two years into the pandemic. Authorities cite drivers’ anxiety levels, larger vehicles and fraying social norms.” From that article:
Going into the pandemic, some traffic specialists were optimistic that pedestrian deaths would decline. . . . The opposite happened. Empty roads allowed some to drive much faster than before. Some police chiefs eased enforcement, wary of face-to-face contact. For reasons that psychologists and transit safety experts are just beginning to explain, drivers also seemed to get angrier.
The very first reason given—Empty roads allowed some to drive much faster than before—is exactly what Marohn said above. So it’s not like the Times is ignoring the speed thing or trying to hide it. But, then, as the article continues the issue of speed does not return, nor does anyone mention the safety benefits of congestion. After all, as Marohn notes, congestion causes frustration, but it’s in the uncongested, less frustrating setting, that people are free to drive faster and then crash their cars and run people over.
Just to be clear: these other issues such as big dangerous cars, lack of enforcement, a pedestrians-don’t-matter culture, etc. . . . sure, I can believe these make a difference. And I can’t blame it when a psychiatrist and a psychologist are interviewed for a news article and they give psychological explanations. That’s their job, right?
But, all that said, I think Marsh has a good point that this is a case where a conventional narrative—in this case, about the harms of social isolation—is driving the story in a misleading way. It’s not that this Times article is wrong that isolation and disruption have fed the problem of vehicle crashes. The term “fed” is vague enough that it could represent a very small influence, and we don’t have any evidence that isolation and disruption haven’t had this effect. But there’s no evidence or strong argument that they have, either—all we have are speculations by a psychiatrist and a psychologist, without an acknowledgment that faster, freer driving could be enough alone to cause these traffic accidents, without need for these psychological explanations.
The problem here is not anything specific to those NYT articles; it’s that a conventional narrative can take over, it can be so much in the air that it’s hard for people to see anything else
A message that people don’t want to hear?
Another thing, perhaps, is that the counter-story told by Marsh, Pareene, and Marohn—that our roads are inherently dangerous, and congestion has been a major force in keeping the accident rate down—is not a message that many people want to hear! What we want to hear is that congestion is bad, frustration is dangerous, and that we can have it all—smooth traffic flow, calmness, and safety. But what if happy, smooth driving leads to more dangerous roads? What if the way to safety is to make driving less pleasant, less like a car commercial? That would be annoying, no?
I’m not saying that the NYT authors focused on “isolation and disruption” as a way to avoid thinking about tough choices in road safety policy. I just think that the conventional story avoids some potentially upsetting ideas, and that’s one reason why the story can continue to stay afloat. I appreciate the work of journalists such as Marsh and Pareene in pushing back against this.
I will have to go back and read the arguments but I believe the phenomenon has been recognized as Smeed’s Law.
Jk:
Not quite, I think. From the linked wikipedia article:
In this case, the decrease in volume and increase in speed corresponded to an increase in fatalities per capita, which is the opposite of what Smeed’s law would say, right?
What’s funny is that I don’t see the post mention how much was the spike compared to what we see in a usual year of the typical year to year variation.
Rahul:
There’s a graph of those data in one of the linked NYT articles.
I posted about this in a thread downstarirs, but seems appropriate here re poorly suppported narratives related to the pandemic:
https://www.theatlantic.com/ideas/archive/2022/01/university-covid-policies-remote-learning-mental-health/621159/
Where Oster engages in some pretty loose speculation about the impact of remote learning on the mental health of college students.
In a related Twitter thread, I thought this was an interesting discussion with reference to considering how to analyze for causality and how Oster treats causality within the narrative she presented.
https://twitter.com/Joey__Schafer/status/1478988462879494144?t=E0HlAC9YIMXJX3XKe9pYFA&s=01
Mahron invented the word “stroad” for roadways that result from the design model he identifies – urban streets configured to move traffic at high speed as if they were were highways. You can watch a good video that illustrates the design and the problem it creates here:
https://www.youtube.com/watch?v=ORzNZUeUHAM
It appears that there is factual evidence that fatalities (in total, and in the rate) are up and traffic volumes are down. It also appears that there is evidence regarding speeds increasing (I haven’t looked at such evidence but I know some such data is collected and I am willing to assume it shows average speeds have increased). So, there appears to be agreement on these points form both the journalists and the psychologists. But the psychologists then go on to attribute at least part of the increase to increased anxiety and anger. Do they have evidence that these have increased – and, that these increases are contributing to decreased road safety?
I doubt that they have this evidence. At best they have the sort of “evidence” we often bemoan – small contrived experiments that show that things like anxiety lead to poor decisions, which can then be claimed to “show” that increased pandemic anxiety is causing more road fatalities.
But I wonder if the psychologists’ leap here is really so bad. After all, their expertise in psychology probably gives them plenty of latitude to speculate on these connections. And, real “evidence” of this connection is not likely to be collected. So, is their speculation harmful? Is it helpful? The only thing I really find disturbing is the lack of clarity for claims that are based on data and those that are based on professional speculation. They do have expertise and I think it permits them to speculate – and for that to be worth reporting. I would only ask that they be clearer about what type of evidence we have for what types of claims. Other than that, I don’t really find this to be something to criticize.
pretty sure it’s unambiguous that airplane passenger rage is WAT up.
Pardon my ignorance, but what does “WAT” stand for? (Can’t find it in a web search– keep getting “WAP” instead.
WAY except fat fingers and a cracked phone screen
I had the same question! Could be some new internet slang, but I think it is just a typo (Y and T are adjacent on the keyboard).
Dale,
Yes, I agree that speculation is fine. As a political scientist, I’ll speculate about politics, and it makes sense that psychologists can be knowledgeable spectators about psychology.
So I don’t think it’s bad that the NYT articles talked about isolation and disruption. But I do think that Marsh has a point that those articles were over-focusing on this particular story without recognizing the importance of more mechanical factors such as street design and traffic speed.
How would an increase in road rage shootings fit into this?
https://www.economist.com/united-states/2021/12/04/as-americans-get-back-into-their-cars-road-rage-shootings-are-spiking
https://www.healthline.com/health-news/road-rage-has-risen-during-the-pandemic-how-to-stay-safe
A much higher proportion of the people on highways in eastern PA were driving much more aggressively (at whatever speeds) during the months in which traffic levels were much reduced. The danger on the highways was greatly increased by berserk behavior, not merely by higher speeds. I never felt in danger driving on German autobahns, but driving here during the height of the pandemic was white knuckle stuff.
Smintheus:
I see your point. Marsh and Pareene discuss it a bit in their article: what they say is that the biggest risk is not on freeways but on mixed-use surface streets. Also, regarding reckless driving: lack of congestion can contribute to that too: when the streets are less crowded, this reduces drivers’ motivations to be careful.
When I look at traffic fatalities over the years, I see a dramatic decline over time.
https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in_U.S._by_year
Perhaps we have reached some asymptote, and the current increases are just statistical chatter around a rate that is remarkably low compared to the historical record.
Oncodoc:
I wouldn’t quite call it just statistical chatter, but I agree that long-term trends are important. I guess one reason why the topic of increased traffic deaths gets attention is that it is, to a large part, avoidable.
I wonder if there might be data related to injuries and deaths per accident. I suppose it would be possible that there is interaction between trends. Maybe the injuries and deaths per accident has increased due to higher speeds somewhat independently of the number of accidents (and the associated change in the amount of trafific) – and the interaction between speed and number of accidents and number of cars on the road and number of casualities is complex.
LOTS of data here: https://www-fars.nhtsa.dot.gov/Main/index.aspx
Just one bit, after a brief look at the data: From 2019 to 2020 (annual totals), fatal crashes increased by less than 4%, but fatal crashes involving speeding increased by around 15%.
> (Pareen) but then we took a bunch of cars off the road and they somehow got worse—and you have a theory as to why.
> (This post) One thing that Marsh and Pareene get into in their article is the difference between what is actually happening—a lack of congestion allowing people to drive too fast and then crashing their cars
Maybe a bit nitpicky but it seems like the congestion is set up as a theory also. I guess the most meaningful difference from the NYTimes articles (which I didn’t read) is probably in how you’d try to measure the variables concerned. Dale points to a bunch of road data — so maybe this is easier to get at than the psychological variables behind the NYTimes theory (certainly seems like it would be to me)?
Also after I read this I walked out of my suburb and crossed a busy throughway and a car squealed its tires a bit to stop. I recall I had the cross light, but I could be mistaken. I suspect this person might have been driving in a bike lane to take advantage of right on red. Who knows. The car wasn’t really that close to hitting me — I think they must have just been zoned out and slammed the breaks, but mistakes!
I like “casualities” in a comment on causality of casualties, at a blog about causal inference. Could belong in the Handy Statistical Lexicon?
Heh. Lest anyone think my typos aren’t intentional.
Imagine how hard I had to work to avoid Muphry’s Law in that comment :)
well I think it’s clear that the pandemic effected affect, which affected the effect of driving on pedestrian safety.
“in a comment on causality of casualties, at a blog about causal inference”
You missed a chance. You should have written “at a casual blog about causal inference.”
Could this instead be due to a decline in the performance of our EMS / medical services during this period? Our health care sector was clearly impacted by the events of the last 2 years which leads me to wonder if an explanation along these lines is more appropriate.
How about reduced vehicle and roadway maintenence? Or “defund the police” emboldened some risky drivers?
Im sure there are dozens of plausible post hoc explanations we can come up with.
So you just deep six information that doesn’t accord with your models?
Smintheus:
I don’t think that Marsh in her criticism was saying the NYT was throwing away any information; I think she was just saying that they were pushing their favored theory without good evidence or without clear reflection of the speculativeness of the theory. But I don’t see anyone deep sixing any information here!
Decreased traffic enforcement in 2020 (along with decreased traffic and increase in drug/alcohol usage) could be part of story and could explain the opposite trends in USA vs Western Europe in 2020 following same pattern of murders.
Here is a good article on traffic fatalities and decreased traffic enforcement in Connecticut along with other factors.
https://www.wshu.org/connecticut-news/2021-12-13/connecticuts-traffic-deaths-are-on-the-rise-since-last-year
This Vox article does a good job summarizing increase in murders.
https://www.vox.com/22578430/murder-crime-2020-2021-covid-19-pandemic
I can’t actually find EU homicide data for 2020 but Vox article states that they at least didn’t experience double digit increase as in USA.
One counter-argument is that many crimes fell during 2020, but there is a new paper that cites observational evidence that overall crime was actually up after accounting for decreased opportunities related to pandemic. This hypothesis was mentioned in the Vox article as well.
http://maximmassenkoff.com/papers/victimization_rate.pdf
Definitely:
Good point about enforcement. I guess the people interviewed in that traffic article might argue that this is increased evidence of how dangerous our streets are, if a high level of enforcement is needed to keep down the rate of accidents and deaths. From an engineering standpoint, you’d want a system that was safe by default and didn’t require continual police intervention.
Dont they have the data segmented by accidents in the city/town, accidents on highways etc? This will resolve the issue quite easily.
Anon:
According to the linked article, most of the fatalities are on surface streets, not freeways. But I haven’t looked at the data myself.
Interesting post. I always believed congestion was one reason why I saw fewer severe accidents on local indian roads, where traffic rules are less well-defined and followed less.
It’s worth keeping in mind that India’s traffic fatality per car is 10X that of the US (https://en.wikipedia.org/wiki/List_of_countries_by_traffic-related_death_rate). Congestion perhaps “helps,” but the current level of death/injury is very high.
Indian traffic is truly depressing. I’ve wondered if the most high-risk situations I’ve ever put my kids (and myself) in involve crossing roads in India, which often involves some measure of giving up on waiting for things to improve and blindly dashing across.
My scariest driving experience (I was not the driver) was on a highway in india between vijayawada and hyderabad. It was a narrow two-lane road with two-way traffic; the driver decided to pass a truck going uphill, when you could not see what was coming from the other side of the hill.
Inside vijayawada, on local roads, in a car, I never felt too unsafe even with all the chaos. I agree one has to be extremely careful when walking and crossing roads. Thanks for the Wikipedia link Raghu.
Political scientists are so precious with their models, never to be subjected to facts about how people actually behave.
Smintheus:
I don’t get what you’re saying here. All the discussion above, including the linked articles, my post, and the comments that follow, are all discussing different aspects of how people actually behave. It’s just that behavior is complicated, and lots is gong on at once. I appreciated the Marsh and Pareene article because they focused on something I hadn’t though about before, which was the idea of roads being so dangerous that traffic congestion was what was keeping things from being even more dangerous—but there are other things happening too, hence this discussion. I don’t see anyone here being “precious with their models,” indeed I don’t even know what that means.
Then why sit on one person’s comment for hours while posting other, later comments to the thread?
Smintheus:
We have a spam filter that catches some comments and puts them in different places, then every once in awhile I go thru the spam directories and release the non-spam comments for publications. Sometimes it takes awhile. This is something I do for free!
The podcast from Marsh and Pareene has guests doing the same sort of thing that Marsh accuses Leonhardt of doing on Twitter. It’s fine that she doesn’t find his argument convincing, but it’s kind of ridiculous that she accuses him of constructing his argument poorly, then goes on with Pareene to do no better.
Anyway, I think part of the problem here is that there’s not going to be any sort of direct proof one way or the other. With that in mind, I think Leonhardt is reasonably on point when he says that “The only plausible explanation for most of the recent surge is the pandemic,” even if the specific ways in which it did this are up for debate.
Or maybe it’s something to do what deaths versus injuries. Not all people that get into accidents die, of course. The coverage seems to focus on deaths rather than just injuries, and perhaps one reason the number of people that died went up due to some specific reasons (in addition to just bad luck). Perhaps one of those specific reasons is that hospitals were stressed due to COVID and thus couldn’t treat people as effectively. Of course, this is entirely speculative on my part.
It’s nice to see something in my wheelhouse. As I was telling my class last week, I like Marohn’s distinction between roads (designed to move traffic) and streets (designed to facilitate interactions). As an engineer, I do find him a tad harsh on the profession at times. Many of us now recognize that you can’t build your way out of congestion.
On the safety point, I agree that it’s unlikely we’ll reach a definitive conclusion on the relative role of behavior vs. road design for recent trends. A colleague brought up another interesting point. Drivers may have been different during the peak of lockdowns. If we had data on who was driving, we could get a better sense for how behavior affects accidents. Higher accident rates are found among non-Saudi nationals in Saudi Arabia vs Saudis (work by Fred Mannering, who is prominent safety researcher. He also did some interesting work on seat belts). Similar patterns are found between Europeans and Americans (Europeans assume they must yield at uncontrolled/no stop sign intersections vs. Americans think they have the right of way). I recently moved to the US and certainly notice a difference as a pedestrian. I have a project with my state DOT looking at ped/bike accidents in Nebraska and how road features affect it. I’m not holding my breath anything will change in design practices based on our results though.
> Americans think they have the right of way). I recently moved to the US and certainly notice a difference as a pedestrian.
The pattern of interaction between cars and pedestrians varies quite a bit across different regions in the US. E.G., in my experience, in California you can more sometimes expect cars to wait for pedestrians. If you tied that in Philly where I few up, you’d get flattened. Also, in my experience jaywalking is much less frequent in Cali (I’m assuming they give out tickets for jaywalking there where I’ve never even heard of that in Philly).
Also… I lived in Seoul a while back and it was interesting to see how even late at night when there was zero traffic on the road, pedestrians would wait at traffic signals and never cross the street before the light had changed. In Philly that would be considered insane behavior.
If it’s true that traffic congestion produces safer roads, you should be able to see that by noting the time of day when crashes occur. You can look at the ratio of number of crashes to number of vehicles on the road at that hour. If the ratio is lowest during rush hour, it will be pretty strong evidence for the hypothesis.
It would also be interesting to look at number of fatalities in rural counties. Has that number stayed flat during the pandemic? If so, it would also count as pretty strong evidence.
Kevin:
You can see some data here, but unfortunately that page only gives number of crashes of different types by time; it doesn’t give total traffic density as a denominator. Anyway, the peak times for fatal crashes are late at night and early Saturday and Sunday morning; that’s also when many drivers are tired and many are drunk.
Interesting numbers, and I think they do support the “congestion produces safety” hypothesis to some extent. Notice that there always seem to be relatively few fatal crashes during weekday morning rush hour. Also notice the relatively high number of fatal crashes on late Sunday afternoons.
The pattern is quite different when you look at the chart for nonfatal crashes. Now Sunday afternoon looks like one of the safest times. It’s possible that when traffic is congested, accidents are just as likely but they’re less likely to be fatal because the speed of collision is lower.
> but they’re less likely to be fatal because the speed of collision is lower.
https://statmodeling.stat.columbia.edu/2022/03/12/auto-accidents-and-isolation-and-disruption/#comment-2047315
Wow! A sad end to a road-hater’s fable.
This data obliterates the “congestion = safety” argument. Quick read: prior to the pandemic, death was strongly associated with periods of congestion and even more strongly associated with “party hours”. The total number of accidents is **very strongly** associated with congestion. Death is most common later in the day but accidents overall rise throughout the day and there is a peak in fatalities associated with the morning commute every weekday.
In my experience the afternoon commute is always longer and more congested. I’d say on average, even with the benefit of an express lane in the afternoon, my commute is 15-20% longer in the afternoon. I suspect this is due to rising commercial vehicle traffic and the general peak of daily human activity, which is probably sometime in the mid-afternoon to early evening.
As others have noted the fatality / nonfatality framing is likely misleading, as it could just as well be determined by the availability of treatment as by the severity of the accident, the former of which would be dramatically shifted during the pandemic. Also the relative proportion of high-risk drivers on the road could be dramatically altered by the pandemic, as older more experienced workers (low risk drivers) are more likely to be able to WFH, while younger less experienced workers (high-risk drivers) dashed to find new jobs. Another important factor could be the sudden emergence of millions of delivery vehicles. Delivery vehicles have low visibility, tight schedules, likely have less experienced drivers and commonly drivers don’t know the neighborhoods they’re operating in. They’re often parked willy-nilly along roadsides blocking visibility and darting the opposite direction of their blinkers as they look to find addresses.
This is a great story about how “Just-So” stories created with half-baked data “analysis” deliver unreliable answers. If you miss one or two critical dimensions in an n-dimension problem, you’ll come up with nonsense.
What are you talking about? The total fatal crashes peak on Saturday, Sunday, and Friday night. Particularly, midnight to 3 A.M. on Saturday and Sunday have the most fatal crashes and also the least cars.
If you constrain yourselves to working hours, only then can TOTAL fatal crashes be said to be associated with congestion–but then, total number of cars on the road also peak sharply during congestion, and the rise in non-fatal crashes during those hours is higher than the rise in fatal crashes. If you consider “given a randomly sampled traveling vehicle, how likely is it to be in a fatal crash”, it’d be lowest during congested hours.
Are you looking at some other link I’m not seeing?
“Are you looking at some other link I’m not seeing?”
Probably not. You just don’t get it. You didn’t read the explanation, so I’ll clarify for you:
Both fatal and non-fatal accidents peak at peak traffic hours (4-8p) in the afternoon/evening., except Friday and Saturday nights. There is also high point in fatal crashes during the morning commute, followed by a dip, then increasing into the afternoon and evening.
Evenings are higher risk for fatality than traffic levels would otherwise predict because the risk profile of the drivers changes from low-risk drivers (higher earning commuters and family people) to high-risk drivers (younger people, more likely engaging in drinking and drug use, and possibly criminal activity). The peak fatalities are on Friday and Saturday night because these are also peak times for young people partying.
There’s nothing here that suggests gridlock is protection from fatalities at all. Peak commute times have the most overall fatalities and the most total accidents. Friday and Saturday nights are risky, but only because the driving population changes substantially, from a population weighted to older, low-risk conservative drivers to a population weight to younger, higher risk, and more reckless drivers.
???
Did you miss somebody’s point about different numbers of cars on the road? Peak commute times might have higher total numbers of fatalities, but they also have more cars. Look at http://www.cmt4austin.org/Traffic_By_Time_of_Day.htm
During the 4-8pm peak, there’s up to twice as many cars at during noon to 4pm. Looking at monday on that graph, there’s 8% more accidents. This means the risk per car of getting into a fatal accident declined by something like 45%.
Here’s a question for you Anonymous. If you’re driving on a highway, it’s clear, no cars, you might feel safe cruising along at, say, 70 miles an hour. Suppose now, that every lane is filled as far as the eye can see, with 2 cars length between each car, but everyone is still moving at 70 miles an hour. Do you feel as safe now?
Here’s a second question for you. Suppose there is a new obscure highway, I-X, in the middle of nowhere. Every year, 3-5 people attempt to take that highway and all of them die. Is that highway safer or less safe than the I-80? By your logic, there are fewer total traffic fatalities on the I-X every year, but multiple on the I-80, so the I-X is clearly safer. Which is complete nonsense.
If there is a bridge that’s never been finished and is raised in two parts, zero people will take it so there will be zero total fatalities, so the data lead us to believe it’s the safest road ever constructed.
What part of your comment do you think I didn’t understand? All you’re doing is repeating the total numbers. I’m saying the relevant metric for road safety is fatalities per person-mile traveled. “Given that I drove a mile between these hours, what is the probability that died.” Not “what is the probability that there was an accident somewhere on the same toad as me.”
“Did you miss somebody’s point about different numbers of cars on the road? ”
??? So what? The question at hand has nothing to do with the number of cars on the road. It has to do with the composition of the driving public.
Here’s the question I’m answering: “Does congestion / gridlock provide protection against accidents / fatalities?”
Here’s the question you’re answering: “what are the odds of becoming a fatality at different times of the day?”
They aren’t the same question.
Zhou: what do you think would happen at 6pm on Tuesday if we filled the roads with 3am Sunday morning drivers, in the same condition they’re typically in driving on Sunday morning? Tons more accidents and fatalities, that’s what. It’s not the congestion that’s providing the protection at 6pm on Tuesday. It’s the fact that the composition of the driving public is far lower risk on Tuesday at 6pm than it is on Sunday at 3am.
If congestion were providing protection as you and somebody claim, then we should see a spike in fatalities on every night of the week from 8pm to 4am. We don’t see that, we see just the opposite: fatalities decline 8-12p and are virtually non-existent when roadways are wide open at 12-4a.
It’s a classic Simpson’s Paradox. The causal factor of accidents changes as the risk profile of the driving public changes throughout the day. As traffic builds through the day, the increasing volumes cause more and more accidents. But as traffic wanes in the evenings the risk profile of the drivers rises – especially on Friday and Saturday night. If those same risky drivers were on the road in the same proportions at peak traffic volumes, the accidents at that time would be much worse.
cheers folks!
I am not claiming that, I’m just disagreeing with your misinterpretation of the data as debunking the theory. It’s totally possible that your theory is true and the entire effect is differential composition of drivers. But the data does not disambiguate the two theories, and does not “obliterate” the congestion => safety theory. The data show that the safest times of the week in terms of accidents per person mile is during peak commuter hours—you claim it’s because the drivers are safer, it could also be because those hours are more congested. Both can be supported by these data, which is why your confidence makes no statistical sense.
Total decrease by a factor of 3 during those times relative to peak hours while number of cars decreases by a factor of 10 or more. So you do see a big spike. Against, this could just be more people driving sleepy and drunk, so your theory still holds.
If you accept that the relevant agent for protection is individual drivers, and not the road or cost to society write large.
> Zhou: what do you think would happen at 6pm on Tuesday if we filled the roads with 3am Sunday morning drivers, in the same condition they’re typically in driving on Sunday morning? Tons more accidents and fatalities, that’s what. It’s not the congestion that’s providing the protection at 6pm on Tuesday. It’s the fact that the composition of the driving public is far lower risk on Tuesday at 6pm than it is on Sunday at 3am.
This is specifically why I did not make the comparison with 3am drivers, but instead between peak commute drivers and early afternoon drivers during weekdays. Peak commute drivers are still almost twice as safe as drivers out between noon and 4pm. I don’t think you can as easily blame teenage drunk joyriders for accidents happening during this time period.
Zhou, Somebody:
Click on the “age of driver” link in the data Andrew linked to. You’ll see that the number of crashes is concentrated in the 24-34 age group. There’s no data on the number of fatalities, but my guess is that drivers age 16-44 are responsible for a very large proportion of fatal crashes – which is why they pay much much more for insurance. These are the drivers out on Friday and Saturday nights, but not out during the morning commute.
Of course it’s true that the fraction of risky drivers changes at different times of the day and different times of the week. As far as I can tell, no one has denied that. It’s also true that some of the very same drivers may get riskier at times. Thursday morning’s sober and cautious driver may be Saturday night’s reckless drunk. There are a lot of factors at play.
Now, is “congestion reduces fatal accidents” one of those factors? You say it isn’t, but your argument still sounds pretty weak to me. The key premise you rely on seems to be the following: “If congestion were providing protection as you and somebody claim, then we should see a spike in fatalities on every night of the week from 8pm to 4am.” But that’s not necessarily true. The protection is relative, not absolute. The lack of congestion late on weeknights might still produce many more accidents than would have otherwise occurred, without producing more accidents than during evening rush hour.
Furthermore, you can actually see there are more late-weeknight fatalities than morning rush-hour fatalities. That looks to me like strongly suggestive evidence. Of course, you can always tell a story about how other factors are responsible instead. And for all I know, some other story is true. “Strongly suggestive” is as far as I’ll go; all in all, you seem to feel much more certainty on the subject than I do.
Schematically speaking, you could imagine a graph showing number of traffic fatalities versus number of cars on the road, all other things being equal. The curve would surely start at zero: If there are no cars, there will be no accidents. My expectation is that the curve will then rise linearly, or maybe quadratically. Then it will reach some peak, and start to decline. It’s on the far side of the peak where congestion provides protection. When there’s total gridlock, there won’t be
fatal accidents because cars won’t be able to move at all.
Where is the peak, then? My tentative answer would be that it’s the level of traffic you get an hour or so after the evening rush winds down. Of course, since other factors matter, there’s not a single absolute peak you can look for. But it’s plausible that rush-hour traffic is always on the far side of the peak, while traffic late at night is always on the near side. An even more tentative conjecture is that most vehicle-miles are travelled in circumstances on the far side, where fatalities decrease with the presence of more cars on the road. That would certainly be consistent with the spike in fatalities we’ve seen during the pandemic.
Kevin Nelson:
It’s odd to me that people are suggesting that, when risky drivers have more cars to smash into, the risk of fatality or serious injury magically declines! That makes absolutely no sense.
The claim that congestion prevents people from driving fast and thus reduces fatalities is rejected by a simple comparison of the morning and evening commute, which have about the same traffic levels but dramatically different fatality levels. There is simply no evidence in this data that congestion provides any kind of protection.
The total fatalities likely increase as a function of the ratio between aggregate driver risk and the number of cars on the road. During the day aggregate driver risk rises steadily as more and more 20-somethings take to the road, and peaks when the number of both 20-somethings and the number of cars peak. On weekend nights the number of 20-somethings on the road skyrockets, causing more fatalities.
Thanks folks!
So I downloaded the dataset of drivers involved in fatal car crashes during 2019 from the NHTSA’s FARS reporting system and plotted a stacked area chart of count of fatal drivers by hour of week, segmenting out under 30s
https://imgur.com/a/auYQxA0
Needless to say, while under 30s are of course overrepresented as a fraction of fatal car crashes, the distribution does not shift temporally in anything close to the way you’re claiming.
If you’re going to write things like this
it would help to consult any data at all on the age distribution of drivers and fatal car crashes. Again, the point of all this isn’t to advance any particular theory or agenda as conclusive–many things change, all at once. Is it alcohol, is it drugs, is it police enforcement, average speed, age distribution, darkness and visibility? All of those things probably influence a non-zero number of fatal car crashes, each of those theories is on the margin, true, including yours. But it takes a lot more than 5 minutes looking at some absolute numbers to pick out one particular variable as being responsible for all of a particular trend, and again, these data do not in any way “obliterate” the congestion theory.
And the theory really isn’t all that outrageous. If I’m stuck in 15 mile an hour gridlock, I could close my eyes and spin the wheel at random and it’d be pretty surprising if I died, while if I did the same at 55 I’d be lucky to survive. That’s just how collisions work.
I just can’t get over your complete unwavering confidence. It led me to believe maybe you were speaking from some kind of experience, but it turns out it was just guesswork. Clearly though, you must have known you were making things up — why do that? What do you get out of it? The satisfaction of wasting an hour of my time?
“And the theory really isn’t all that outrageous. If I’m stuck in 15 mile an hour gridlock, I could close my eyes and spin the wheel at random and it’d be pretty surprising if I died, while if I did the same at 55 I’d be lucky to survive. That’s just how collisions work.”
Your description of rush hour traffic is **wildly** inaccurate. Contrary to the parking-lot safety you describe, rush hour creates unusually dangerous conditions – conditions in which impulsive high-risk drivers can wreak havoc.
On our interstates during rush hour, it’s extremely common for traffic to go from full speed limit to full stop in half a mile – and extremely common to see accidents where that happens. There are numerous situations this: at the interchange from one Interstate to another, the leftmost westbound lanes are wide open while the right lanes, which access the ramp to the northbound interstate, are at full stop. Cars are constantly trying to dodge in at the front of the line or get out of the back up and go the other way. We also have carpool and pay lanes, which can move at dramatically different speeds than the regular lanes, and people dart in and out of those all the time.
The problem with speed variability at rush hour is so acute that the state installed automated speed limit reductions in *many* places to slow down the free-flowing traffic, but even these still often leave a substantial difference between the speed of adjacent lanes.
“So I downloaded the dataset of drivers involved in fatal car crashes during 2019 from the NHTSA’s FARS reporting system and plotted a stacked area chart of count of fatal drivers by hour of week, segmenting out under 30s”
Great! What does “drivers involved” mean? You need the age of the driver that **caused** the crash. And please remember that I indicated “higher risk **and** younger drivers”. I’m claiming that the aggregate risk increases – and that younger drivers are *one* component of this aggregate risk.
And, no, I didn’t just make all that up. As with your inaccurate imagined idea that rush hour is just creeping gridlock, your lack of knowledge from the real world is preventing you from understanding what the data mean. You must know that bars are busy in the evenings and not in the mornings; that restaurants are much busier in the evenings than in the mornings. And you must have at least some idea that full-timers at Google and other 9-5ers on average likely have better driving records than the waitstaff and kitchen staff at Red Robin, regardless of age. This information is just as relevant as what’s in the spreadsheet, and it’s more than reasonable to infer from it that the average risk rating of drivers rises in the afternoon and evenings.
Last but not least, you haven’t countered my observation that fatalities are substantially higher (25%-40% higher!!!) in the afternoon commute than in the morning commute, even though traffic levels are about the same.
So your basic argument for the “gridlock” theory is that
1) cars are always stopped or creeping at rush hour.
2) outside rush hour people can drive at high speeds because traffic volumes are light, and smash into other unsuspecting people.
I’ve shown you that both arguments are wrong. In fact, as I contended above, rush hour creates *highly* dangerous situations, where a single risky driver can cause very serious crashes involving a much higher number of vehicles. “Congestion” is so dangerous that the state has been forced to respond with new traffic controls. I’ve also provided a strong rational from common experience that it’s very likely that the risk rating of drivers changes as people’s activities change throughout the day, and that higher risk drivers are more likely to be on the road later in the day, whether that’s age-related or not.
As far as I’m concerned, you have a lot of work to do to show that “gridlock” or “congestion” confers any significant measure of protection. The fact that it covaries with other factors doesn’t make it responsible for the variation in fatality risk.
FARS does not assign blame, because that’s complicated. But here is a plot of the total count of drivers in single-vehicle fatal crashes, with under 30s in red and over 30s in blue. Green represents 9 A.M., yellow represents 5 P.M.
https://imgur.com/a/LRHgDm2
Still, the two counts peak in unison, and there are many more over 30s than under 30s. There simply are not enough under 30s to drive the whole trend. Maybe all those over 30s single-vehicle crashes were swerving to avoid some young hothead–but I doubt it.
Here’s the share of single-vehicle fatalities by under 30 drivers vs over 30s. Now we’re getting somewhere
https://imgur.com/a/C14XMCZ
It looks like the under 30s share of total accident peaks in the early morning and late at night, after commuting hours. This is indeed contemporaneous with when the roads are most dangerous on a per-vehicle basis. But it’s also when the roads are the fastest and when visibility is the lowest. Lots of things changing at the same time here–it’s hard to actually say. What’s definitely not true is that “risk peaks when the number of both 20-somethings and the number of cars peak”, since the share of 20-somethings clearly peaks after the total number of cars peaks.
Here is what you actually said
You made no mention of any other factor in aggregate driver risk. You strongly implied that aggregate driver risk rises through the day primarily because of the number of 20 somethings on the road, erroneously stated that risk, total cars, and share of 20-somethings peak contemporaneously, and explicitly stated that the number of 20 somethings causes the rise in weekend fatalities. But there’s a huge rise in the number of over 30s causing fatal car crashes on the weekend and weekday evenings as well. That’s just wrong.
I know all those things. But you made specific claims that the temporal distribution of accidents is driven by share of under 30s that are on the road, and I am saying that that doesn’t make any sense because there just aren’t enough under 30s causing accidents to do that. You made specific statements about how the age distribution of drivers changes throughout the day and what that causes, facts that are false and originated from your brain. Yes, it was based on a general intuition, but the confidence was misplaced and misleading.
Now you’re retreating into the much broader claim, that the periodicities are driven by changes in driver quality. I’m sure that’s true, but it’s such a tame, vague claim that it’s also completely useless. Even so, I have to add the caveat that I’m sure other things like visibility and differential rates of traffic enforcement also play a part, as well as within-driver changes in driver quality (a day worker who’s low risk during the day might be a worse driver after drinking at night). Lots of things are at play here.
No I don’t. Because I’m not pushing that theory. I’m simply stating that your confidence in your pet theory is misplaced and lacks empirical support, and that the other theory is not completely wacko.
And yes, rush hour traffic creates hazards. More cars are more obstacles. It’s also true that people slow down to protect themselves when they see more obstacles and go faster when they see open road. The reason you get traffic at all is because it’s safer than the same number of cars moving at regular speed. And a common reason for gridlock is because people have to slow down a lot to safely merge around some kind of road hazard, be it a lane closure or a previous accident.
The contention is whether or not the degree to which people slow down is an “overreaction” from the perspective of maintaining a fixed level of risk, where the risk reduction from the act of slowing down exceeds the risk increase it’s in response to, and whether or not that explains the jump in pandemic fatalities. I don’t know and I don’t have an opinion, I was just pointing out that your interpretation of the data as debunking this theory was you falling for the base rate fallacy.
To reiterate, in case you get lost again:
1. Your claims about the change in the age distribution of drivers over time are false, and your theory that said change is the leading driver of periodicity in traffic fatalities is clearly falsified by data
2. I’m sure that your amended theory has some truth in it. Changes in driver quality and driving quality are probably part of the periodicity in traffic fatalities, along with many other things
3. I am not pushing the congestion causes safety theory, and I certainly don’t think it’s the only thing that matters, but I don’t think it’s ridiculous either
A tangential comment: I’ve found I can save myself a lot of time by skipping past newspaper stories that use the words “spike” or “skyrocket” in the headline. I’m sure I miss some useful stories, but those words tend to be associated with over-interpreted claims from spotty data. That was my impression, at least, before I stopped reading them.
+1
I’ve crunched a lot of data on the surge in traffic fatalities since the first half of 2020 to write a couple of columns about it. Last year, I found an astonishing number:
“… motor vehicle fatalities among blacks similarly soared 36 percent in June–December 2020 [i.e., the seven months following George Floyd’s death] versus the same period in 2019, compared with a 9 percent increase among the rest of the population.”
In recent years, there has been a pretty strong correlation from one year to the next between traffic fatalities and homicides. The general trend for both was down in this century, but then both went up in 2015-16 (for homicides, this is known as the Ferguson Effect), then drifted down again, then shot up in the summer of 2020 and stayed high in 2021.
It appears that police stops of bad drivers dropped sharply in 2020 due to the two historic events: cops started social distancing due to covid in March 2020 and pulled fewer drivers over, and then the racial reckoning following George Floyd’s death on May 25, 2020 encouraged further police forbearance of bad driving, especially by black motorists.
Thus, traffic fatalities among blacks were 55% higher in June 2020 (the biggest month for the mostly peaceful protests) than in June 2019. The massive June 2020 increase is something of a fluke — June 2019 had unusually few black traffic deaths — but the general pattern of black traffic fatalities going up four times more than nonblack traffic fatalities according to federal NHTSA data for Jun-Dec 2020 over the same time period in 2019 is a spectacular phenomenon.
I graphed the NHTSA traffic fatalities by race data here:
https://www.takimag.com/article/the-racial-reckoning-on-the-roads/
My January 12, 2022 column reported on a sample of the first 30 states to publish year-end 2021 traffic fatality totals:
“Last week, I reported that my tabulation of year-end homicide counts from 44 of the 50 biggest cities found that killings were up 45 percent in 2021 versus 2019. This week, I find from a sample of 30 states and selected cities that have already reported year-end traffic fatalities that car crash deaths grew 17 percent from 2019 to 2021. Those are both unusually huge two-year increases. …
“What’s going on on the roads?
“In 2020, the quantity of miles driven dropped versus 2019, which should have reduced car deaths, but the quality of driving plummeted. Total miles driven declined 13 percent due to lockdowns and work-from-home, while deaths per million miles driven went up an insane 23 percent, leading to a 7 percent overall increase in car crash deaths.
“In 2021, miles driven rebounded to only a few percent less than in 2019. Yet, despite the return of traffic, deaths per million miles continued at nearly the same reckless (and thus wreckful) rate as 2020, improving by only a few percent.
“When everything shut down in the spring of 2020, some drivers took advantage of the empty streets and the suddenly reclusive police to drive like bats out of hell. For example, despite no doubt wanting to socially distance in 2020, the Utah Highway Patrol wound up issuing 31 percent more tickets than in 2019 to drivers speeding over 100 miles per hour.
“Despite the worse driving, American cops prudently ticketed drivers less. To take a random example from the Indianapolis Star:
“‘Overall, Indiana State Police issued about 25,000 fewer speeding tickets in 2020 than 2019, a 41.1% decrease.’
“Similarly, WSHU in Connecticut reported:
“‘For both state and local police, the combined number of stops statewide dropped from 512,000 in 2019 to about 188,000 in 2021.'”
https://www.takimag.com/article/american-driving-in-2021-reckless-and-wreckful/
That’s a pretty good theory, but unfortunately I think the 2021 data complicates things a lot. 2020 increases indeed were largest for black Americans, and were also collocated with urban areas and contemporaneous with civil unrest. But 2021 has also had a dramatic increase. They haven’t released their sub-population breakdown, but this increase comes:
1. without much evidence for a further retreat in policing after 2020
2. after homicides have seemed to nearly plateaued
3. Are more dispersed across the country with large increases in regions with almost no black population.
The 2020 report
https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/813115
The 2021 report
https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/813240
Also interesting: in the subcategory link
https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/813118
a whopping 29% decrease in traffic fatalities for Asians, a larger absolute change than for any other category except for the “Other” race, which had an even larger 54% increase. I *have* to guess that the Asian change is due to proportion working from home or something.
Honestly I’d be super cautious about stuff like this because there seems to be a high risk of Simpson’s paradox here. The experience of the pandemic across the US is not the same. The states (or counties, even) that experienced traffic reductions may not be the states that experienced increases in accidents. Different states also experienced peaks in the virus and thus lockdown activities at different times. This extends further to people. For example, risk averse individuals might avoid travel, but risk takers (either by choice, or economic necessity) might actually travel more. After all, we can all picture the meal delivery worker getting a bunch more work during this period, and they have an incentive to drive fast.
>This is a pathology exclusive to Americans. What we saw in European countries is that as driving went down, crashes went down; as driving went down, fatalities went down; there was not the correlation we saw in the U.S.
and
>Prior to the pandemic, the one ubiquitous condition that we experienced across North America is high levels of traffic congestion
seem to suggest that ubiquitous congestion is somehow a N.A. specific phenomenon which I highly doubt.
A. Congestion is not so ubiquitous in N.A. unless you are talking about major cities like NYC, LA, Chicago, SF.
B. Even for major cities, while NYC congestion is the worst in the US, it is comparable to cities like Paris, London, Rome.
Congestion as an explanation for increased traffic deaths in the US would require them to
show that either US had particularly high congestion pre-COVID (compared to other comparable countries with good data) or
that US had experienced particularly low congestion after COVID.
Or at the least that US has a higher proportion of DUI (probably difficult to compare across countries).
What’s relatively unique to the United States is that suburban sprawl actually dominates city centers, pushing up both car ownership, miles traveled, and minutes spent in traffic relative to European counterparts. Learning about the lack of suburbia in Europe, at least in the American sense, was a major shock for me.
Also, I think the original post is referring to road design, not congestion. To take NYC as an example, you don’t really find that many roads like Northern Boulevard, Queens Boulevard (https://goo.gl/maps/TRDgy1kJ1ScWuYfU8), Atlantic Avenue (https://goo.gl/maps/9cdYZKggsjzrVmJZA), or Flatbush Avenue (https://goo.gl/maps/MhqwMEkE8S6t5xeH7) in, say, London, where this
https://goo.gl/maps/Li1NZswPNtAEd2JBA
is their widest street
European countries tend to be much less reliant upon police stopping speeders to deter speeding. They use automated radar guns that send tickets in the mail to speeders. So, traffic enforcement goes on even when the cops are social distancing due to fear of infection or are retreating to the donut shop in the wake of George Floyd.
I got stopped by cops a couple times in Europe, but not for speeding. Once was for going through some red lights and once was for biking in a pedestrian zone. One thing I remember was that when I was stopped in France, there were 4 cops in the car, 2 in the front and 2 in the back. In the U.S. you’d never see a cop sitting in the back seat. In both cases, cops were nice and didn’t give me a ticket. My weak language skills either annoyed them or motivated them to take pity on me; I have no idea which.
They may have been just checking for more serious infractions like driving while impaired or so forth.
I was pulled over on a busy urban interstate (in USA, not Europe) around midnight on a Saturday while driving far over the speed limit (irresponsibly so, I’m ashamed to admit). As soon as the patrolman seem convinced I was not under the influence of anything he told me to slow down but he was too busy to stop and write a ticket.
Well, if this is true, this strongly supports a switch in traffic policing models, then.
This seems like a fun article related to this topic for a friday.
“Sexual behaviours among dyadic vehicle occupants”
https://www.sciencedirect.com/science/article/pii/S1369847821002709
“This project was supported by Queensland University of Technology (QUT) and the Motor Accident Insurance Commission (MAIC) Queensland [and] an Australian Research Council Discovery Early Career Researcher Award Fellowship (DE200101079).”
Your tax dollars at work.
Wait, it’s not April 1 yet is it?
Nope. They published two papers based on it. The journal is a top one in the field, too.
I’m astounded that (1) anybody would do this work to begin with, (2) that anybody would supporot this research, (3) that the maligned peer review process approved it, and (4) that the journal published it. Now, perhaps the article explains what relevance the subject has but I’m not willing to spend any money to read it. From the abstract, it seems like they’ve discovered that most porn films that involve sex in a car while driving involve a man driving and being the recipient of an act by a female passenger. I guess this naturally leads to discovering policy options to improve safety on the roads. Or maybe the point is that we need to achieve gender equity in automobile sex acts or in porn films. Or????
Dale:
I could see it as a student project in a data-science-for-humanities-students course or something like that, but, yeah, I’m genuinely baffled that these official funding bodies would be paying for it. I respect the value of sex/gender/media studies—hey, I wrote a paper on gaydar myself!—but this one seems pretty contentless.