US Collision Patterns in 10 Pictures
Where do car crashes happen?
...Everywhere
When do they happen?
Aug 27, 2019
Mar 19, 2020
July 1, 2020
Sep 10, 2020
This chart shows the number of collisions in the continental US every day from Feb 2016 to New Year's Eve 2020. Traffic collisions spiked in mid-2019, jumped again in March 2020, and kept on spiking.
While it's tempting to blame the rise in car crashes on out-of-practice drivers getting back on the road after quarantine, the dates show that isn't the whole story.
​
That fuzzy caterpillar on the left-hand side of the plot gets its fuzz from a very regular weekly pattern. Strip it away to see the underlying trend:
There is much less short-term variation on the left when the weekly pattern is removed. Even after August 2019 when the number of collisions increases, most of the "fuzz" is gone -- until mid-March 2020 when both variation and total collisions increase! Whatever that variation is, it isn't the regular weekly pattern of the first four years of data.
​
The method here was to define a linear regression model that uses only the day of the week to predict the number of collisions that day. The prediction is subtracted from the actual number of collisions, leaving a residual for the day. This very simple model works well until the beginning of the pandemic.
So what changed?
Before the pandemic, there was a very strong pattern of traffic collisions on weekdays between 6 and 9 am, and again between 3 and 6 pm (earlier on Fridays). The driving (heh) factor for pre-pandemic crashes was weekday rush hour, which averaged between 50 and 55 collisions per hour.
​
When people started staying home, rush hour collisions should have decreased, right?
After COVID-19:
Well. Clearly, traffic accidents did not decrease during the pandemic. The evening rush hour collisions are still present, but instead of gradually increasing from 3 to 6 pm, collisions increase sharply around 1 pm and stay elevated.
​
That strong weekday/weekend pattern has faded a little, and the morning rush hour collisions are very faint compared to a new bloom of crashes between 11 pm and 2 am on nights after weekdays.
Extra Accidents:
Average collisions increased for every hour of the week during the pandemic. Although there were just 4 additional collisions on Tuesdays between 9 and 10 am, from 1 to 2 pm a few hours later there were 140 extra collisions. Then there is the really surprising rash of more than 100 extra crashes every hour from 10 pm to 2 am on weeknights.
Okay, so which roads are most dangerous?
These are the 100 roads with the most collisions on them. The top 10 most crash-prone roads in the country, in order, are
I-5
I-95
I-10
I-80
I-75
I-15
I-405
I-94
I-84
CA-99
(in light blue, above).
​
This is actually a pretty silly chart. Of course the roads with the most crashes are not the same as the most dangerous roads. A hundred-mile stretch of road will naturally have more collisions than your local Main Street!
By the way, when taken together, "Main Street" is the 31st most collision-prone road in America (not pictured).
A better measure of danger would be collisions per driver per unit time, but in the absence of those numbers, we can normalize in other ways.
This graphic normalizes by distance. Collisions mostly happen where the blue dots are, below. Each dot on this map represents at least 15 crashes in a 5-year period within a 21-meter radius (a circle just larger than the diagonal of a very large intersection).
This doesn't look so bad! There are still spikes in the major cities, the interstate roads in Oregon are all lit up, and in California, every highway around San Fransisco and Los Angeles is covered. The coasts of Florida, and the Northeast Corridor from Washington DC to Boston are a haze of blue. But many of the most crash-prone roads look much less dangerous when they are normalized for distance.
​
The most frequent collision sites were identified using the DBSCAN algorithm to separate the signal (clusters of at least 15 collisions) from noise (collisions near fewer than 15 other collisions).
​
Seasonal Change:
Summertime collisions are a little more likely in urban areas, while wintertime crashes are more likely on major roads between cities.
​
Minnesota is interesting. The presence of so many collisions up to the state line suggests that it's better at recording crashes than its neighbors. If that's so, Minnesota can give us some insight into crash patterns in Iowa, Wisconsin, and North and South Dakota.
TL;DR:
Collisions jumped in 2019 but followed a regular weekly pattern until mid-March 2020. Then collisions jumped again, and again, until they had more than quadrupled in frequency.
​
Crashes mostly happen in cities, especially Los Angeles, San Francisco, DC, Philadelphia, New York, and Boston, or along the main highways in Oregon and Florida. There is some evidence that reporting varies by state.
​
Across the country, collisions are more likely in fall and winter than spring and summer. This is especially true in Montana, North Dakota, and Minnesota. Oklahoma and Massachusetts are unusual for having more accidents in summer than winter.