Counting Accidents

My recent forays into algorithms that predict how drivers will perform forced me to dive deeper into statistics, including how we count accidents.

The traditional method relies upon NASCAR’s accounting of cautions. Each week after a race, NASCAR releases a summary of the race that includes the caution block, as shown below.

People like me translate this block into numbers and codes.

We count the number of planned cautions, the numbers of spins and stalled cars. We count accidents.

Through The Years

These numbers exist going back to the 1980s, but let’s just look at the last 10 years. The graph below breaks down cautions by year and by type.

There are a lot of interesting things to notice in this graph. The things that stand out to me include:

• Debris cautions (dark blue) have decreased significantly thanks to the damaged vehicle policy
• With the advent of stage racing and the damaged vehicle policy, the overall number of cautions has decreased
• Boy, there sure are A LOT of spins this year
• The accident count is way down relative to 2017.

Counting Accidents in 2022

We can make a similar graph for just 2022. Accidents are up a little. By if you consider each race independently, something should jump out at you.

The Indianapolis road course race was a crash fest. Yet the official accident count is… one?

I reviewed the race video because, well, I’m probably unhealthily obsessed with data. At road courses, incidents that would normally cause cautions often don’t. Unless an accident is so big that it blocks the track, or a car gets stuck, there’s no need for a caution.

That doesn’t change the fact: There was an incident.

So I counted all the incidents.

All the accidents.

This is an admittedly subjective process. I decided to only count accidents that caused a car to lose positions or caused enough damage that the car had to make an unscheduled pit stop.

By my count (in addition to the official incidents), the 2022 Indianapolis road course had:

• 10 accidents
• Nine spins
• Five off-track excursions
• Two miscellaneous incidents

The one ‘official’ accident, plus the 10 I counted, makes 11 accidents — more than any other track this year. No track has totaled nine spins in one race, either. And the off-track excursions on a road course would be hitting the wall at oval tracks.

I did the same exercise for three other road courses. (I haven’t gotten to Watkins Glen yet.) For the four additional road courses, I count:

• 19 additional accidents
• 24 more spins
• 10 off-track excursions that would have been hitting the wall at an oval
• 2 ‘other’ incidents

It’s Not Just Road Courses

Relying on cautions to count accidents also misses incidents on ovals. If a wheel comes off a car on track and the car stalls, that’s counted. But if a wheel comes off on pit road, it isn’t counted.

A bump-and-run, like Joey Logano’s turn-4 offensive against William Byron in the first Darlington race doesn’t get noted. Nor does a driver who hits the wall and falls back five positions. Or a driver who has their power-assisted steering fail and wrestles the car for an entire race.

If an incident isn’t significant enough to bring out a caution, you may think, why bother with it?

If you’re interested in counting cautions, you shouldn’t consider these incidents. But if you’re trying to understand why a driver is doing well (or not doing well) so that you can decide whether to put him on your fantasy team or wager on him, it’s important information.

And I guarantee you that the driver doesn’t think of it as insignificant.

It also skews the historical record. 2021 only had 78 accidents after 27 races. 2022 had 99. But there were only two road courses in 2020 and seven in 2021. It looks like the number of accidents went down, but in reality, they just didn’t all get counted.

An Example

William Byron started the season with a win at Atlanta, but his performance stalled. His average finish in the second half of the regular season was 7.7 positions worse than in the first half of the season.

According to the official accident count from cautions, he’s been involved in five accidents, no spins and no stalls. But if you include the uncounted incidents from road courses and even from ovals, he’s had 12 total incidents.

Byron does have six DNFs this year. DNFs are way up this year in the Cup Series. You might note that those other incidents are likely reflected in the DNFs. But so are some caution-causing incidents.

So Why Do We Do It This Way?

The short answer is because it’s the easiest. It look a lot of time to go back and check the videos, plus decide what to count and what not to count.

If I were still at a university, I would figure out a data-science project, get an internal grant for undergraduate research, and hire a couple of undergrads to watch and code races, and then analyze all the data.

Until we have that data, keep a skeptical note in the back of your mind when people start using accident statistics to justify things.