Rrel Explained: Speed Score

It’s difficult to devise catch all metrics for composite processes. The goal in autoracing is to win races, and the simplest way to do so is to drive faster than everyone else - but driving faster means a lot of things. You can give it 100% all day, but you’ll find yourself wearing through your equipment. You can run safely and consistently, but can you run quickly at opportune moments? How do you measure what it means to drive fast in a 500 mile race? Is it who has the fastest peak, and thus records the fastest lap? Is it who is most consistently fast and records a strong average speed? Or is it somewhere in the middle? That’s what speed score is looking to solve.

Speed score is the most upfront metric at rrel.ca. We show it on every tab and almost every page of the website, because it’s the most easily digestible and translatable to reality. There’s three versions on offer:

Cumulative speed score, which you see under the Counts report. This represents a period-long (e.g. season-long - we have a bunch of filters you can apply, like all short tracks in the playoffs) evaluation of speed for a driver, car, or team (depends which page you’re on!). For example, taking a look at the 2020 driver report, you can see speed score has Kevin Harvick as the fastest driver all year, with Chase Elliott is second. Harvick up top doesn’t reveal anything new, but considering the understanding was that he and Denny Hamlin were favorites for the championship, Elliott being second and Hamlin in fourth could have revealed some opportune betting situations for the 2020 playoffs (where Elliott was fastest).
Per-race speed score (or average speed score), which is under the Rates report, and is simply the cumulative score per race (a la yards per game for a quarterback). This is more useful when you’re looking at a report where the number of races isn’t uniform, like if you want to see how much Kyle Busch dominated the trucks in 2020.
Adjusted average speed, also seen on the Rates report (and individual race pages), which is average speed score scaled to miles per hour. The intent of this is to make it a bit easier to understand.

Speed score attempts to compromise on the issue presented in the introduction by settling between average and fastest lap to evaluate drivers. It first takes the 95th percentile lap for a driver in a race (the lap where 5% of the driver’s laps are faster and 95% are slower). Then it takes the 95th percentile of all laps in the races. Finally it takes the ratio of the two lap speeds (note: not lap time), subtracts 1 (thus representing how much faster or slower the driver’s 95th percentile lap was than the race 95th percentile), and multiplies by 1000 to amplify the value. As a pseudo formula:

((Percentile(95th, driver's lap speeds) / Percentile(95th, all race lap speeds)) - 1) * 1000

The adjusted average speed simply adds 180 to that number.

There are some fair concerns here. The 95th percentile is somewhat arbitrary. I chose the 95th percentile specifically because it does two things: it requires drivers to run very quickly and requires them to do so fairly consistently. A driver can’t run a hot lap with fortunate drafting and put up a great speed score. They have to have a decent number of great laps. This maps to racing reality. You aren’t going to last long by running your equipment into the ground. You have to run quick all day (thus pushing up the floor of your speed score) to put you in a good position so that you can turn up the heat during opportune moments, like when you’re trying to make a critical pass to win a stage or a race (thus pushing up the ceiling of your speed score). Multiplying by 1000 can overamplify differences between drivers, but I think that’s fair. Partial seconds and few miles per hour can be the difference between a good day and victory lane. The biggest concern to me, the author, is that adjusted average speed score arbitrarily adds 180. When I was developing the stat, converting it to miles per hour was helpful to get a grip on what I was seeing, so I kept it, but I appreciate the concerns that it could be misleading or useless and so the metric is under review.

Finally, some caveats. Speed score means well, but it’s not infallible. Just because one driver is above or below another doesn’t mean that driver is better or worse than the other. In fact, this stat makes absolutely no attempt to suss out the contributions of the driver from the car, the manufacturer, the crew chief, or anything else. It’s a composite stat that measures the performance of the whole team. Large magnitudes of difference are likely representative of reality: I’m am very confident that Kevin Harvick was much faster than Aric Almirola in 2020 based on the 6 point difference between them. But am I certain of the difference between Joey Logano and Denny Hamlin? No - my only inclusion is that they were too close to call. Further, I’d advise breaking things down further: filter by package or by track type!

The hope is that speed score is a decent measure of car speed, and when used in conjunction with the other metrics and context available on this website, it should provide a good jumping off point into the world of autosport advanced stats and help in gathering a picture of overall car and driver value.