Introducing NASCAR Composite Winning Percentage
A new way to look at the best drivers in the sport.
When analyzing racecar drivers, I’m a big fan of trying to measure performance against a baseline of expectation using teammates and other comparison points. I’ve tinkered with this in NASCAR before by looking at adjusted points relative to various groupings of fellow drivers, but today I want to introduce a new (and in my opinion, more elegant) way of comparing drivers by looking at a weighted series of head-to-head winning percentages.
Broadly speaking, we can look at a driver’s win-loss “record” against any grouping of opponents in a race — some of which serve as better comparison points than others. If a driver wins a race with a 40-car field, for instance, they would have a 39-0 “record” in that race (beating 39 opponents and losing to no one). We can add this up over the course of a season or any subset of races, to get a general sense of how dominant a driver has been against the competition. Among drivers in the modern Cup Series (since 1972), the best winning percentage against all opponents belongs to Dale Earnhardt Sr., who won 73.4% of his head-to-head “matchups”.
Other comparison points include records against opponents in the same car manufacturer (i.e., Ford, Chevy, Toyota, etc.), since certain cars run better at particular tracks and there is a quasi-teammate aspect to drivers that share a manufacturer, even if they have different owners. Bobby Allison, who drove for a wide variety of different manufacturers — from Chevrolet and Ford to Matador, Buick and Mercury — is the all-time leader in record against drivers in the same car type, with a 75.3% winning percentage.
And in the spirit of fantasy games that measure start-versus-finish differential, it’s useful to look at a driver’s record against opponents who started within +/- 5 positions on the grid to begin a given race. (Although qualifying is also tied up in driver skill, it can also speak to inferior equipment — which can be overcome to a degree through superior racecraft.) Earnhardt Sr. was the master of outperforming his initial positioning, beating 66.5% of drivers who started near him on the grid.
Finally, we come to one of the most interesting comparative categories: Performance against teammates. Besting your teammates can mean many different things, from shining on a bad team to holding one’s own within a star-studded group. (Some drivers don’t even have teammates, particularly in earlier eras of the sport.) Kevin Harvick has been the best in that regard, with a 70.6% winning percentage against teammates in his career. But we can’t simply use that number in a vacuum, since the quality of teammates can vary dramatically depending on the team. To account for this, I also track the W-L records of a driver’s teammates against everyone else in a season — basically, category 1 from above, which I’ll call Strength of Schedule or SOS% here — and adjust using the following formula:
Adjusted W% vs. Teammates = (SOS% * W% vs Tm) / ((2 * W% vs Tm * SOS%) - SOS% - W% vs Tm + 1)
This effectively adjusts the raw winning percentage against teammates to its equivalent value against a perfectly average (50% winning percentage) set of teammates. Harvick is still the leader after that adjustment — and in fact, his adjusted rate rises to 76.6%, since he posted his 70.6% winning percentage against teammates who collectively won 58.0% of their matchups against everybody else. Turns out good old Happy Harvick can drive!
The only step left is to combine all four winning percentages into a composite measure. In my initial attempt at a metric such as this, I assigned the following weights, mostly through intuition (plus what I thought I was trying to reward/highlight:
Teammates (50% weight)
Drivers who started within +/- 5 spots on the grid (32.5% weight)
Drivers in the same car manufacturer (12.5% weight)
All drivers in the field (5% weight)
But as you will see, the value of these weights can change wildly depending on what we use to test/optimize them against. After further research that involved looking at what best predicted performance in opposite halves of the season (even vs. odd) and among the subset of drivers on new teams (a good natural experiment for isolating the true value of various metrics), I have amended the weights to the following values:
vs. All drivers in the field (35% weight)
vs. Teammates (25% weight)
vs. Drivers who started within +/- 5 spots on the grid (25% weight)
vs. Drivers in the same car manufacturer (15% weight)
This weighting scheme strikes a better balance for what we’re trying to accomplish: mixing a driver’s overall performance with their performance under various specific conditions that each tell us something different about their abilities.
According to this new metric, the best retired driver in modern Cup Series history is Earnhardt Sr., with a Composite% of 71.2%, followed by Jeff Gordon, Carl Edwards (!), Cale Yarborough and Tony Stewart. But the best active driver is even better than Dale Sr. under this metric — and it’s the man who was his replacement after that tragic day in Daytona 22 years ago: Kevin Harvick.
(Note that Harvick and Earnhardt have a bit of a Simpson’s Paradox situation going on, whereby Dale Sr. has superior winning percentages in each category, but has a lower overall Composite% because of the different weights assigned to different categories — particularly the teammate comparison, since Harvick had teammates his entire career and Earnhardt spent most of his career without them.)
This is far from a be-all, end-all type of stat, but it will hopefully provide a framework that enables comparisons of drivers under different conditions, and allows us to split out where each driver added the most to their performance. And if you want all-time season-level data for Composite W% (or any of the data that went into calculating it), you can find that on my GitHib.
Filed under: NASCAR