2025 Open Wheel Model Update

I'm finally teaching myself some slicker production values...

I couldn’t visit Mom on Christmas because the North Syracuse bus doesn’t run that day, so I didn’t really celebrate and decided to finish up my touring car model instead. Well, that’s not really true. I did go to Christmas parties at the nursing home and the autism group and I went to the 10 p.m. Christmas Eve vigil mass, but what I guess I meant by that is that I didn’t do any decorating (for whom?) or buy presents or send cards or anything else. I think Christmas is for being with other people anyway rather than exhibiting a display of various consumerist accoutrements like “holiday inflatables” (I’m still laughing at the idea that those ever became a thing). So, I mostly just spent the actual holiday working on my touring car model, my top 200 list, and having an endless stream of diarrhea. Fun, fun.

I still think I’m finishing this year a little better than I started it (which is probably the first time I can say that in at least four years since my typing book was published). I cleaned up my mom’s hoarding to the village’s satisfaction, landed multiple jobs, set up payment arrangements for some of my credit cards in default, started to open myself much more to the real world for the first time in decades, read more books than I have in years on my bus rides to visit Mom as well as reading to her personally, lost some weight, and I think my mom and I have been treating each other somewhat better since she went into the nursing home as absence made the heart grow fonder I guess. Although I’m not going to quite finish the top 200 list by the end of the year, I think I will finish all the research for it by then. I just completed my British Touring Car Championship table, and I’m going to try to finish DTM, WEC, and IMSA over the next few days. I probably won’t bother with the European Le Mans Series this year. And thanks to a new paid subscriber I got yesterday, I have now crossed the $1,000 mark in terms of annualized income from the Substack, so it looks like there might be some real potential here. This week’s been mildly annoying as there was a massive snowstorm yesterday and I had to shovel the walk with a shovel that I was unable to put together. My neighbor helped me finally screw the top on after I had been unable to figure out how to do that for weeks (apparently, there was a missing part). Even worse, there was a water main break in the neighboring town of Cicero and a lot of the local towns in the Syracuse area have been told to reduce their water usage. The village of North Syracuse isn’t part of that, but I’ve decided to refrain as much as possible so I haven’t taken a shower, done my dishes, or washed my laundry since even though all those things are well overdue. Pretty nasty that that happened a few days before Christmas in what was probably the worst time of year to make a repair. I’m handling this better than I thought I would though. I’m probably a lot more stoic than I once was at this point.

While stalling to give myself time to finish the list, I’ve decided to post updates for my open wheel, stock car, and touring car models. These will all be free posts. If you’re new here, the open wheel model includes all drivers who made at least one start in any of the four major league open wheel series (Formula 1, IndyCar, Formula E, Super Formula, or any of their predecessors) in the post-World War II period, the stock car model includes all drivers who made at least one start in the NASCAR Cup Series, and the touring car model includes all drivers who made at least one start in the TCR World Tour, Supercars, BTCC, DTM, Brazil’s Stock Car Pro, Porsche Supercup, or any of their predecessors. For each driver, I collect head-to-head teammate comparisons in the races both drivers finished (excluding DNFs/retirements and disqualifications) in any series (including minor league races), as long as both drivers within each teammate pair made a major league start at some point.

I understand a lot of other models like this remove crash DNFs, but I decided not to do that because honestly, it’s hard for me to argue that every crash is the driver’s fault (I realize other analysts like David Smith disagree). Plenty of crashes are caused by mechanical breakdowns or tire failures, and since drivers on weaker teams are more likely to have mechanical issues, these sorts of models are going to be intrinsically biased against drivers with weaker equipment. Isn’t identifying diamonds in the rough for weaker teams one of the points of these sorts of analyses? If you are an experienced driver, can you have better odds of avoiding a crash? Sure, but it still isn’t guaranteed, and often whether crash damage takes you out of the race or not can come down to sheer dumb luck. Ryan McCafferty’s analysis of NASCAR crashes argued that only 19% of Cup Series DNFs were due to driver error and only 24% of all crashes resulted in a DNF for the driver at fault. I realize those numbers will probably be higher in open wheel series, but they’re still far too low for me to feel comfortable rating drivers on that basis. His data supported my own assumptions that all DNFs (not even merely mechanical DNFs) have a significant luck component, and this reinforced my own decision to exclude all DNFs, not merely mechanical ones. I do think I am going to consider whether a driver crashes a lot to some extent in my 1,000 greatest drivers list, but I definitely think that metrics on how clean a driver is should be completely separated from rankings based on pace. I prefer ranking drivers primarily on what they do in their best races (not their worst), and these models do a good job of reflecting that.

Although I never gave him the credit publicly at the time, I originally got the idea for these models from Joe Lunardi’s adjusted scoring margin. Lunardi’s college basketball statistic rated teams by comparing their actual margins of victory with the average margin of victory for each team’s opponents. According to ASM, if Duke beat a team by 25 points and that team lost their games on average by 5 points that season, Duke would have an ASM of 20 for that game. I adapted the same concept here, plugging in each driver’s winning percentage against their teammates where Lunardi used scoring margins. So, if one teammate finished higher 90% of the time in races where neither driver DNFed, but that driver won 30% of the time against all their teammates throughout their career, that would produce a rating of .2, reflecting the actual winning percentage of .9 - the expected winning percentage of .7.

The big difference between my models and Lunardi’s is that I reiterate my models thirty times and plug in each driver’s rating from the previous iteration to determine the expected winning percentage instead of the actual winning percentage, which helps clean up discrepancies so that drivers with strong winning percentages against exceptionally weak teammates will decline and drivers with weak winning percentages against exceptionally strong teammates will rise. I think 30 iterations is sufficient to produce convincing results for the most part. The model is defined so that the average driver at the major league level is rated 0. You can therefore predict the probability a driver will beat an average teammate at the major league level based on their average career performance by adding .5 to each driver’s career rating. You can likewise predict how two teammates will perform against each other by subtracting one driver’s career rating from the other. So, if you wanted to compare Max Verstappen and Álex Palou, say, my model states that Verstappen would be expected to beat Palou 71.2% of the time, because that equals .5 plus the difference between Verstappen and Palou’s ratings of .469 and .257 (.212). I do think Palou was slightly better this year, but most other years Verstappen was a lot better, so that makes sense.

While there will always be outliers, I can at least understand where the numbers come from. You will all likely find Lance Stroll far too high, but it makes sense because he was both Sebastian Vettel and Fernando Alonso’s teammates when they were in their decline period. For the most part though, I think my open wheel model especially nails which drivers are above average and below average. No, I don’t think Oscar Piastri and Ritomo Miyata are below average and there are probably a couple drivers rated above 0 I’d say are below average, but not many. My models have been very successful at predicting drivers’ futures: Mitch Evans was the highest-rated Formula E driver in my open wheel model for its entire history until his really subpar season this year, but he only had two wins at the time I launched the model, and he’s now tied for the most Formula E wins ever. My stock car model also forecast the future successes of Chris Buescher and Ross Chastain; I believe Carson Hocevar will be next. And I’m especially proud that my model had Alex Albon ahead of Carlos Sainz, Jr. before this season when I don’t think a lot of people were expecting Albon to finish higher in points.

For each driver, I list their ratings at the end of both the 2024 and 2025 seasons for comparison’s sake while also listing each driver’s single-season rating for 2025 only. In most cases, if a driver’s 2025 rating was higher than their 2024 career rating, they will go up, while they will go down if their 2025 rating is lower. There are exceptions in cases where the second-order effects end up having a larger effect than the first-order effect. For example, Fernando Alonso dropped from .423 to .416 even though his rating was .437, because the second-order effects of his past teammates Lewis Hamilton and Esteban Ocon’s significant declines ended up overpowering the first-order effect of Alonso slightly outperforming his previous career average. There are other drivers who improved despite lower single-season ratings because other drivers were integrated into the model for the first time. Obviously, drivers who have a lot of teammate comparisons already will have ratings that are more fixed than drivers who have fewer. Even though Scott Dixon and Josef Newgarden started with the same rating of .235 and both of them significantly underachieved with Newgarden posting a slightly higher 2025 rating, Dixon didn’t fall as much because he had a lot more teammate comparisons to begin with. With this update, Taylor Barnard overtook Mitch Evans to be the highest-rated Formula E driver and Álex Palou overtook Pato O’Ward to become the highest-rated IndyCar driver. Despite losing the title in the last race to Ayumu Iwasa, Sho Tsuboi remains the highest-rated Super Formula driver.

Tonight, I discovered the website Datawrapper, which produces snazzier data visualizations than I have ever been able to produce, and data visualizations were one of the key things I was missing in my toolkit. Now that I have that, maybe my work will start looking a lot more professional now. Yeah, I caught a few mistakes after I finished the table. I had already entered all the announced 2025 drivers in my December 2024 update, so I had a rating for Kimi Antonelli based on his Formula 2 season against Oliver Bearman (obviously, both of them crashed to earth once they had real competition, but they’re still correctly above average) but I technically should have left that blank, and I forgot to put the umlaut over the u in Maximilian Günther. Those are both minor and forgivable I guess. This still looks better than anything I’ve produced before here.

I certainly won’t have such long intros before the stock car and touring car model updates. I might just post the tables, but I felt this post deserved a longer intro since there are probably some new people from the typing world or whatever who haven’t seen my work before and also because this is my first serious attempt at data visualization. Enjoy.