Daily Fantasy NASCAR: Is Practice Data More Valuable for Some Drivers Than Others?
We've got a pretty unique leg up in filling out daily fantasy NASCAR lineups that we don't get in other sports. We get to watch and quantify practice.
Imagine if you knew how much spin Gerrit Cole was getting on his fastball before using him in MLB DFS. Or if you knew that (shocker!) Josh Allen's warmup throws were a wee bit erratic on a Sunday morning. We would probably make better decisions if we had that data at our fingertips.
We get exactly that with NASCAR as most races will include at least a pair of practice sessions that tell us who is fast that weekend. We can pair that data with thoughts on current form and a sprinkle of track history, and we're going to have a good idea of who will run well when the green flag drops.
But there's a pretty major pitfall here. In practice, not every driver's going to be pushing as hard as they can. Allen Iverson's probably not putting down jaw-dropping 10-lap averages. And if they're not pushing their hardest and are lollygagging a bit while others shove the pedal to the floor, we may wind up underrating them if we put too much stock in their practice times.
There's never going to be a perfect way to know this, unless you were to ask each driver about their individual practice philosophy. It's hard for me to do that without confronting my greatest fears and interacting with other humans, so we'll try the data route, instead.
Each week, I have a model that ranks drivers based on their practice times, current form, and (to a lesser degree) track history so that I have an idea of who's going to compete for a win and who may underperform. I can split that model into different modules and see which tests best at different tracks, etcetera.
Testing the model shows that, overall, practice times are very valuable. For the entire season, the correlation between each driver's practice mark in the model and their average running position in the race was 0.737. That was lower than the correlation for their aggregate model score -- 0.805 -- and their current-form score -- 0.767 -- but practice by itself did tend to correlate with the strength of drivers during the race.
That doesn't mean it was true for individual drivers, though.
There are a couple of reasons that the stickiness would slip once you look at individual drivers. First, it's a smaller sample size, and that's never great if you want good data. Second, more volatile drivers -- I'm lookin' at you, Ricky Stenhouse Jr. -- aren't going to appear volatile in their practice data because you don't usually wreck in practice.
With that said, it can still be helpful to have an idea for which drivers' practice data is most representative of how they'll race that upcoming weekend. We just have to keep the limitations of this in mind before we overreact.
There are 27 relevant (aka non-back-marker) drivers who ran the full schedule last year and are returning for a full season in 2020. The table below shows the correlation between their practice marks in my model last year and their average running position and finish for each race (sorted by average running position). The higher the correlation, the tighter the relationship between their practice times and their speed in the race. The lower the correlation, the more random things were. I've excluded races at Daytona and Talladega as practice data there borders on being worthless.
|Driver||Finish||Average Running Position|
|Martin Truex Jr.||0.469||0.530|
|Ricky Stenhouse Jr.||-0.223||-0.187|
I, for one, am stunned -- STUNNED! -- that Stenhouse is at the bottom of this list. Words alone cannot articulate my shock.
It's also not a surprise that Matt DiBenedetto is near the top. There were a lot of tracks last year where DiBenedetto was effectively a non-factor because his equipment wasn't good enough. Alternatively, when the series went to tracks where the driver had a bit more influence, he popped both during the race and in qualifying.
For example, his practice mark in the Bristol fall race was the 19th-best mark for practice in my model the entire season. He proceeded to almost win the race and post a fifth-place average running position. When they were at the 1.5-mile tracks, DiBenedetto's lack of speed showed up in practice and helped us stay off of him in DFS.
Rather than crowning DiBenedetto as practice god, we can instead look to other drivers who will fit this archetype: those who have equipment that isn't up to the task at some spots but have enough skill themselves to wheel it elsewhere. Nobody will ever accuse Bubba Wallace and Michael McDowell of having great equipment, and both graded out as drivers whose practice times were more predictive of their output. DiBenedetto is now in better equipment with Wood Brothers Racing, so his practice times may now be less representative of how he'll run in the race, but it seems like those still in lesser equipment will likely provide us with high-quality data.
It also feels noteworthy that all four Hendrick Motorsports drivers were in the top 10, and three Joe Gibbs Racing-affiliated teams were in the top five. This could be an indication that certain teams put a heavier emphasis on practice times than others.
This theory seems to hold a bit of value when you look at the flip side of the table, as well. The five drivers with the lowest correlation between their practice ranks and their average running position were all Fords: one Roush-Fenway driver, two Stewart-Haas drivers, and two from Penske. For all of them, there was really no relationship between their practice runs and how they did during the race.
Perhaps more important, though, than team affiliation was age. You'll notice that some of the sport's spring chickens -- Wallace, William Byron, and Chase Elliott -- all had higher correlations between their practice data and their average running positions. Meanwhile, the bona fide vets in Kevin Harvick, Clint Bowyer, and Ryan Newman were all fairly random.
If we group the drivers into three segments -- those in the upper third of correlation, the middle third, and the bottom third -- we see this supported by data. The upper third -- the nine drivers with the highest correlation between their practice data and their average running position -- had an average age of 30.2 years with a median of 27 in 2019. That's despite having Jimmie Johnson in his age-43 season in that sample. The lower third were at an average of 33.0 years and a median of 31. That's a pretty significant gap.
It's not hard to come up with an anecdotal explanation for this. Drivers who are younger tend to have less experience, meaning there's a greater need for them to find the edge between speed and overdriving the car in practice. It makes sense that they would push it. The veterans, meanwhile, can work on more refined approaches, test the drivability of their cars in traffic, and more.
We're working with anecdotes and small samples here, so there's a decent chance that this assumption winds up being faulty. But it's not all that hard to buy into the narrative that younger drivers will have more reliable practice data while the elder group can partially be ignored.
It's also easy to explain away some of the outliers. Erik Jones was near the bottom of the chart despite having been in his age-23 season, but he -- like Stenhouse -- has a stronger affinity for crashing than you'd like. Practice speeds won't predict when he's going to put it in the fence, so he's naturally going to have a bad mark here.
For Kyle Busch and Martin Truex Jr., there isn't an easy explanation for why their practice data would be tied so well to their finishing positions. They're less volatile than guys like Jones, but the same is true for Joey Logano, and he skewed far more toward random than predictable. It could be the team-based narrative discussed above, or they could just be drivers who test their limits in practice. Regardless, we should note that they fit as exceptions to our rules, at least in a one-year sample of 2019.
Instead of focusing on the names of the drivers themselves for our takeaways, let's go through some buckets where practice times are more noteworthy than for others.
Based on what we saw last year, it seems like drivers who are young or who drive for teams with lesser funding are the ones for whom practice data is more predictive. So, if we see a driver in either or both of those buckets lay down some sick laps on Friday, we'll want to give them extra consideration on Sunday.
If they're more of a veteran, we should consider putting less weight on what they do in practice. So if Keselowski is merely having a middling weekend on the speed charts, we shouldn't overreact and completely omit him from our player pool and betslips.
There will always be exceptions to these, and it's based on just one year of data. But it at least helps to have some guidelines in our head for when practice may matter a bit more. We should always account for practice when making our decisions as it is broadly predictive of success during the race, but at least now we know a few instances when we can either pump the brakes on a potential panic or start to buy in aggressively.