Its Clustering Time! 2024 NFL Draft QB Edition
Long time no see substack! Let's cluster some QBs
After another cycle of prospect evaluation, discourse, and our favorite draft rumors, the NFL Draft is finally upon us this week. Personally, the NFL draft is one of the my favorite times in the entire year. Watching players that we watched all throughout their college career see their dreams come true as they walk across the draft stage is a magical moment each and every year. Its also the time where virtually every fan base sees a surge in optimism as fans get to lie about the prospects their teams drafted.
As it is in normal football discourse, QBs dominate the draft conversation nearly every cycle. Hitting on a QB prospect is a teams one way ticket out of the basement and into contender status. While it does not guarantee a team a title, 9 of the last 12 Super Bowls featured a team with a QB on his rookie contract (including the 5th year option).
Figuring out which QBs are going to succeed and which ones are not going to pan out is extremely hard, and any individual’s hit rate on picking QB’s is going to be extremely low. That being said, evaluating QB’s is extremely fun and im not letting some low hit rate stop me from ranking my QBs. First things first, let’s see what this draft class looks like from an overview:
Im using the draft grades given out by Lance Zierlien on NFL.com as a way to compare this class with previous draft classes. This draft class compares to last year in that there are two QBs who stand out from the crowd (Caleb Williams and Jayden Daniels), followed by another tier of QBs (Maye to Penix Jr.) and then a bigger group of players occupying the mid to late round spots. Caleb Williams owns the top draft grade in this class, but his grade is nowhere near as high as Trevor Lawrence, Joe Burrow etc.
Overall, this class seems a bit stronger at the top vs. last year’s class. According to Arif Hasan’s consensus big board, six QBs (Caleb Williams to Bo Nix) are in the top 40. Sportsbooks have set the line at 4.5 QBs selected in the first round, and it is likely that all six of these QBs will be gone before the end of the 2nd round.
The question on everyone’s mind is… Who is good? Who is overrated? Through the power of data we are going to try and get some answers on these prospects. The best way this newsletter knows how to accomplish this task is through good ole fashioned K-Means clustering. To keep it brief, we will be taking a bunch of variables and using an algorithm that will group QBs together into clusters based on how similar their numbers are to a point in the middle of a cluster. If you want to see what the results of this on last’s years draft class, you can view it here.
Just how many variables will we be using? Im glad you asked. I have complied ONE HUNDRED AND SIX stats/metrics etc. on the QBs from the 2017-2024 draft classes. Everything ranging from HS recruit rating, efficiency, accuracy, measurables like height and weight, you name it I threw it in there. I also included the data that I used on my play creation tiers that was featured on PFF.com here. The variable list has a mix of film based metrics (like PFF grades and accurate throw rates), raw statistics like expected points added per play and completion percentage over expectation, and opponent adjusted stats like ESPN’s QBR. Every year I try to cover as many bases as I can and this year I had my biggest expansion to date.
Clustering Results
This is the final result after all of that clustering. To make it a little easier to view the clusters, they were ordered by average NFL.com grade. At the top you have the QBs who were elite for the entirety of their time as a starting QB in college, all the down to QBs that really didn’t have that great of a college career according to the numbers. By far the most interesting tier in my opinion is there one right here:
Among the 75 QBs in this pool of prospects, these four had the highest uncatchable throw rates in their college careers. Their biggest connection is big, strong armed athletes that have tools you cannot develop. Josh Allen is the 2nd best QB in the NFL, and Anthony Richardson had a promising start to his career before an injury sidelined him. DeShone Kizer did not have a good career by any stretch of imagination. This year’s mega athlete is the 6’5” (93rd percentile among QBs) 246 lbs (97th percentile) Joe Milton of Tennessee. While his college career was not great, the tools keep you up at night. His odds of succeeding are low, but you still are intrigued to see where he falls in the draft and which team attempts to form him into the QB we all wish he would be.
Clustering By Draft Year
This is an easier way to see where the 2024 draft prospects fall in the clusters. At the top is JJ McCarthy, who led Michigan to a national championship and a 27-1 record as a starting QB. You might ask “How can you say he was elite Michigan rarely had to use him!”. While that is true, when he was used he was extremely efficient. In 2023 he ranked 3rd in the country with a 89.2 QBR, and his career 0.37 EPA/Play ranked 11th among the QBs in this dataset.
The next grouping of 2024 QBs are Caleb Williams, Michael Penix Jr. and Bo Nix, who occupy the tier that includes Patrick Mahomes and (sorry Tuanon) Justin Herbert. All of these QB’s appeared in the same cluster in my play creation tiers, which is the main characteristic of this tier.
Pressure to sack rate has been studied a lot recently due to its ability to translate over in the NFL levels. In this dataset, Michael Penix Jr’s 6.5% pressure to sack rate is the lowest by over a full percentage point. Bo Nix 11.4% rate is 7th behind Patrick Mahomes (11.3%). As for Caleb Williams, his ability to create under pressure is the reason his name will be called first in the NFL Draft this week.
Drake Maye, Jayden Daniels and Jordan Travis were all highly productive in college and occupy the third tier in these clusters. The biggest calling card for this tier is protecting the football. In this data set, Jayden Daniels owns the lowest PFF turnover worthy play rate at 1.7%, followed by Drake Maye at 2.1%. For Jordan Travis, his -10.5 expected points lost due to turnovers was the 2nd lowest in this dataset of QBs.
Tier Strengths/Weaknesses
Averaging out the variables based on tier, we can get an idea of the strengths and weaknesses of each tier. Here are those strengths (ranking in the top 2 in certain variables among tiers) and weaknesses (ranking in the bottom 2). It should be noted that just because a 2024 QB falls into a particular tier does not mean those strengths and weaknesses necessarily reflect them. Nobody would consider JJ McCarthy short but having Kyler Murray and Bryce Young in your tier will bring down your average.
Smushing 106 Variables Into Two Dimensions
One final way we can look at the clusters is by using principal component analysis to essentially combine all of our variables into two components that can be plotted on a scatter plot. This plot makes it easy to see the divide between clusters such as the Burrow/Fields/McCarthy cluster and the Daniel Jones/CJ Beathard Cluster.
Conclusion
Clustering QBs allows us to sift through a massive trove of numbers and group QBs based on those numbers. Not all QBs will fall into the clusters that we think they should fall into, but for the most part clustering has had a solid track record with this newsletter. Overall, this looks to be a strong QB class and it will be interesting to see where they go in the NFL draft.
Buy Me A Coffee!
https://www.buymeacoffee.com/CFBNumbers
As long as I do this newsletter I will keep it free so that as many people as possible can enjoy and join us on our CFB data adventure. However, if you did want to show additional support the newsletter (which you 100% do not have to!) you can always buy me a coffee here!
If you want to dive in to the data like I do, check out @CFB_Data and @cfbfastR on Twitter, where you can learn how to get started in the world of College Football data analysis!
If you want to see more charts and one off analysis, follow my twitter page, @CFBNumbers