Clustering the 2023 NFL Draft QBs: Part 1
Jamming a bunch of numbers together to see who is similar to past QB prospects
Long time no see substack! First off I do want to apologize for the absence, life has been a little hectic lately and I haven’t been able to do much data exploration. Now that we’re back, its time for one of my favorite articles of the year: clustering some QBs!
The NFL draft is coming up at the end of next week. All of the rumors, mock drafts, and trade ideas all come together next Thursday. For the fans of the NFL draft, its like the series finale of your favorite TV show. For those that hate it, your timeline will soon be cleansed of all draft content. As it is on the field, the main draw to the draft starts at the QB position.
Last season we only saw one QB, Pittsburgh Panther turned Steeler Kenny Pickett, selected in the first round. This year, we have four QBs expected to go in the first round: Bryce Young (Alabama), CJ Stroud (Ohio State), Anthony Richardson (Florida) and Will Levis (Kentucky). Before we dive into some clustering, let’s get a broad overview of the 2023 draft class.
Looking at each prospects NFL.com Draft Grade, we can see how this class stacks up to previous classes. This is what the numbers mean:
Bryce Young and CJ Stroud has separated themselves from the rest of the pack, and represent the two “elite” prospects in the draft. Last year’s draft class, which was considered a down year, did not have prospects in that “elite” category. Richardson/Levis/Hooker are all in that middle category that has produced everything from the superstars of the league to guys fighting for roster spots. The rest of the class have grades lower than a 6, which means backups and lower priority guys.
The sheer amount of information on prospects continues to grow each and every year. Whether its film, data, combine measurements or anything else, the overload of information can be hard to sift through. Luckily, we have a neat little trick we can use to get some quick comparisons to past prospects: K-Means clustering.
Clustering TLDR and Variables Used
Clustering is something I have done many times on this newsletter, its basically the bread and butter of this whole thing. If you want to see how last year’s draft class was clustered, you can check that out here. Essentially we are going to take a bunch of variables, standardize them so they’re all on the same scale, then feed those into the algorithm which will give us our grouped clusters. These are the variables we will be feeding into our clustering algorithm:
Raw Efficiency Stats: This includes traditional QB stats (Comp %, Yards, TD-INT etc.), EPA/Play (both passing and rushing), Success rate, Explosive play rate (plays that are above the 75th percentile of EPA). Early and late down splits were also included. These variables are only interested in what happened on the field, with no opponent adjustments included.
Adjusted Efficiency Stats: These mainly come from ESPN’s QBR, which is essentially a per play efficiency metric that adjusts for opponent and game situation. The actual average adjustment (QBR-raw QBR) was also included. Adjusting in CFB is hard due to the sheer number of teams, talent disparity and schemes, but this gives us a little bit more context into their play on the field
Charted Stats: PFF charts every single play of every single game, and that information is transformed into things like PFF Grades, Big Time Throw Rates, Accurate throw rates etc. All of that is included here. While some scoff at PFF grades, they have constantly stood the test of time. This also adds a little film based perspective to our clustering.
Misc Variables: 247 Composite Recruiting rankings, combine measurements, and pre draft rankings. This rounds out our variables and adds in where they stand pre-draft.
Clustering Results
Now that we have all of our information, we’re ready to cluster. There are many ways to visualize the clusters, but I have found two that really showcase the clusters in the best way. The first is by arranging them into tiers, using one variable for the arrangement. In this case, will use the cluster’s average NFL.com draft grade:
This is the final result. If you don’t want to go hunting for the 2023 NFL draft prospects, here is where they ended up:
Video Game College Numbers (1): CJ Stroud
These QB’s were the best of the best in the CFP Era. For CJ Stroud, he finished 1st in QBR in 2021 and 3rd in QBR in 2022. Very rarely did this QB’s have off performances, which was the case for Stroud. His off “games” were really just off halves (Minnesota in his first start in 2021, Notre Dame in 2022).
Highly Productive In College (2): Bryce Young, Hendon Hooker, Stetson Bennett, Jaren Hall
You may be wondering “Why is Bryce Young here?”. After a heisman campaign in 2021, Young’s numbers slightly dipped in 2022, which most likely sent him to this tier. This was due to a combination of “theres nowhere to go but down” and his lack of support in 2022. While a small sample, this gives you a little idea of the different situations Young was put in vs. his Alabama counterparts:
The Josh Allen High Roller Tier (3): Anthony Richardson
Josh Allen is the most used comparison for Anthony Richardson this draft cycle. These QBs are all either big or insanely mobile or in the case of Richardson, both. While the names in this tier aren’t exactly comforting, none of these QBs (even Josh Allen) reach the athleticism that Richardson has to offer.
Not QB1’s Predraft… But? (4): Clayton Tune, Dorian Thompson-Robinson
None of the QBs in this tier were ranked 1st in their respective class coming into the draft (though Kenny Pickett ended up as the first QB taken). However, when you see the names on this list, you see some established superstar NFL QBs, and Kenny Pickett who showed some promise in his first year. Clayton Tune and DTR aren’t ranked nearly as high in pre draft rankings as the other QBs in their tier, but both have been floated as the “next Brock Purdy”.
Big Guys, Big Arms (5): Aidan O’Connell
For the most part, this tier includes the prospects that came from air raid backgrounds, or mostly won with their arms as opposed to their legs. Aidan O’Connell comes from the Jeff Brohm air raid system which got him up to 5th in ESPN’s QBR in 2021. While Justin Herbert is the only established star QB on this list, most of this list have taken snaps in regular season games.
Daniel Jones Got A 2nd Contract (6): Will Levis, Tanner McKee
Will Levis finished 2022 ranking 61st in ESPN’s QBR and 93rd in PFF Offense Grade. It isn’t really a shock that he is lower in a purely numbers based clustering. His draft stocked is fueled solely on his physical tools and his experience in a McVay tree offense, which in conjunction with Shanahan concepts are the predominant schemes in the NFL. Another factor to consider for Levis and Tanner McKee is their support in situations were not the best in 2022, which can partially explain the poor showings on the stat sheet.
JAGs: Jake Haener, Max Duggan
Well… Brock Purdy is in this tier! Sam Howell and Desmond Ridder are (as of now) in line to start in 2023, but their drops in the 2022 NFL draft signal the league was not as high on them as we were pre-draft. As for Jake Haener and Max Duggan, they’ll probably need that mixture of excellent situation + unfortunate injuries that we saw with Brock Purdy.
Just looking at the clusters in a tier form can leave you wondering how close the tiers are to each other. One way to solve this is by looking at the clusters through two dimensions (Principle Component Analysis). As you can see, the first tier has clearly separated itself from the rest of the group, which has some separation but is mostly blended together.
One final thing we can do is look at the averages of each cluster among variables to see how each tier separates from each other. Earlier I said the “Big Guys, Big Arms (tier 5)” tier won with their arms as opposed to their legs. Looking at their PFF Rush grade, you can see they are well below each of the other tiers in rushing ability. Tier 7 (“JAGs”) is an interesting tier because they aren’t last in everything as you’d expect from the lowest tier, but their average draft grade is the lowest. This shows that while their college careers were pretty good when looking from a pure numbers perspective, their overall draft profile did not stack up to their fellow prospects.
This marks the end of the first part of our dive into the NFL Draft QBs. Next week, we will take all of our information and create final draft comparisons for the top 4 2023 NFL Draft QBs. By next week, we will have a complete understanding of where these draft QBs stand from a numbers perspective.
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