The NFL’s Big Data Bowl posted videos yesterday of the finalist entry presentations for the 2021 event, which asked “what happens after a quarterback takes a snap and drops back to pass?” Submissions used NFL Next Gen Stats to find data-driven ways to characterize and predict successful pass defense, typically focusing on separation data (distance from targeted receiver to nearest defender) and targeting/completion rates over expectation data (how likely the receiver is going to get thrown at and catch the ball compared to other players in similar situations).
Eight submission videos were posted by the NFL’s official YouTube account:
- Wei Peng, Marc Richards, Sam Walczak, Jack Werner – Evaluating Defensive Player Coverage: A New Framework
- Jill Reiner – Evaluating and Clustering Coverage Skill
- Meyappan Subbaiah, Dani Chu, Matthew Reyers, Lucas Wu – Let’s Eliminate the Defense
- Zach Bradlow, Zach Drapkin, Ryan Gross, & Sarah Hu – Check the TAPE, He’s Wide oPENN (Target-Agnostic Player ELO)
- Joe Andruzzi – Defender Evaluation: One-cut Routes and Double Moves
- Asmae Toumi, Marschall Furman, Sydney Robinson, Tony ElHabr – WADE (Weighted Assessment of Defender Effectiveness)
- Ella Summer – Random Player Effects: Defensive Backs
- James Venzor, Matthew Gartenhaus – Defender Performance by Pass Coverage Attribute
Each of the presentations are quite different and use their own approaches and methods, showcasing how creative analysts can be with raw data. A few of the results in these studies stood out as being both easy to understand (at least for me) with intuitive applications in a “real football” context.
Joe Andruzzi’s presentation did a great job of really involving football knowledge on route mechanics and defenders’ break on the ball. The end result he comes up with is a way to actually quantify coverage aggressiveness and categorize cornerback types by figuring out from angular and acceleration data how good the corners were against routes that only have one change of direction at most versus two (the double moves). At the top of the double move list? Darius Slay, who turns out to be one of the best overall cornerbacks in Andruzzi’s evaluation system.
Venzor and Gartenhaus found that more than 60 percent of the defender’s ability to make a difference (expected points saved in their modeled play outcomes) came from being able to stay close to the receiver (tracking) and do something once the pass arrived (ball skills). While that’s not surprising, what was kind of interesting is that the ball skills attribute was actually more important than tracking: fighting for the ball turned out to be slightly more valuable than separation.
Ella Summers looked at the split between how often players were targeted and how often completions were made against them, noting that there was no correlation between the two in the data. This is interesting, because she comes up with an implication that some players are getting their value from shutting down the route before the pass is even thrown (i.e. lower than average target probability) but that some players perhaps ought to be thrown at more often (i.e. high completion rates against them). This meshes well with what Venzor and Gartenhaus found, and led to a neat proposition by Summers: deliberately put coverage defenders on the field who funnel throws (good tracking, low target probability) toward other defenders who are good at making a play on the ball (good ball skills, low completion rates conditional on being thrown at).
Another special mention goes to Jill Reiner, whose presentation made extensive use of the Los Angeles Rams on offense in the examples. Like Andruzzi, the Subbiah-Chu-Reyers-Wu submission, and several of the others, Reiner does a multiple phase breakout to classify defenders into types: before the throw, after the throw, and closing to the receiver. Is a player good or bad in each phase, and what type does that make them? It’s pretty fun to see Jared Goff and his receivers on the Rams roster being used for demonstrations.
For folks really interested in statistical analysis, these videos are really nice projects showing how clever folks have applied the mathematical tools to distill something practical about a decidedly non-mathematical problem. For those who aren’t into the data and math side, even if you aren’t able to follow every step of the way, these are still cool to watch because they drive home the point that analytics is more than simply rattling off numbers.
There’s a ton of creativity and outside-the-box thinking that goes into modern data work, and it’s gratifying to have good predictive models line up and agree with the eye test. When you see which players do best at the end of the Peng-Richards-Walczak-Werner submission or the Toumi-Furman-Robinson-ElHabr one, the names in the rankings make a lot of sense. A lot of fine thought and work went into these projects, so check them out if you have the time.
Now, on to the rest of today’s Notes:
- Lots of good stuff going on:
Team mascot Roary was on hand for Meijer’s turkey giveaway event at the Corktown Detroit Police Athletic League headquarters: