Stats Illustrated - Rushing Yards Over Expected

This season, I’ll be taking on a little experiment where I take a play from the most recent Green Bay Packers game each week in order to explain an advanced statistic and how the inputs to that statistic actually appear on the field. Too often, we speak in theory rather than in practice about advanced metrics like EPA and DVOA or concepts like quarterbacks v. offensive linemen as being the primary cause of sacks.

Seeing exactly how these statistics and concepts are relevant to the actual game is a good idea for two primary reasons. The first, and most obvious, is that seeing anything in the actual game makes it easier to understand. The second is that applying a metric to the actual play can lead to a new way of looking at a play-call and help to explain not just its success or failure, but whether it was a good idea in the first place.

Rushing Yards Over Expected On The Field

Today we have just such a case, as we’ll take a look at AJ Dillon’s failed 4th-and-1 run from the shotgun in Sunday’s loss to the Vikings through the lens of NextGenStats’ “Rushing Yards Over Expected.”

You can find up-to-date metrics on RYOE and its cousins, RYOE per attempt and ROE%, here, and an outstanding video explainer that NextGenStats put together here.

The model NGS uses was born in the 2020 Big Data Bowl, wherein contestants were charged with predicting the outcome of running plays based on various inputs available at the start of and throughout the play. Using player tracking data, the winners, Philipp Singer and Dmitry Gordeev, created a model that quantified the effective acceleration of the back, combined with the relative positions and acceleration of each defensive player and offensive blocker, to determine how far the run was likely to go. Grand Finalist Matt Ploenzke came to a similar conclusion with his model, which used over 40 variables. Notably, however:

Among roughly 40 input variables, a ball carrier’s “effective acceleration” was the most important for estimating yards gained on a handoff play.

The most shocking finding in my opinion was that effective acceleration at the time of handoff was a huge predictor of success on the play, and that is as good a place to start as any. Five years ago, if you asked any given football coach, or player, or analyst if they thought acceleration at the time of handoff was crucially important, I suspect they would all tell you “no”. I suspect most would still tell you “no.” But understanding why this strange factor actually is important can tell us a great deal about what kinds of runs are better in certain situations, and as a result, it can lead to calling better plays.

This run to AJ Dillon is ostensibly an option, but given that Aaron Rodgers was probably never keeping the ball, it’s really more like a shotgun draw. Rodgers, to his credit did say after the game that he probably should have kept it as he could have walked into the end zone, but he didn’t. Instead, AJ Dillon gave us a textbook case on the importance of effective acceleration at the time of handoff. Here we see Dillon taking the handoff just outside the 5 yard line from a standing start. Za’Darius Smith (55) comes around the edge virtually unblocked while Eric Kendricks (54) is there to fill the most obvious hole.

Dillon is in a bind. The play is not well blocked, and because he took the handoff from a stationary position, any forward force he generated will be dwarfed by Smith and Kendricks heading downhill with a full head of steam. Dillon decides to try to hit the Kendricks hole, but by the time he does the hole is rapidly closing and safety help in the form of Camryn Bynum (24) has arrived. He doesn’t get there fast enough or with enough leverage. Instead, Kendricks adds his heft to the pile and Smith undercuts Dillon from behind. It was a play emblematic of so many “shotgun draw” short yardage runs that it’s a wonder that they ever happen. NextGen doesn’t make the expected rushing yards for every play publicly available, but I suspect on this play it’s 0 yards. To see why, let’s look at what the model sees.

First, all of the Vikings are headed the right way, and have gotten push on their Packers. More than anything it sees Kendricks and Smith, with plenty of speed, plugging the most plausible holes and sandwiching Dillon in between. The NextGen player tracking data is good at identifying which way, and how fast everyone is moving, and in this case, it will see the Vikings winning the play, heading towards Dillon with decent velocity. Flipping to the Packers:

Dillon is, at the point of handoff, not moving nearly as quickly as his pursuers. They are gaining on him with a significant speed advantage. If he accelerates into the Kendricks hole, Kendricks will have a force advantage. Dillon cannot plausibly bust this outside. The Viking defensive linemen have outside leverage and a well-set edge. Bynum is also crashing down should Dillon get any ideas, and as soon as he turns laterally, he’ll be toast. He also can’t cut back because he’s a sitting duck with no velocity and limited acceleration, while Smith is already a heat-seeking missile.

It’s hard to hit moving targets, and easy to hit still ones. “Effective acceleration” is the simple encapsulation of this idea, quantified, and Dillon is a stationary target for far too much of this play. In addition to the physical problem of being stationary while everyone else is moving, we actually get one additional piece of information from acceleration at handoff, that being whether the back has a clear idea of where he wants to go. On a standard, well-blocked run, the back’s acceleration will be higher when the hole is obvious, and less so when he has to make a decision on where to take the run. Acceleration is therefore not just a physical stat — it also serves as a proxy for how well the play is blocked. You can see on this play that Dillon hesitates for just a moment before crashing into the pile. This happens because the obvious holes are rapidly closing, and he’s forced to settle for the best available option.

Don’t Run Out of Shotgun on Fourth and Short Near The End Zone!

There can be good reason to run out of a spread formation like this on 3rd or 4th and short. If the defense is using lighter personnel, you may simply power your way to a first down upfront. With the legitimate threat of a pass, you may even break a bigger run as the receivers carry their men downfield. However, on the goal line, with the field compressed as much as it possibly can be and with the threat of a run more dangerous than the threat of a pass, that logic breaks down. And sure, Aaron could have pulled this out and run it in (and should have), but him not being willing to do so means there is little to no benefit from running this type of play. Options without the actual threat of an option don’t work, especially when Za’Darius Smith can tell.

If Dillon has a full head of steam, and hits the Eric Kendricks hole a half second earlier, he probably scores. Heck, it’s AJ Dillon, if he can get himself singled on almost any defender with a full head of steam, he’s going to score! He still almost scored here even though he was a sitting duck. Here’s a standard under-center run from earlier in the drive where Dillon gained 3ish yards. You can see the difference his momentum makes. It’s not a great run, but when you only need a yard, a little push goes a long way.

Anyway, the lesson from the Big Data Bowl, and NextGenStats, is that generally speaking, running plays where the back is gaining speed rapidly at handoff are better than anything that looks like a draw play. If you’re going to run some type of draw or option, the strategic factors around size of opposing personnel, your QB’s ability to run, and position on the field need to weigh heavily in your favor. Otherwise, there’s a good chance your back is going to lack push at the point of contact, and a high probability that contact occurs well behind the line of scrimmage. It’s exactly what you do not want in short yardage.