Riding the Wave of Analytics

Everybody expects the NFL to be changed the way baseball was changed by Moneyball, but as Sam Monson points out, football's analytics revolution is more subtle

| 4 years ago

Riding the Wave of Analytics

The NFL data landscape is changing. People want more information than ever and everybody is scrambling to keep up with the demand. No longer are fans satisfied with sitting back and letting a few talking heads tell them a few banal platitudes, they want hard facts, information, and real knowledge.

(7:40) K.Kolb pass deep left to L. Fitzgerald for 37 yards, TOUCHDOWN.

This is the typical information that an NFL game book gives you about a given play. By the end of a game the box score just tots up the various totals and averages.

At one point box scores were the best people could manage from some games, maybe the odd highlight, but do those numbers really tell you anything?


They can’t tell you what kind of coverage the quarterback faced, whether he was throwing in the face of pressure, how tight the coverage was, who was in coverage, whether somebody in the secondary bit on a play action fake and took themselves out of the play entirely. While other sports can be distilled down into a series of repeatable scenarios, football is far more complicated than that. Every play is the culmination of 11 guys on offense trying to get one over on 11 guys on defense, and the number of times the exact same scenario repeats is minimal.

Pro Football Focus really isn’t a stats company.  The site began as a way to really evaluate player performance beyond what talking heads on TV try to feed people, or the farce that has become the Pro-Bowl vote. We have a whole host of unique data, but we record it almost incidentally to the main focus of the site – player performance and analysis.

People love to use statistics to prove their point, but the truth is numbers lie all the time, and at the very least all numbers can only tell part of the story. A sack can come as a result of JJ Watt beating a double team, tossing aside the running back and bringing down the quarterback within three seconds of the snap. It can also come as a result of the quarterback holding onto the ball for five seconds in a clean pocket and then falling over as he tries to escape towards the sideline. On the stat sheet those plays look exactly the same, but in reality everybody knows they are dramatically different.

Even the various advanced formulas and metrics we can create with PFF stats can lie at times. The Elusive Rating will always underrate players with the speed to burn defenders without needing to make them miss or break a tackle. Pass Rushing Productivity doesn’t take into account how much help a player had or how quickly the pressure came, neither does Pass Blocking Efficiency.  There is no perfect stat, advanced or otherwise.

The moving parts in football are so multiple and unpredictable that it becomes virtually impossible to tell the real story with numbers alone. The only way to know for sure is to watch and analyze everything for yourself. As Raiders GM Reggie McKenzie put it “You can never dispel what you see on tape.”

We are the only people out there watching and grading every single player on every single snap in the NFL season. By the time each week of games has been accounted for this coming season each game will have been through in excess of 24-man hours of work. That is 16 full days worth of tape watching and analysis for a non-bye week. The amount of work involved is so vast that NFL teams are coming to us for the data rather than attempt the undertaking themselves.

In order to get an accurate idea of exactly what happened and who is playing well or badly, tape really is the only option. Moneyball for football simply doesn’t work the way it did for baseball.

Why not? The numbers just don’t stack up.

A typical baseball season has 677,000 simple plays with which to collect data and analyze numbers. With a 16-game schedule the typical NFL season has just 43,000 extremely complex plays with dozens of factors affecting each one. Clawing through the white noise to find meaningful data is just not viable, the information is too distorted with too many things affecting any single play.

That’s not to say that there is nowhere that statistics and sabremetrics can improve a football team, but I don’t believe that you can accurately scout using numbers alone the way it is possible to do in baseball. The cutting edge in the NFL won’t come from analyzing numbers in isolation, and teams know that.

The franchises that are able to gain an advantage from the new influx of data are those that are best able to marry the old with the new.

Remember the old curmudgeon scouts from the movie Moneyball? The guys that scoffed at the numbers and math telling them which players they should be targeting while their trained eyes told them something else entirely? They were swept away by the Moneyball movement, rendered an anachronism by a game that moved on without them. Now people expect the same thing to happen in football, with some treating those that refuse to accept the inevitable truth with the same scorn poured over the baseball scouts before them.

But the revolution in the NFL is going to have to incorporate those guys, not cast them aside like Betamax, Minidisc or HDDVDs; a failed technology consigned to the history books as the world moved onwards and upwards. Numbers can add context, and places like PFF can provide yet more with grades from tape, but the league will still need traditional scouts and coaches to use that data and exploit it.

The teams that are at the forefront of working out how to incorporate all of the PFF grading and data are the ones that are seeing the benefit from it already. NFL teams have already used our grading to assess their own coaches and the grades they give out to their own players internally, improving the quality of the feedback they give in-house by providing an independent check to the inbuilt bias their coaches may have had in the past.

Other teams have used the data PFF collects to analyze exactly how blockers surrender their pressure. We can tell them whether a player is regularly beaten to the outside, inside or via bullrush, and they can then go away with a cutup of those plays and work out why, and if it’s something they can exploit, or fix if it’s a player on their team. It’s something they may have spotted themselves using the old traditional scouting methods, but PFF is able to instantly draw attention to it from the numbers. And grades.

The NFL is modernizing, and the wave of analytics has come upon the league, but unlike in baseball where that wave swept away the old guard, the best NFL teams are learning to ride that wave, harnessing its power and using it to improve their practices, not destroy them completely.

Follow Sam on Twitter: @PFF_Sam

| Senior Analyst

Sam is a Senior Analyst at Pro Football Focus, as well as a contributor to ESPN and NBCSports.

  • Jimmy

    Chilo Rachal. That was some fantastic analysis, explain that one. Did he at least cut you guys a cheque?

    • PFF_ColeSchultz

      If he did, it must not have been for very much. He graded out at -12.1 last year in 8 games for the Bears.

  • Chris B

    Love this stuff. Wish I could join the team and get paid to do this kind of stuff. Love it.

  • Jake B

    I don’t think I’ve seen screen grabs with the grade for that play on them before…hope to see more of that as part of in-season analysis this year.

  • MrOodlesnNoodle

    I would love to edit videos and upload for you guys. Do you guys have a Youtube account? I can’t do every team of course but I can record and upload analysis of the Chicago Bears upcoming season

  • Boyfromoz

    I think its a great idea to show a graded play as an example. As a part B could you talk through what each player did to explain why they graded as shown in the diagram?

  • BruisingChargers

    Very insightful. Thanks for the great read!

  • Sam

    I think that the pass-blocking efficiency rating should incorporate time into the equation. Divide the rating by the number of seconds it took for the defender to beat the block. Kind of like UZR in baseball. It would have to be more complicated than that, obviously, and unfortunately I don’t have the knowledge of math required to make an equation, but I think that it would make the stat significantly more accurate.