Is the Analytics Revolution Coming to Football?

In the New Republic, Nate Cohn explores the small but growing role of advanced statistics in football. Projects like Football Freakonomics notwithstanding, the NFL isn’t usually thought of as a realm where stats hold all that much sway, in part because the game is so much more of a complex-dynamic system than, say, baseball. Here’s Cohn on one big change fans might notice if more coaches start relying on statistics:

The one place where fans could see analytics at work is in play calling, which also happens to be the place where analytics could impact the average fan’s experience of the game. The numbers suggest, for instance, that teams should be aggressive on fourth down, and that it’s better to go for first down with a lead in a game’s final minutes than to run the ball on third down to run out the clock. Yet even the teams with well-regarded analytics departments, including San Francisco and Baltimore, largely adhere to a conservative and traditional play calling approach: the coaches “just aren’t listening to them yet,” [Brian] Burke says. And the few coaches with a reputation for following the statistics, like New England Patriots coach Bill Belichick, aren’t even close to as aggressive as the numbers would advise.  

If coaches begin to adopt the lessons of advanced football statistics, the changes would be noticeable to even a casual fan: Teams would go for it on fourth down, stop running so much on first down, go for the jugular with a late lead, and take big risks as an underdog in the first quarter. In that sense, statistics might promise more fundamental changes to football than baseball. Fans watching a data-driven baseball manager might not notice any big changes at all, unless they were fans of bunting. 

Steve Levitt will be happy to hear this.

(HT: The Big Picture)


One problem in some of the football quant blogs I've noticed is that with respect to expected yardage value of certain play calls, isn't addressed in terms of the actual game. It's assessed in terms of infinite trials, e.g. "These subset of plays leads to the most yardage per play over a simulation, therefore, these plays should be utilized." But it doesn't address that games are small trial, finite, and thus high variance. That the goal isn't just total yardage or points gained over 1000 snaps, but over ~70 snaps and accounting for opponent and game flow. A pure statistician would say it doesn't matter, but anyone with enough experience with the game has developed heuristics to understand that it does. Of course, it'd help to quantify them so it's just not a case of "this is how we've always done it, and it works."

lemmy caution

Somebody should try this. The owner needs to be on board first though.

Bill Walsh used to script the first 15 offensive plays to probe the defense. He then used the results of these 15 plays to help call the rest of the game. Maybe Walsh was just looking at noise. I doubt it though. teams are different. players have different tenancies.

paul o.

Adding to Brad's point, some of the football quants lump expected yardage not accounting for specific game scenarios. For instance one can expect to see better passing yards on 1st and 10 as opposed to 3rd and 5. It was nice to see Baltimore try to put away the binders with a fake field goal at 3rd and 9.


There are no football quants who confuse the difference between 1st and 10 and 3rd and 5. You are just wrong there. Maybe they have a stat for all passing plays, and that might get quoted. But they also very much break those stats down in the 1st and 10 and 3rd down issues. Everyone knows those are completely different situations.

But yes, the Ravens were the underdog and they tried a risky play in the first quarter. It almost came back and bit them in the but.

Mike MacDonald

Modeling a massively complex dynamic human system with some statistical analyses? What a swell idea! How very economics-like! The statboys should stick to checkers (baseball) and the econoboys should stick to sandboxes (micro). Football and macro should be left to the dynamic human masses - or have you forgotten 2007 already?


Seeing as this model just goes off league average and ignores whether an offense or defense is great or horrible, or weather, or injuries, makes this exercise impractical in terms of putting a coach's face and decision to it.


ESPN is already using some simulated / predictive analytics to compile their QBR rating. Essentially they review each play a quarterback makes and assign points based on how other QBs did from the exact same situation. More points are awarded for critical inside the 20 yard line (redzone) and 3rd or 4th down. A QBR score of 50 means you are an average QB a score of 100 is the best. By way of example perennial Pro Bowler Peyton Manning's rating for 2012 was an 84 and perennial NY media target Mark Sanchez was a 23.

Golf is also using a similar analytic called strokes or putts gained where they compare the result of an individual putt to the field's average. Since most of these guys at the Pro Level are pretty good at Driver thru Wedge shots if you are sinking 10 footers while everyone else is settling for a two putt then you are really creating distance from the field.



Yah, but.....

I am very into sabermetrics. It took decades to even begin to get all the important issues in their like park factors. etc. And the key factor in 80 percent of the statistics there just revolves around mano-a-mano batter vs. pitcher.

In football you have 21 other players involved, the personnel differs on almost every play, and very often key players are performing with some form of nagging injury.

There are just so many ways to pick this NFL stuff apart. Weather is very often a factor in football and rarely a factor in baseball. So many things happen outside a QBs personal control. For instance, about halfway through the seasonm Superbowl champion Joe Flacco had suffered a league low 3.8 percent of pass attempts dropped. Tennessee's Jake Locker had a league leading 11 percent.

Do you go for it on 4th and 4 from the 32 or kick a field goal? Well, what if your second string left tackle is in, and limping, and your QB is almost getting killed by the All Pro defensive end on that side combined with the blitz packages they are throwing that way. So maybe that takes half your playbook away and plus you'd have to keep an RB or FB in to help block on any passing play.

And is your kicker David Akers? Well, the 49ers kicker had an interesting year in 2012. The only kicker to miss more field goals in one season the history of the NFL was -- nobody.

So to tell Bill Belichick that the league average of going for it from the 32 on 4th and 4 is a result of 2.73 points versus a result of 2.54 points on attempting a FG is just gibberish.


Erik Jensen

Yes, the fact that there are 22 people on the field does make it much more difficult to assess individual performance than baseball. Yes, there are subtle differences for each situation that averages don't take into account. But for a millionaire coach to be unaware of even the base percentages (when we have huge amounts of data) is appalling. Mike Tomlin even said that he goes strictly on feel for 4th down decisions.

If coaches were rational (wanting to maximize chances of winning) while taking into account the particulars of the situation, then they would be more aggressive than statistically indicated about the same number of times as they are less aggressive. For example, maybe the defense has a key injury, so it's time to go for it on 4th and 8, statistics be damned. What is found is that they are almost always less aggressive.

I don't think it's even debatable whether or not coaches are maximizing their chances of winning with their 4th down decision making. They are not. The interesting questions are why they choose not do so, and how long it will take before they change.



The only team that would try something like this would be a losing team. Why? Because the cost of failure is high, for a winning coach. There's no reason to risk an untried decision-making system.

Once it's proven to make losing teams better, then it'll be adopted across the league.

That said, what winning coaches should do is try this system in parallel, as a data point to use when decision making. "What does the box say?" should be what people hear on the sideline.