I am a fan of the Freakonomics podcast and they have been doing a series on sports. The latest in the series had a portion talking about data analytics in terms of NFL play calling. You can find the whole pod cast here:
http://freakonomics.com/podcast/think-like-a-winner/. The part I am referring to starts around 33:45
I've copied the relevant portion below. Levitt is
Freakonomics co-author and economic professor at the University of Chicago Steve Levitt:
Levitt has also studied the details of in-game decision-making in professional sports. In a paper he co-wrote with Ken Kovash — a one-time University of Chicago research assistant, now a Cleveland Browns executive — Levitt looked at how good N.F.L. offenses are at confusing or deceiving the defense.
LEVITT: We looked at whether teams are making the right choices between running plays and passing plays, and our conclusion is that we see that on average teams seem to make a mistake.
The mistake is, essentially, being too predictable.
LEVITT: They seem to not throw enough passes, relative to running plays. And N.F.L. teams in particular have lots of serial correlation.
“Serial correlation” meaning the previous play is too predictive of the next play. Ideally, you want the play-calling to appear nearly random, to maximize the element of surprise.
LEVITT: It turns out the play that you just ran has a very large influence on the next choice. And interestingly, and almost embarrassingly funny to an economist, is that it’s exactly when that play did badly. So if you do a running play and it didn’t do well, you are much less likely to call a running play the next time because of some belief that, “Well if it didn’t work last time, it’s not going to work this time.”
There are, of course, other factors to consider: if a run play fails, you’ve got more yardage to make up, which might argue in favor of a pass play. But in the final analysis, Levitt and Kovash found that teams were leaving points on the table.
LEVITT: We think the stakes are really high. The magnitude of the mistakes that they made were associated with scoring an extra point a game, which doesn’t sound like much, but an extra point a game would turn into about half of a win per season, and that’s worth about $5 million a year. So it looks like roughly a $5-million-a-year mistake. Currently in the N.F.L, the number of pass plays is about 58 percent. And our belief is the optimal looks much more like 70 percent. And so there really should be dramatically more passes in the N.F.L.