Lessons in Prediction, Analysis & Deep Learning From The NFL On FOX
Or Why These Guys Should Be Getting Smarter, But They Are Probably Not…
Each week the NFL on FOX pregame ends the same way, a set of predictions by their team of football “analysts”. These former players and coaches — the subject matter experts pick the winners and losers of a subset of the day’s games. Their accuracy is far from heartening and it doesn’t appear to be improving. Likely, they are doing it wrong.
Predicting the Winner of A Subset Of Games Is Easy
The methodology for making predictions like this is simple enough. Build a model. Make a prediction. Score the outcome. Adjust the model. Repeat. If I gave the job to a Wall Street Quant with no subject matter expertise, there would be little contest. If I added a true subject matter expert, we should be seeing near perfect performance by now.
Whoa! If someone could do this perfectly, they would make a killing…
No, that is actually proof that my statement is true. First, I am not saying you could do this for ALL professional games each week. Football games are a complex model of interactions among players, coaches, refs, and the elements with a healthy dose of luck. But statistically speaking, you should be able to nail four games a week with near certainty.
As far as making a killing goes, that is why there is a line. Well, one of two reasons. The line makes it exponentially harder to predict the outcome of a game. But it is also a tell that even this is possible, the line is primarily used to balance the money wagered on either side of a bet. You see Vegas and other establishments are admitting that some people can nail even this, consistently. So they worry more about keeping enough money on the other side of the bet to break even.
Ok smart guy, so why should these guys be getting smarter?
Well they can’t get much dumber… jk. I like watching FOX each week. While I often rant about the overuse of the “analyst” title, these guys do “break down” the games and deserve to wear it, if they chose.
They also engage in some analytic best-practice. They collect data. They interpret it. They measure and report statistics. And they make predictions. This is pure analytic genius — and a recipe for learning.
Even if you believe they are talking mouthpieces for a behind the scenes layer of more serious, less charismatic, and famous football analysts — it is still a great example. In this case, it is the early stages of Deep Learning. The analogy gets better if there are teams of analyst behind them with a hierarchy… but I digress (mildly).
Analyzing the Analysis and Further Digressions with Jillian Barberie
Digressing further, we get a little new perspective. Many of you remember the days of Jillian Barberie’s game day national weather forecasts. Analytically speaking, weather is an important component in the outcome of most games, with the exception of domes. But Jillian gives us deeper insight into what is going on here — entertainment. The NFL on FOX really has no reason to predict every game correctly — in fact, if they did, you might stop watching the game.
They are there to inform and entertain you. Jillian’s role emphasized that. But she was a subtle indicator. To leave zero doubt, FOX provided Kimmel, Caliendo, and Riggle. Not only are these guys the comic relief, they are the lead in to the predictions segment of the show (most of the time).
Stated more analytically, FOX is not optimizing their predictions to find the winners of the days games. They are optimizing on the entertainment value of their segments based on viewership and advertising sales. So perhaps the guys are getting smarter… or at least more entertaining? They are simply playing by a different set of rules. Thanks for reading!
