Skip to content

Football Analytics

Preparing for the Superbowl Football Analytics

Your team is at midfield, you have the ball, it’s 4th down with 2 yards to go. Should you go for it? (Apologies in advance to our many readers, especially those outside the U.S., who are not aficionados of American football, but it’s Superbowl week in the U.S. A quick guide to the game is here.)

Years ago, the answer would have been no – punting (kicking) the ball away to the opponent was the usual, risk averse practice. Now analytics has taken hold, and, increasingly, teams are going for it (gambling that they can run a play to gain the 2 yards necessary to retain possession of the ball; failure means the opponent takes over possession). It turns out that punting (which returns the ball to the opponent) in that situation has an expected point value of -0.55, while going for it has an expected point value of +.50. What does this mean?

Expected point value (EPV) is calculated by evaluating a large number of situations where a team is faced with 4th and 2 at midfield and tallying what the next score was (i.e. +6 for you, if you end up getting a touchdown, -6 against you, if your opponent ends up getting a touchdown). With all those results in hand, you average them to come up with the EPV of being in that particular situation. Team analytics staff or statistics services are now able to supply that EPV for the average team for a wide variety of situations. Of course teams differ in their offensive and defensive capabilities, but this EPV for the average team can serve as a baseline for decisions.

Another choice that comes up is whether to opt for a 2-point conversion attempt after a touchdown, versus a 1-point attempt. The optimal decision depends on several factors, mainly the score and how much time is left in the game. Touchdowns are worth 6 points, plus any points obtained in the point-after-touchdown conversion: 1 point for a kick, which succeeds 99% of the time, and 2 points for carrying the ball into the end zone, which succeeds 60% of the time for the average team.

Let’s suppose you were down by 7 points against an equally-matched team in the final moments of a game, and have just scored a touchdown (6 points) so are down by 1 point. Should you attempt the kick for 1 point, or try for the 2-point conversion?

If you do the kick, you are pretty well assured of tying the game and sending it into overtime where you have a 50% chance of winning (recall that you are equally-matched). If you do the 2-point attempt, though, you have a 60% chance of gaining the lead and, since it’s the final moments, probably the game as well. So, on average, you should do the 2-point attempt.

Loss Aversion

But this is where the concept of loss aversion comes in. Most coaches are reluctant to throw away the sure trip into overtime when faced with the 40% probability of an immediate loss if you try the 2-point conversion. Objectively, the 60% chance of an immediate win should outweigh the 50% chance of an overtime win. However, psychologically, you already possess the tie, and the perceived pain of losing the tie outweighs the potential gain from the gamble.

Consider the following gamble on the toss of a coin:

  • Heads, you lose $100

  • Tails, you win $150

The expected value of the gamble is positive – if you make it repeatedly, you will come out ahead. Nonetheless, most people will turn down the gamble, since the perceived pain of possibly losing what you already have exceeds your perceived gain from taking the risk. This is loss aversion. Put another way, probabilistic gains are down-valued, compared to definite outcomes.

Player Evaluation

Analytics has its biggest pure impact in decision points like those described above. Identifying a very specific, easily defined scenario (whether to go for it on fourth down) and trolling through a database of thousands of games can yield empirical evidence that may be neither obvious nor intuitive. Analytics, in the sense of analyzing data, is also integrated into evaluating player capabilities. In the player draft, for example, a potential new player coming from college will be assessed according to a raft of metrics (e.g. how quickly they can run 10 yards, how much they can bench press, etc.).

A systematic review of these player capability metrics and comparison to existing NFL players can exert discipline on the player selection process. A general manager’s “gut feeling†that a certain college player is just what they need, and worth a certain amount in draft picks, can be benchmarked against a sizable set of prior data with other similar players and their subsequent performances.

Still, the best general managers are those who have a good intuitive (and perhaps explicit) sense of new players. Paraag Marathe, a senior consultant with Bain, was engaged by the San Francisco 49’ers to advise the team on player acquisition, trades and utilization. He started by examining historical data, to predict player moves in prior years. He was astounded to find that player moves dictated by his analytics were nearly a perfect match for the trades that team president Bill Walsh had actually executed over the years. (Marathe is now Executive Vice President of Football Operations).