For years we have been relying on points and yards to tell us how effective football teams are, but every fan knows that those numbers can be deceiving.

For example, an offense could rack up a bunch of yards but end their drives settling for field goals or turning the ball over. This same offense could also score a bunch of points but commit costly turnovers that directly lead to points being scored on their defense.

Fans and analysts try to take all these factors into account when evaluating offenses, but they are forced to do so by using multiple metrics that are attempting to describe the same thing but don’t tell the whole story.

A more comprehensive stat for offenses would instead describe how effective the team is at moving the ball down the field, picking up first downs when necessary, and either putting the ball in the end zone or pinning the opposing team back towards their own goal line.

Thankfully, our Expected Points model allows us to do just that, and by the end of this post you’ll never look at yards and points the same way again.

Expected points are the average number of points that will be scored next given the current game situation.

The Game situation is exactly what it sounds like: the down,
yards to go, field position, current score differential, and time remaining in
the game. As fans we all know that 4^{th} & 1 at the goal line,
down by 6 and 3 seconds on the clock is vastly different from 1^{st}
& 10 from the 20 at the start of a game, so our model evaluates those
situations differently.

This seemingly complicated model we keep talking about is actually a pretty simple concept that just uses some advanced statistical methods to get the answer we expect. The model basically just takes the current game situation as the input and outputs the average points scored next based on over a decades worth of NFL play-by-play data.

This would be a good time to clarify what scoring next actually means: it is the next points scored by either team. So if the team with the ball scores on the current possession that would be 7 points scored next, but if they had to punt and the opponent eventually scored a touchdown it would be -7. If no one scored before the end of a half then the next points would be 0.

An example should help clarify this; if we
wanted to know the expected points for 1^{st} & 10 for every field
position we could simply take the average next points scored in all those
situations as so:

Hmmm… This data makes intuitive sense; as you get closer to the goal line the average points scored next increases, and increases even more rapidly 10 yards away from the end zone. You can also easily see the randomness of the data preventing a perfectly smooth line that would describe the true relationship.

Our model simply takes this historical data and creates a
function that will allow us to get a smoother expected points curve for
literally any possible game situation. Check it out for the 1^{st}
& 10 situations we just talked about:

Our expected points model is the most comprehensive and accurate measure of any football situation, and it is what separates our stats from any other stats you can find out there.

Once you understand that expected points is just a single number that describes a particular game situation, it is easy to derive any of our production stats: Productivity = Expected points after a play – Expected points before, essentially how much did the play change the state of the game.