NFL analytics for dummies (like me)

I know, I know: You’re used to football performance being measured in terms of yards and touchdowns.

Maybe, if you get fancy, you’ll use passer rating or pass-blocking win rates.

But if you’re looking to evaluate the importance of an individual play, the two measurements you need to know are win probability and expected points.

Win probability is a statistic that gets thrown around a lot, but people tend to focus on its predictive power, which is understandable.

The statistic itself is constructed to tell you how likely it is that a given team will win.

I do not find this predictive ability to be all that useful. Getting all concerned about a team having 67-percent chance of winning a game in the first quarter seems kind of silly for a couple of reasons.

1. This means there is a 33 percent chance that team will lose, which means a defeat wouldn’t be all that surprising. It will happen one out of every three times.

2. You don’t have all that much information to go on in the first quarter. The two teams haven’t been playing that long so most of the data used to inform the estimate will be drawn from other games. That doesn’t mean it’s useless, but it does mean it’s not necessarily as applicable to what will happen in this particular game. 

What I love about win probability is not in terms of its predictive ability, but in assessing how much a given play affects the likelihood a team will win. How much does it help – or hurt – a team’s chances of winning.

Now, the source I’m using for win probability is NFLGameData, a site which was built during the pandemic by a group of statisticians that includes my economist friend Ben Baldwin.

This is not because I believe numbersFire Live has the best model, but because it’s the most easily accessible of those I looked at.

There are lots of other places that calculate Win-Probability Added. You can see it on ESPN’s Gamecast for its college and pro-football games.

Expected points is a little more complicated to explain, but actually a better measurement of stand-alone importance of a given play.

The reason is because win probability winds up being heavily influenced by when a play occurs in a game. The later a play occurs in the half, and especially in a game, the more important it will be if only because the opponent has less time remaining to counteract the effect of that play.

But Expected-Points Added (EPA) is not as dependent on time.

Expected Points is a measurement that tells you how likely a team is to score in a specific situation.

If a team has first-and-10 at midfield, its Expected Points would be 2.5. This means that if a team has the ball, first-and-10 at midfield, that drive will produce an average of 2.5 points. This estimate is based on decades of play-by-play results.

Now, Expected Points Added tells you how much an individual play affected the likelihood of one team scoring.

Let’s use two examples from Seattle’s game in Week 2 at New England.

The first one comes in the first half when Seattle had the ball on its own 44, facing second-and-5. Geno Smith then threw a 56-yard touchdown pass to D.K. Metcalf, who ran right by two defensive backs and caught a 56-yard scoring pass.

Expected Points Added = (Result of a play) - (Expected Points before this play)

In this case, the Expected Points Added = (6 points resulting from the TD) - (Expected points for a team facing 2nd-and-5 from its own 44, which is 0.9).

Expected Points Added = 5.1.

OK. But the vast majority of NFL plays don’t result in points. So how do you calculate those?

Well, you use Expected Points in both sides.

Let’s use another play from that same game.

Early in the third quarter, the Seahawks faced fourth-and-1 at the New England 23. The Seahawks handed the ball to Zach Charbonnet, who was stopped for a 1-yard loss, resulting in a turnover on downs.

So how do you determine Expected Points Added?

Well, you’d started by calculating the Expected Points from when Seattle had the ball, fourth-and-1 at the Patriots 23. This is pretty good scoring position as it’s not only a makeable field goal, but a team could choose to go for it — as the Seahawks did — in hopes of scoring a touchdown. Let’s call this EXPECTED POINTS BEFORE, meaning it tells us the Expected Points before Seattl eran the play we’re masuring.

Next, you’d determine the Expected Points for a drive in which a team has first-and-10 at its own 24. You’ll notice this is the situation that the Patriots were in after making the stop. This is the EXPECTED POINTS AFTER.

The Expected Points Added for this play = EXPECTED POINTS AFTER - EXPECTED POINTS BEFORE.

The result was -4.2 EPA for Seattle.