# A Game of Inches

Two posts in one day, you lucky bastards.

Today we mourn the finale of the 2022 CFB season. September seems like just yesterday and after today we will have to wait 230 days until Week 0 for 2023.

This was prompted by a discussion that happened on our Discord and a nudge by one @Y_Soup for more data driven newsletters.

## What I Said

After Jeff posted his earlier article, I said that over the last 10 days since the Fiesta Bowl I have realized that how somebody views that game is a good litmus for how they view the game of football in general.

## What That Meant

TCU made the plays, but if you look at projecting the title game, the reproduce-ability of not 1 but 2 pick sixes, a TD called off the board with a fumble at the half yard line the next play, and another 4th down attempt inside the 5 that is a trick play that gets blown up is about 0.0000000001%.

And then they won by 6.

TCU made a lot of massive plays.

They stuffed the run other than the first play where the LB went for a kill shot and got out of his lane. but they also gave up a ton, especially over the top and needed those 4 plays to win by 6.

They can do it, but just about everything that COULD go their way in the Fiesta Bowl did. And Vegas has reacted accordingly - the money line is currently sitting at Georgia -450, which implies the Dawgs win ~80% of the time.

But the Frogs only need 1.

## Why I'm Writing This

Everybody's heard the phrase "it's a game of inches" so I decided to take a look at our handy-dandy EPA data after somebody mentioned that they found anything related to EPA arousing.

### Expected Points Added

Developed by Virgil Carter when he was playing for Bill Walsh in Cincy, EPA is a measurement of how many points each single play, in theory, should be worth.

The way it's calculated is pretty intuitive - let's say you have 1st and 10 on the 25 yard line. If you look at all (assume 100 for the sake of round numbers) the drives with that exact scenario, and say 15% of them end in a TD, and 20% end in a FG - that means 15/100 got 7 points and 20/100 got 3 points - or rather, that scenario is worth 15% of a TD and 20% of a FG.

15% of 7 = 1.05, plus 0.6 from the 20% of 3 points, and our expected points for the drive is now at 1.65.

Make a play, do the same calculation again and say our new expectation for the drive is 2 points becuase we got 20 yards, that difference (2-1.65 = .35) is our expected points `added`

(or subtracted if there's a penalty, etc).

Note: You can have a negative expectation - for example, if you are punting from your own 1 yard line, the expected result is the next score would be by your opponent kicking a field goal, so you could be snapping the ball where the expected average result is you end up giving up points on the next drive. Similarly, pick sixes will be over 7, because if you go from Michigin +2 points to TCU scoring 7 on the same play, that's a difference of 9.

### My Burning Question

**How many plays typically wins a game?**

TCU won by six. Take away any of the 4 plays mentioned above and they very likely lose. So if we took all the positive plays while you have the ball, and the other teams negative plays (where it was good for your defense) - how many do we have to take off the winner before they lose?

Looking at the Fiesta Bowl, the biggest plays per EPA:

- First Pick Six (-8.94 for Michigan)
- Second Pick Six (-7.25 for Michigan)
- Quention Johnson 78 yard house call (7.18)
- Michigan fumble at the 1 yard line (-7.11)
- TCU fumble at the 27 yard line (-5.91)
- The bomb TD that got called back that happened to set up #4 (+4.4 for Michigan)

You get the idea. You could take off any of the first 4 plays and Michigan likely flips the game result.

### The Answer

5.8 (for FBS vs FBS games in the CFP Era)

Surprisingly small, right? Especially when you consider that including PATs/2pt conversions/kickoffs there's roughly 140-150 chances for recording EPA on a game event in every game.

Winning is really hard. It's gonna come down to less than 5% of the plays. If you filter out the "blowouts" where it was 10+ plays as outliers, it knocks it down to 3.88 plays.

Four plays can be the difference between a win, a loss, a national championship berth for the Ohio State Buckeyes or the agony of defeat.