Week 17, Ravens at Bengals
4th-and-19, 9:30 remaining in 2nd quarter, Down by 3
The Bengals punted on a 4th-and-19 on their 32. The NYT 4th Down Bot signs off on that.
If you disagree

The coach and I agree that punting is the thing to do here, but you may be thinking it makes sense to go for it. That would be the right call if you think the Bengals’ chances of converting on fourth down are greater than 52 percent. But based on my analysis, I’d give the Bengals only a 10 percent chance to get a first down here.

Here's the full breakdown of my calculations:

Option Chance of converting
Chance of winning
Before play
After play Change
Punt 59% 59% -
Go for it4th and 19, own 32 10% 59% 52% –7%
Field goal tryA really, really long kick <1% 59% 48% –11%
My decision in context

Along with some circuitry to come up with a win probability for every game situation, all you need to figure out what you should do next is an estimate of how likely you are to make a field goal or convert a first down.

My estimates for these are based on the results of thousands of similar plays, but you may think you're smarter than I am. This chart shows you how changing those estimates would change my recommendation.

What to do on 4th-and-19 on own 32
Down by 3 with 9:30 remaining in the 2nd quarter
What coaches usually do
Punt >99% of the time
Go for it -
Field goal try -
Based on about 2,601 fourth downs in similar situations since 2001.
What happened

Kevin Huber punts for 38 yards to Bal30. Kevin Huber punts for 38 yards to Bal30.

Where did these numbers come from?

To estimate a team’s chances of winning, I use a mathematical model that accounts for a whole lot of variables — including the difference in score, the time remaining in the game, and the number of timeouts each team has left. On top of that, I have models for the likelihood that a team makes a field goal and the likelihood that it will convert a first down.

By combining all of this information, I can come up with the best decision a team can make, according to math.

If you want even more details about the numbers behind my decisions, my full model is available on GitHub. Help make me better!