Did Jimbo make some mistake-os?
Evaluating fourth down decisions from the Texas A&M-Alabama game.
As I was watching the Texas A&M-Alabama game on Saturday, there were two fourth down decisions in the fourth quarter that caught my attention.
Normally I would have checked one of the fourth down decision bots that tweet out the win probabilities of each fourth down decision, but sadly, those bots do not seem to exist anymore. However, there is a nice tool for the NFL that allows the user to input the game state and returns win probabilities if the offense goes for it, kicks a field goal, or punts.
Now, this is an NFL tool, and it requires the user to input the teams that are playing, so I took the pregame spread of 2.5 points in favor of Alabama and chose the 2022 Chiefs to represent Alabama and the 2022 Bengals to represent Texas A&M as the Chiefs were 2.5 points better in ESPN’s FPI in 2022. Reasonable people can disagree on this choice, but I imagine running this for a few different teams would produce close results.
Now to the decisions made by Texas A&M.
4th & 5 on Alabama 31, 9:06 left in the 4th quarter, down by 7. Decision: field goal attempt.
Watching this one live, my thought was that a field goal does not help a ton in this situation. You are fairly deep in opponent territory, five yards is not an insane amount to get, and being down four isn’t much better since it still requires a later TD. The field goal in this case ended up being blocked, but that is not important for the evaluation of this decision.
Now to the model results, remember that the Bengals are A&M here.
This field goal attempt is actually not as big of a mistake as I thought it would be. College kickers are probably slightly less likely to make this kick than NFL kickers (so this win probability change is probably a lower bound), but ignoring that, it is only about a 1% increase in win probability to go for it. The model does not output to the user how the variables are impacting the result, but I would imagine the time left in the game here is the big one impacting a field goal attempt not being that bad of an idea.
The second decision by A&M:
4th & goal on Alabama 2, 2:14 left in the 4th quarter, down 9. Decision: field goal attempt.
On the two yard line! Two score game!
This decision I was pretty sure was incorrect when watching, the reason being that although a field goal cuts it to a one score game, you still need a TD, time is low, and you are two yards away from six points. If you do not make it, your opponent is backed up in their end zone to start the next possession. But let’s see what the model says:
Going for it increases your win probability by 6.5% in this situation, a huge difference. Now, one thing this demonstrated that was mentioned on the Wharton Moneyball podcast a few weeks ago is that coaches might have a tendency to make a decision that prolongs their chance of winning a game, even if it decreases their overall probability of winning the game.
As seen here, kicking a field goal and making it results in a win probability of about 4%. Low for sure, but you know you are in a one score game and it “feels” like there is a chance you can win, because you can imagine a situation where you recover an onside kick, score a TD, and win the game. Compare this to going for it and not getting it — you now have a 2% win probability, which is not that much different than making a field goal, but to a coach a two score game with two minutes to go and not having the ball probably feels like a 0% chance. Jimbo is prolonging his chance of winning the game (keeping it above 0%, at least in perception) at the cost of decreasing his overall win probability.
Checking your own in-the-moment thoughts on fourth down decisions is a fun exercise to calibrate how good your mental model of these decisions is. I thought there were two clear (+3% or so) win probability swings had A&M gone for it on fourth down, but it turns out the first was really not a tremendous difference. As for the second decision, I don’t think there is a lot of good justification for getting that one wrong.
—Drew