How to read the 4th down numbers

Every 4th down recommendation on this site comes down to one comparison: the offense’s chance of winning the game if they go for it, against their chance if they kick. Not the chance of converting. The chance of winning. Those are different questions, and the difference is where most of the arguments start.

Conversion is not the point

A team facing 4th and 2 at midfield might convert around 60 percent of the time. That number alone tells you nothing about whether to go. Failing at midfield hands the opponent a short field, and punting hands them a long one, and the cost of each depends on the score, the clock, and how many possessions are left. The model’s job is to fold all of that into two win probabilities and report which one is bigger.

The margin is the story

When the two numbers are close, say within a point of win probability, the call is a coin flip and the recommendation is closer to a shrug than a verdict. Coaches who kick in those spots are not making mistakes. The interesting calls are the ones where the gap is two, three, four points of win probability and the sideline still sends out the punt unit. Those are the decisions this site tracks.

So when you see a recommendation here, read the margin before the verdict. A “go” by 0.3 points and a “go” by 3 points are different claims, and only one of them is worth being loud about.

What the record will mean

Once the model is live, the sidebar will keep a record of graded calls: situations where the coach’s choice differed from the model’s recommendation by at least 2 points of win probability. Each one is graded by what the choice cost, the win probability the coach gave up at the moment of the decision, not by what happened next. A go that fails and a go that converts were the same decision, and a punt that works out was still the worse bet. Single plays are noisy, and grading the decision instead of the result keeps that noise out of the record.

How well the model’s win probabilities match what actually happens is a separate question, and it will get its own season-level check rather than a play-by-play scorecard.

The calculator itself is in progress. When it goes live, you will be able to set any situation, down to the timeout count, and see the win probability of going for it, punting and kicking a field goal, and the margin between them.