Coach Aggressivness

The code here is ported over from Ben Baldwin's public repo. His 4th down bot is powered by this model, which I'm using for this analysis. The model itself, which quantifies the value of each decision a coach might take measured in gains or losses in win probability, is xgboost based and is defined in R/_go_for_it_model.R.

My analysis, which is very simple, can be found in analysis.R.

The README below is from Ben's original repo.

Fourth down calculator

This is the repository for the fourth down calculator introduced in this piece on The Athletic. Here are the main files of interest:

The code that powers the Twitter fourth down bot is in this folder here.

Features

  • The go for it model gives probabilities for possibilities of yards gained and includes the possibility of earning a first down via defensive penalty
  • The punt model includes the possibility for getting blocked, returned for a touchdown, or fumbled on the return
  • The field goal model is a simple model of field goal % by distance and roof type

Current limitations

There are some edge cases that are not accounted for. These should only make a marginal difference to the recommendations as they are largely edge cases (e.g. the possibility for a field goal to be blocked and returned).

  • The go for it model does not allow for the possibility of a turnover return. However, long returns are extremely rare: For example, in 2018 and 2019 there were only four defensive touchdowns on plays where teams went for fourth downs out of 1,236 plays, and all of these happened when the game was well in hand for the other team.
  • The punt model doesn’t account for the punter or returner, ignores penalties on returns and ignores the potential for blocked punts to be returned for touchdowns
  • The field goal model doesn’t account for who the kicker is, what the weather is (only relevant for outdoor games), or the possibility of a kick being blocked and returned for a touchdown