Open tunesmith opened 4 years ago
The WP model is documented in Section 3.2 here: https://arxiv.org/pdf/1802.00998.pdf
There is also a separate repository for the models: https://github.com/ryurko/nflscrapR-models
Just a guess, but it probably has something to do with the use of expected score differential in these edge cases.
Denver/Chicago, week 2 2019. One second left, 35-yard line, Chicago is down by 1. nflscrapR WP model says Chicago has a 95.5% chance of winning. This is nowhere near the probability of making a field goal at that range. Next Gen Stats says it's a 53% chance (https://twitter.com/NextGenStats/status/1173382032128266240) to make the field goal, plus whatever chance of scoring a touchdown had they gone for it. Is there more information somewhere on how the WP model is calculated?