📈2023-24 NFL Playoff Predictor🏈
Who has the best odds of winning the Super Bowl?
The following table shows my 2023-24 NFL playoff predictor model. For more info about how it works, read below.
How it works
This NFL playoff predictor model uses team ratings and quarterback adjustments to simulate the rest of the postseason 5,000 times and track how often each team wins the Super Bowl.
The ratings work in a somewhat similar fashion to Pro-Football-Reference’s Simple Rating System, which adjusts a team’s points-per-game differential for its strength of schedule. But there are a few twists in these ratings:
Unlike the SRS, which uses a basic linear regression model, we use a ridge regression to account for the fact that NFL schedules are short and, in the case of quarterbacks, some starters may have only seen limited action during the season. By penalizing ratings that are less certain (shrinking them toward an effect size of zero), ridge regression allows us to avoid large errors for ratings that suffer from multicollinearity.
These ratings are also adjusted for quarterback quality, based on the (schedule-adjusted) Expected Points Added per game by the team’s primary QB in a game. Each team carries a QB EPA rating on offense and defense, also determined by ridge regression, along with a rating for the QB who had the majority of attempts in a given game. When predicting subsequent games, a bonus or penalty is added to a team’s offensive rating if the EPA rating of its current starter differs from the team’s season-long QB EPA rating.
Finally, these ratings are weighted more towards recent results, using the following function for a given game:
| Weight = 0.95^(Weeks ago)
This returns a weight of 1.00 for the most recent week’s games, a weight of 0.95 for last week, and so forth. This has been shown by Wayne Winston’s research to optimize future predictions better than assigning an equal weight to each game. The only other adjustment here is to also downweight Week 18 games if a team was resting its starters; games where both teams rested starters receive a weight of 0.3 (lower than Week 1 of the season), while games with only one team resting get a weight of 0.65 (effectively the same as Week 9) and all other games receive the normal weight.
Together, these factors yield the following ratings and adjustments:
Filed under: NFL, Statgeekery