For the first time since we started this work we’re going to use priors in our rankings. We were driven to this because we wanted to make week-1 predictions, but we’re also hopeful it will boost our performance early in the season.
We are doing this in the most simple, light-handed, and in some ways obviously wrong, way possible – we are simply regressing last year’s end-of-season rankings to the mean. How much we regress (“shrink”) depends on the statistic – some stats lose all predictive value (e.g. scoring defense) and some are stronger than ever relative to the other stats (e.g., rushing defense). How much we regress is based entirely on what history suggests is optimal for predicting the next season’s performance.
While this approach will be fine in the aggregate, we know we are missing many important off-season details for some teams. Some have added or lost key players, suffered new injuries, had coaches suspended, etc. We don’t account for any of this, above and beyond normal year-to-year transition. While omitting these events is imperfect, to be sure, we believe they often receive too much weight in the public’s mind. We’d rather keep our expectations in check and respond to this year’s data as soon as it starts coming in. ‘Sides, how are we supposed to value the suspension of a head coach? 😉