Move to exclusively using the Gonzalez system to rank teams, but adjust the season to season carry-over, involving reverting to the mean.
I propose that we nix the other three formulas in the ranking system and just go with the Gonzalez formula.
Rationale:
Our current Algorithm is very and explicitly complicated. Each system takes direct reasons people consider important to a team’s performance, then spits out an ordered set. This ordered set is then averaged between the other three ordered sets, to produce a new ordered set which we consider our NCDA Algorithm Average. We use this for Nationals seeds. The problem with this is that is actually provides less predictive energy than some of the better systems, compromising the ideal predictability we seek in assigning fair seeds for the Nationals bracket or even the non-bracket play of Saturday.
The success of a ranking system is determined by how often it predicts the correct outcome, given two teams. This is what we are striving for. I think the Gonzalez system is the best case for this. At current, there are only 171 upsets of 920 ranked matches, resulting in a 81.4% success in predicting the correct winner. I think this average will go up when we adapt the system to account for season-to-season turnover.
The Gonzalez system will simplify our approach to ranking and most importantly setting accurate seeds for Nationals. Using one system, it will be easier and more intuitive to determine what a team needs to do to move into a certain placing. Lastly, the benefits of the other systems carry into the Gonzalez exchanges, but are not diluted or lost when translated into an averaging algorithm.
How Reverting to the Mean works:
This is a solution to the our issue with how teams evolve from the last game of Nationals to the opening match of the next season. Players graduate, players are recruited, rosters change. Currently, the Gonzalez system does not actively account for any teams possible different in rating in the off season. Since the system is reactive, it relies on future match exchanges to move the rating where it needs to be.
My proposal involves reverting a team’s Gonzalez ranking to the mean after Nationals. Teams will retain 75% of their rating (based on how FiveThirtyEight handles their ELO rankings) and the remaining percentage is applied to the League mean rating. Say a (super) team’s rating after Nationals is 55.400 and the League average rating is 40.500.
(.75 * 55.400) + (.25 * 40.500) = 51.675
We can use existing data to fiddle with the percentage that is retained to find a better and a more accurate carry over. FiveThreeEight notes the NFL formula has the mean weighted by a third, so teams retain at least two-thirds of their rating going from season to season.
