How to Include 1AA Results

© Copyright 2007, Paul Kislanko

28 September, 2007
The mistake of allowing games against teams from the Football Championship Subdivision (who're we kidding - 1AA) every year with respect to bowl eligibility has caused problems with respect to ranking Football Bowl Subdivision (ok, 1A) teams. As I noted last year, even though 12 games is the correct number for a 120-team field, so many 1-A teams use the 12th game to schedule a 1-AA opponent that the connectivity of the field is less than it was with 11 games and the "count one game vs a 1AA team every four years" rule.

For rating systems that rank FBS (Division 1A) teams, there are basically three options for addressing games against teams outside of the FBS/1A field:

  1. Ignore those games when calculating the rating
  2. Treat all games involving a FCS team as if they were against the same FCS team
  3. Include all division 1 teams in the rating

Historically, the correct approach for the ratings I publish has been 1 - ignore games against teams that aren't being rated. For the most part, those do not contribute useful information because there's usually not a connection from the 1AA opponent to another 1A team, and we're comparing all pairs of 1A-vs-1A teams.

Some ratings adopt the second approach, and for formulaic systems it might be appropriate. But in iterations over games this approach skews the results because it makes the field appear to be more "connected" than it really is.

The third approach can be made to work for systems that explicitly take into account the degrees of connectivity. When this approach is taken with iterative systems such as the ISR, however, wins by teams that haven't played a good team tend to be over-emphasized.

Alas, as Wes Colley (whose system also ignored games against teams outside the 1A field) notes, because of the "wins over 1AA teams can count towards bowl eligibility every year" rule there are now too many such games to ignore entirely.

Topology to the Rescue

Since there are now too many inter-divisional games to ignore some means of accounting for them is necessary. It is simplest to just include all FCS teams in the ranking, but that changes the schedule topology in a negative way. Including all 242 teams and all games between division 1 opponents skews the convergence because there are so many 1-AA teams that do not contribute to the comparisons.

The approach I've taken is to define a teamset that includes all teams that have at least one Bowl Subdivision team as an opponent or opponent's opponent. This of course includes all Bowl Subdivision teams and any Championship Subdivision team that plays a Bowl Subdivision team or another team that did.

The ratings are then calculated using only games between teams that are included in the teamset. FCS teams games against other teams not in the teamset do not contribute to the comparison and negatively affect convergence, so they are ignored.

There's still a bit of a problem at the end of the season - the FCS teams will not have enough games in the mix for their ratings to be meaningful except in forming ratings for the FBS teams. So, even though these 1-AA teams will be assigned a rating by the algorithm, we do not report a ranking for those teams. As a result, the Iterative Strength Rating and Iterative Strength of Victory reports will still list only "1-A" teams.