Ingale's Recursive Power Rankings
My ranking system is an iterative method that uses the results of games in any league to rank teams based on
on-the-field performance. The system uses three pieces of data about each contest to compute this:
- The two participating teams
- The location
- The winner
This system was created with a few design constraints in mind. Specifically:
- The score is ignored. There is too much variability in scores to draw too many
conclusions from the final score of a contest. Specifically, teams don't play to win by a certain amount but
to just win. If a football team is beating its opponent late in the game and has the ball, the winning team
is primarily interested in running out the clock rather than trying to score again.
- Teams aren't punished for beating weak teams. The big boys like their cupcakes. Many
rankings systems are built such that a powerhouse beating a very week team sees their rating drop. While I chide
the powerhouse for their choice of schedule, that doesn't mean that they are a worse team for it. In my ranking
system, a team is never punished for beating a cupcake. The positive effect might be negligible (or even zero),
but it won't drop. Likewise, teams aren't punished for losing to powerhouses. The fact that you lost to a powerhouse
proves very little, so the effect is small.
My rankings systems is a recursive method. Teams are initially rated based on their winning percentage, and a general
homefield advangate coefficient is calculated based on the percentage of games won by the home team. The algorithm iteratively
updates each team's rating and the homefield coefficient as it factors the strengths of each team's opponents. As the
process repeats, the team ratings and homefiel coefficient settle to their final numbers. Ties are treated as half-wins and
half-losses for both participating teams. The homefield coefficient shown for each season reflects the strength of
homefield advantage for that season and is to be used by multiplying by the home team's rating for that game. For
instance if a team rated 0.640 is hosting a team rated 0.800 and the homefield advantage coefficient for that season
is 1.250, the teams are basically equal, as 0.640*1.250 = 0.800.
I've computed rankings for each NCAA football season going back to 1869 and each NFL season going back to 1940.
As a sidenote (and before I get berated), the best way to crown a champion is on the field, and I do not dispute
that in any way. In fact, one should be wary of anyone that claims anything different.
At the same time, NCAA
football currently has roughly 125 teams that each play only 12 games, so it must always rely on off-the-field
analysis to crown its national champion. The place where these rankings excel is that they don't know the
difference between an Alabama jersey and a Boise State jersey -- just who they beat.
NCAA football
National Football League
National Hockey League
Credits
I did *NOT* compile the game data myself and I want to give credit where credit is due.
|