There are no authoritatively correct rankings, period. It takes a while to understand this, but after enough study of college football (or any competition), it becomes clear that any team really can win on any given day, and winning a dozen games does not prove a team the "best." There is a tremendous amount of luck involved in football. Twelve games is not a large enough sample to accurately evaluate teams - and that is what makes college football rankings an amazing challenge, and one of the most interesting phenomena in all of sports, in my opinion.

That said, I have devoted much time to studying scores over the history of the game. I crafted the system trying to come up with what seemed the best ratings for all seasons. If any team appeared far out of place in any season, I went back to the drawing board and questioned everything. It was a long, frustrating process, and in the end a satisfying experience. I think the ratings are now as close to "right" as can be found, but others must be the judges of that.

The goal here is twofold: high quality predictions during the season, and ratings that make sense to most fans at the end of each season. Explaining in detail how this is done would bore most people to tears, so here's a quick summary...

Quick summary:

1. Scores, locations, and dates of games are the data input.
2. Home field advantage adjustments are applied.
3. Power rating values are on a game point scale. One may make approximate predictions by subtracting a team's power rating from that of its opponent, and then adding 3 points to the home team (4 for basketball), if applicable. However, this is NOT how the official predictions are made. Predictions made by subtracting power rating values should be considered rough estimates.
4. The official predictions come from a complex statistical analysis.
5. Pre-season ratings are mixed into the numbers until about the mid-point of the season.
6. Quoted W-L-T records include only games played against other teams involved in the ratings. If your team is missing a game on its record, it is because the game was against a minor opponent that is not recorded as playing any other team in the rankings that year.
7. For 1977 and later, only Division 1A teams are listed. However, many non-1A opponents are included in the calculations.
8. On forfeits: If the game was completed, it is used in the calculations with the score that was recorded. If the game was not completed, then it is completely ignored.

Regarding Point 3 above: This is a recent change (July, 2007). The power rating values used to be more abstract, having meanings related to probabilities of winning. However, I believe most prefer a point scale that allows for predictions by differences. The ratings still work the same way, but I have converted the old power rating values to this new point scale. In case anyone was confused by the change...

Random Bits:

Predictive systems are not primarily concerned with determining which teams deserve the glory of a "championship." Rather, the goal is to guess at which team would be more likely to win an as yet unplayed game. Therefore, no offense should ever be taken over a ranking position.

Some of the recent rankings that cause the most headaches are found in 1980, 1997, 1998, and 2002. Those seasons each had at least one major team go undefeated and untied without being dominant in many of their games. The teams are Georgia in 1980, Michigan in 1997, Tennessee in 1998, and Ohio State in 2002. Some of these teams appear ranked as far down as #8 by various ranking systems I have checked over the years. Surely these teams deserve recognition and glory, and yet many mathematical systems fail to give them at least reasonable rankings. I do not think any of these teams should be ranked lower than #5. Happily, Entropy puts two of them at #1 (Tennessee '98 and Ohio State '02). This is a sign that the method is a little different than most predictive systems. For more on these teams, click here.

I enjoy a good argument about which team was better than which. To that end, I have worked hard to develop a system that would produce the fewest possible complaints. I want everyone to agree with the lists, but when a disagreement is voiced, I hope for the rankings to be at least reasonably defensible. I invite all to submit any major problems for consideration and debate!

I named the algorithm Bow Down to Entropy because entropy is a concept deeply related to change. Nothing is permanent or easy to pin down. I love the notion that anything can happen, and to appreciate it, I like to highlight this game. Then I encourage everyone to learn about the breakdown of transitive logic. These two short pieces will tell you pretty much everything you need to know - completely "accurate" rankings are impossible, and that's why I like doing this.

Few rankings experts completely divulge their method, and for good reason. It is a lot of work to put together an interesting system. Therefore, I believe the best philosophy is to ask visitors to consider the ratings for many past seasons, and judge for themselves which system does the best job.

Every fan has their own conceptions on how teams should be ranked. Therefore, I am working on a rankings litmus test that will hopefully have standards acceptable to almost everyone.

If these rankings can spotlight some teams that did not get the attention and glory some fans believed they deserved, I'm happy. On the other hand, if the rankings appear to slight some other teams, at least those fans can take to heart my message that we don't really know anything for sure. I happily admit that my rankings are imperfect, but in return I ask that everyone acknowledge their teams are not perfect!

The Entropy CFPOOL pick orderings are determined by estimated odds of winning, not predicted margins of victory. Many assume that if Team A is predicted to win by 10, while Team B is predicted to win by 9, that Team A is more likely to win their game. However, that is not always true.

Enjoy, and feel free to send feedback and questions...

main page