Developed by TeamRankings.com, the Similar Games Model uses data driven algorithms to identify college football games from the recent past that featured statistically similar teams facing each other under similar matchup circumstances. The results of these historical games drive a variety of predictions including win odds, ATS cover odds, margin of victory, and "fair" money lines.

After pioneering this model successfully in college basketball and seeing good results with football back testing, we launched the college football model on our web site this season. In total, the model went 514-169 (75.3%) picking straight up winners during the regular season and 354-317-12 (52.8%) against the spread. However, with this model it is important to recognize the relative confidence in each pick implied by the projected odds. Less than 55% odds are typically not meaningful.

For example, over the last five weeks of the season, teams with at least 55% projected cover odds have gone 67-52 (56.3%) against the spread, better than overall ATS prediction performance.

Strengths: This model incorporates a range of power ratings and team stats as well as several contextual factors including Vegas line implications, travel distances, and game timing.

Weaknesses: The model still does not explicitly consider several difficult-to-model factors such as projected weather, recent injuries, or days rest. If you feel one of those factors may have a material impact on the outcome of a given game, it may be wise to apply subjective tweaks to predictions.

Developed by TeamRankings.com, this Predictive Power Ratings Model iteratively analyzes data on every college football team and game result in the FBS so far this season. In the end, each team receives a simple numerical rating (e.g. 52.4, 31.0, 84.7).

By comparing the predictive ratings of any two teams, you can determine projections for the game winner and the expected margin of victory if those teams played one another. This process is simple; just subtract the higher rating from the lower rating, adjust for home advantage if applicable, and the result is the expected margin of victory expressed in points. (We also have a formula that translates the differences in predictive ratings between two teams into win odds for each team.)

We have confidence in the validity of the algorithms because if you went back and recreated the season retroactively using the final numerical team ratings and prediction methods described above, every team in the FBS would have margin of victory performance equal or very close to how it actually ended up.

Strengths and Weaknesses: Predictive ratings are relatively abstract; they are driven by margins of victory and scoring. They have no idea who plays for what team, how far a team has traveled, or if it likes to run or pass. All that matters is the end result: how many points does a team score and how many points have its opponents scored, adjusted for opponent strength and home advantage. Their predictive accuracy tends to improve as more games are played, especially across conferences.

Adjusted scoring margin (ASM) calculations measure teams based on whether they score more or fewer points than their opponents on average give up, and vice versa. Several practitioners apply ASM to college basketball, and here we apply our version of the statistic to the college football bowl teams.

Every team has both an offensive and defensive ASM. For example, imagine that Team A plays 10 teams in 2008 (the 'Opponents'). Coincidentally, each of the Opponents allows 25 points per game on average. However, when Team A plays the Opponents, it scores 30 points against every one of them. Therefore, Team A's offensive ASM is +5. Team A scores five points more than its opponents give up.

Likewise, if each of the Opponents averages scoring 30 points a game, but every single one manages just 20 points against Team A, then Team A's defensive ASM is +10. (For consistency, we express good ASM's as positive numbers, and bad ones as negative numbers.) When two teams play each other, we can compare their respective ASM's to determine predictions for a game's expected win margin and final score.

Strengths: ASM is a relative measure, while looking at absolute statistics often can be misleading. Giving up 50 points/game may sound bad, but not if the opponents a team has played average 60 points/game.

Weaknesses: ASM is far less meaningful when comparing two teams with large differences in schedule strength or conference strength. (Allowing 35 points/game in the Sun Belt is a lot worse than allowing the same amount in the Big 12.) ASM's accuracy improves as more games are played but college football schedules are short. We have not calculated overall ASM prediction accuracy for the 2008 season.