About The Ratings – Computer Rankings By The Director of Information

The primary objective of these ratings is to minimize Weighted Retrodictive Ranking Violations. I calculate that as the sum of the distances in ranks between teams for any games with a ranking violation, divided by the total number of games played. For example, if Team A defeated Team B, but the ratings have Team A ranked #10 and Team B ranked #7, the difference for that game would be 3. If this sum from all games with ranking violations was 2,000, and 800 games had been played, WRRV as I show it would be 2000/800, or 2.5. This generally aligns with WRRV as calculated by Paul Kislanko, who further incorporates the differences in scores of the violations.

The secondary objective, in cases where the minimized WRRV above has multiple solutions, is to maximize the number of retrodictive wins. If there were 200 games with a ranking violation, out of 800 games played, that secondary Retrodictive Correct Wins metric would be shown as 75.0%.

This approach prefers a solution where there are 3 games with a violation, each with a small difference in team ranks, over 1 game with a violation where there is a big difference in ranks.

The predicted margin of victory for an upcoming game is the difference in ratings plus the home team advantage.

Many thanks to Paul Kislanko for tracking WRRV for each of the college football models in Dr Kenneth Massey’s College Football Composite, Dr. Massey for tracking the ranking violations of those models, and Todd Beck of ThePredictionTracker for tracking the forward prediction accuracy of this and other models in multiple sports.