Sorensen Rankings of Sports Teams

Soren Sorensen

Since 1991 I have enjoyed developing computer algorithms to rank the performance of sports teams. In particular, I have focused on College Football. Below are a few essential links related to these rankings.

• Ranking of college football teams since 1869
• Current ranking of FBS teams
• Description of ranking systems, in general, and a detailed description of the algorithms I am using (PDF)

Sources for college football data

• James Howell's College Football Historical Results. The main source for historical College Football results used in these rankings for the premier division prior to 2000.
• Peter Wolfe College Football Results for the current season updated weekly. This is the main source for all results used in these rankings from 2000 to 2024.
• Kenneth Massey's Data. The best place on the Web for data on sports results. This is the main source for results used in these rankings starting 2025.It has also been an extremely valuable source since 1997 for information on other computer rankings as well as the AP and Coaches' polls. The final annual (since 1997) result of Massey's College Football Composite rankings are also listed together with my rankings
• Wikipedia: WikiProject College Football contains a wealth of historical data for college football. I am slowly working on incorporating these data for the years prior to 2000 in the rankings.
• Nearly all NCAA and NAIA colleges now have webpages with historical data for most college sports. I am also working on incorporating these data, together with the data from the WikiProject and Howel, into the rankings.

If you find errors or inaccuracies in the data, please do not hesitate to notify me at sorensen@utk.edu The data for all divisions from 2000 should be complete apart from hopefully minor errors in conference and division assignments. Prior to 2000 all Conference and Division assignments are very prelimenary, as well as very limited game information for teams in Div I-AA, II, III and NAIA.

The fundamental assumption in this rating system is that each game (between 2 teams) is considered a measurement of the relative strength (= rating) of the two teams. The ratings only use the following information about each game: a) the identity of the two teams, b) the final score, c) the location and d) the date. All other potential information (team roster, score developmemt during the game etc. etc.) is ignored. Historic games labelled exhibition games (typically very early season games etc.) are not included in the ratings, but are listed as part of each teams record. Following the AP and Coaches polls, bowl games prior to 1965 are considered exhibition games.

Teams are rated accordig to three different methods that are using different parts of the available information:

• 1) Using the score and location information(labelled ScoreDif),
• 2) Only considering the win/loss result of the game (ignoring the actual score) and the location (= potential home field advantage) (labelled WinLoss),
• 3) All information is used, including the date, to optimize the capability of the ratings to predict the score difference in future games (labelled Predict) [not yet implemented as of jan 2026].
• 4) In addition, for all teams a strength of schedule is also calculated (labelled Schedule) as the average ranking (using the ScoreDif ranking) of all the teams opponents.

For more information on these rankings, please see my old (~1993-99) writeup describing the essential assumptions and mathematical algoritms used.

In August 2022 this website was moved from a server in the Department of Physics at the University of Tennessee to sorensen.info (hosted by Hostgator.com).

       Soren Sorensen  -  Updated Mon Aug 10 21:05:40 2026