Each team's rating has a defined relation to the ratings of its games' opponents, except for losses to much-higher-rated teams and wins against much-lower-rated teams. Specifically, any losing opponent with a rating more than one below the team's rating or any winning opponent with a rating more than one above the team's rating is not used in the calculation.

For this description, I'll refer to the wins and losses that are not so discounted as the "relevant wins and losses" and I'll refer to the opponents in these games as the "relevant opponents". If a team plays a relevant opponent two or more times, then I'm counting it as that many opponents.

Each team's rating equals its number of relevant wins, minus its number of relevant losses, plus the sum of the ratings of its relevant opponents, all divided by one more than its number of relevant opponents. Furthermore, as a group, the ratings are centered so that the sum of all the teams' ratings is zero.

I go through repeated calculations to come up with numbers that are as close as I can manage to these criteria.

John Wobus, 9/8/03