For each game, 1 "game point" is disputed. The winning team gets the largest part, and the loser gets the rest. The formula is:
(Note: the 11 constant is a typical value for basketball; for other sports it may be smaller).
As an example, here are sample gp values for several margins of victory.
where Ra is the team's rating and Rb is the opponent's rating. Basicly, we gives the percentage of game points that a team should win against a given opponent.
A team's total we and gp can be obtained by summing over each of its games. The new rating is calculated by
where Rn is the new rating, Ro is the original rating, and K is an constant that determines the speed of convergence.
If a team's overall performance is better than expected (i.e. gp > we), then its rating will improve (notice that this does not necessarily require winning!). Conversely, the opponents' ratings will decline. Each team begins with the same initial rating (1000). With each iteration these ratings will adjust themselves until equilibrium is reached.
The published ratings also include a homefield correction variable that likewise reaches an equilibrium. The modified formula is:
As it stands, the formula for we should approximate the expected game points (gp). However, it would be more interesting to instead estimate the probability of victory. This is accomplished by replacing 400 with an appropriate prediction constant (k), which will vary among the various sports. As an example, assume that k = 300, Ra = 1100, and Rb = 900. Then the probability that A would defeat B is
In general, to find the probability that A defeats B use the following prediction formula: