Elo is a ratings system originally developed by Arpad Elo in order to rate Chess players. The system gives additional points to the winner of a game at the expense of the losing team. The system can also predict the result between competitors, and so the size of the adjustment after each win is dependent upon how the teams did against expectations.

In addition to Chess, Elo has also been used to rate soccer, baseball, and NFL teams. Just last week, fivethirtyeight.com released Elo ratings for NBA teams as well.

The ratings are based on all NCAA division 1 hockey games going back to the start of the 1998-1999 season. This includes the results of over 18,000 games. Unfortunately data is not available for older seasons.

A tie is counted as half a win. In practice this results in a (small) point increase for the underdog as they were given less than a 50% probability of winning the match.

No. The prevalence of empty net goals in hockey, means that the final margin of a game is frequently not representative of the skill of the teams.

Resetting all teams at the start of each season would leave the formula without much of the knowledge that it has already learned. However, making no adjustments would leave the formula too slow to react to the many changes that can happen to a team's line-up between seasons. As a compromise each team's rating is kept, but given a one third regression to the mean. This also has the happy side effect that most talented teams will reach their peaks in the playoffs instead of a few games into a season that follows a very successful year.

The K value used for the formula is 30. Those familiar with Elo will notice that this value is a bit on the high side. However a high K value is needed to account for the large amount of turnover that college teams see in their rosters. A bit of trial and error found that a value of 30 had the best predictive power.

Playoff games are given the same importance of any other game. However when calculating a team's end of season score, the results of the playoff games are significant as they are both the most recent games, and are typically played against other highly ranked teams.

Absolutely not. While retaining some knowledge from previous seasons helps make Elo ratings more accurate, it would be extremely unfair to allow the results of a previous year to have any effect on tournament selection.