http://www.hockey-reference.com/about/adjusted_stats.html describes it in detail.
They have 3 adjustments:
- Schedule. Normalize everything to 82 game seasons by dividing totals by season length and multiplying by 82.
- League scoring. For goals, normalize everything to 6 goals per game by dividing league wide goals per game *without* the player in question and multiplying by 6. For assists, normalize everything to 10 assists per game (6 goals * 1.67 assists per goal) by dividing by league wide assists per game *without* the player in question and multiplying by 10. For points, simply add adjusted goals and adjusted assists.
- Roster size. Adjust to an 18 skater roster by dividing allowed roster size in a season by 18. Theoretically this is to account for a player making up a larger percentage of his team and thus being expected to make up a larger percentage of the scoring.
Personally, I have a few issues with this approach. I don't agree with excluding the player from era averages, especially when you're dealing with massively different league sizes. In the 52-53 season they chose, Gordie Howe made up just over 3% of *all* NHL assists, but he also played in 1/3 of all NHL games that season. In a 30 team league a player plays in 1/15th of league games. In 05-06 Thornton had 96 assists in 81 games, but that made up well under .8% of all assists that year, mainly because he played in such a (relatively) small percentage of league games.
I guess my main point is that if you're trying to find the underlying scoring an era, you bias yourself much more by excluding a guy who was on the ice for probably 12-14% of all minutes played as opposed to excluding a guy who was on the ice for ~2% of all minutes played.
I'm also not a huge fan of adjusting assists and goals differently. I haven't seen any evidence to show that assists were inherently harder to get on a given goal in any era or that the variation in assists per goal is anything but random. If it exists I'd love to see it.
I'm also not a fan of the roster size adjustment. I really don't care if rosters have 10 skaters or 100. What really matters is how much time a guy played. Just because Howe played on a smaller roster doesn't mean he inherently got more ice time than Crosby, lines aren't rotated equally.