I think this idea is an excellent framework for evaluating goaltenders. I've played around with similar ideas, but haven't ever put the work in to do it properly.
Ideally the process would go through the following steps, using any goalie rate stat of your choice (winning percentage, GAA, save percentage, or even shutout percentage).
1. Calculate the expected (goalie stat) for each goaltender based on their opponents faced, home/road split of games played, and travel schedule surrounding games played. Use empirical data to find the magnitude of each adjustment.
2. Adjust all goaltender seasonal (goalie stat) by the expected (goalie stat) from step 1. For example, if a goalie has a 0.550 winning % and an expected 0.490 winning %, his adjusted winning % after this step is 0.560.
3. Calculate the difference between goaltenders and their goalie teammates in (goalie stat) for each goaltenderseason.
4. Find the career difference a goalie and his goalie teammates by taking an average of the seasonal differences calculated in step 3, with each seasonal difference weighted by (the lesser of goalie GP or his goalie teammates' GP).
5. All goaltenders now have a career +/ difference in (goalie stat) relative to their teammates. Calculate the career +/ difference for each goaltender's goalie teammates, weighted by the same weight each teammate had in step 4 .
6. Adjust the career +/ difference in (goalie stat) for all goaltenders for their strength of goalie teammates, which was derived in step 5. Repeat this step several times if desired.
By far the biggest problem is that goaltenders don't necessarily have a constant ability level as this method assumes. And the method gives the most weight to half seasons, so this can mean that a longtime starting goalie is rated primarily by his weakest seasons. For example, Tony Esposito's career estimate would be heavily informed by his age 38, 39, and 40 seasons, because they were among the few seasons that he didn't play 65+ games. Any individual player can have his rating easily adjusted by picking which "prime seasons" to include and which to exclude, but that process isn't really feasible for a global rating in which each player's rating is important.
Last edited by overpass: 12112012 at 10:03 PM.
