Mark's Blog

As a fan of the Ottawa Senators, I was very happy today to learn that they acquired Capitals' star Peter Bondra. Bondra gives the highest-scoring team in the league and even greater power play presence and is a fast, skilled player that will fit right in with Havlat, Hossa, Alfredsson, Spezza and company. Bring on the playoffs!

On a hockey statistics note, I've been dissatisfied with Goals Against Average, GAA (goals allowed per 60 minutes played), as a measure of goalie success, since goals against are a function of shots faced, and this is not normalized between teams. Save percentage, or shots saved divided by shots faced, is a better way of looking at a goalie's skill level.

The problem with save percentages, similar (although in general a bit lower) to fielding percentages in baseball, is that they are presented as numbers less than but quite close to 1, which makes quick comparisons very difficult, and obscures significant differences. For example, if Player A makes 2 errors in 40 plays, he would have a fielding average of .950, and if Player B makes 1 error in 40 plays he would have a fielding average of .975. Player A makes twice as many errors, but the untrained eye only sees a .025 difference.

I decided to calculate the league average of shots allowed per game, in order to normalize save percentage to a more understandable number, in the form of a GAA. This was a bit more tricky to do than I initially thought, since the number of minutes a team has played isn't readily available (at least I couldn't find it online). I added 2.5 minutes for a team for every OTL, since there is more or less an equal chance of a goal being scored throughout the 5 minute OT session. I also removed empty net goals from this, since (I think by definition) all shots on an empty net are goals. This gave an estimated league average shots per 60 minutes of about 27.4.

For each goaltender, this number is multipled by 1 minus the save percentage (i.e. 27.4 * (1 - Save%)). I'll call this AGAA ("Adjusted" GAA). The list isn't very interesting in and of itself, since it's in the same order as the Save percentage list, but if we look at the difference between AGAA and GAA then some names pop out. It would also be a much more useful way of talking about save percentage in general.

The five goalies (among qualifiers) that are most overrated by their GAA are (ranked by AGAA - GAA) as follows (using numbers from games as of February 17, 2004). A positive number means that if the goaltender faced a league average number of shots they would have allowed that many more goals per game, based on their save percentage:

1. Marty Turco, Dallas, +0.41
2. John Grahame, Tampa Bay, +0.40
3. Patrick Lalime, Ottawa, +0.31
4. Robert Esche, Philadelphia, +0.29
5. Tommy Salo, Edmonton, +0.29

And the five goalies most underrated by GAA are:

1. Marc-Andre Fleury, Pittsburgh, -0.80
2. Sebastien Caron, Pittsburgh, -0.62
3. Roberto Luongo, Florida, -0.51
4. Olaf Kolzig, Washington, -0.35
5. Sean Burke, Phoenix/Philadelphia, -0.29

Note that these lists do not represent good or bad goaltenders, just those who have misleading GAAs, because they played with teams that allowed much more or much less shots a game than the league norm. AGAA is a normalized, easier to follow version of save percentage. As an example, the current save percentage leader, Vesa Toskala has a GAA of 2.01 and a save percentage of .933. Multiplying 27.4 by (1 - 0.933) we get an AGAA of 1.84. This would mean that Toskala, facing a league average amount of shots per game, would be expected to have a GAA of 1.84.

Being an Ottawa fan, I note that Martin Prusek's save percentage is .928 (GAA 1.80), which corresponds to an AGAA of 1.97. Lalime's .906 (GAA 2.27) corresponds to an AGAA of 2.58 (see how the true difference of just a few points of save percentage, is actually 0.6 goals a game! And while it might at first seem that Prusek, being the backup, might play against weaker opposition, a quick check of his record shows 10 games against playoff teams, 8 against non-playoff teams (and a few minutes of mop-up duty against the Rangers in a game in November), which is an average level of opposition.

One improvement that I could see for this would be to also adjust for the team's penalty minutes taken. It's not fair to penalize goalies for their teammates putting them in a situation where a lot more goals are scored (teams score at about 1.5 to 2.0 times the rate on the power play than they do at even strength).