Wednesday, June 17, 2009

Vezina '09: Why Thomas' GP Doesn't Matter

Anyone who has followed this blog for a while knows that I dislike arguments based on the number of games played by a goalie. If two goalies have a similar level of performance and one has played substantially more games, then it makes sense to take the guy who has more games played because it less likely that his performance was simply a hot streak or a good run of luck. However, often a goalie who has easily outplayed another goalie is considered inferior because he played less games, even though he would have had to play horribly in the remaining games to drop down to the other goalie's level.

Is there a big difference between a goalie who plays 50 games and a goalie who plays 65? Most people probably would say yes, and I know that most award voters would agree with that statement. The problem is that I doubt any of them ever tried to quantify the difference. What if the 50 game guy has a .935 save percentage, and the 65 game guy has a .905 save percentage? What about .925/.915?

I haven't seen any evidence that there is a statistical difference between playing 65 and playing 50 in terms of fatigue affecting your play. Fifty also represents the bulk of the season, which means that the additional games played do not have a major effect on the average.

To demonstrate this, let's take a goalie who has played 50 games at a pretty average level on an average team (say, .908 save percentage), and consider two scenarios, Scenario A: He plays the last 15 games like the best goalie in the league (.940), and Scenario B: He plays the last 15 games like the worst goalie in the league (.880). What is the effect on his seasonal numbers?

Scenario A: .915
Scenario B: .901

In both cases, his seasonal save percentage moves just .007, even though the extra 15 games played were either fantastic or horrific. It is pretty unlikely that NHL goalies will put together results more extreme than either of those two over a 15 game stretch.

Do the same thing comparing 50 games to 75 games (adjusting the save percentage assumptions a little closer to say, .930 for A and .890 for B since the games played sample is larger and extreme results are less likely to occur), and we get a split of about +/- .010 in save percentage.

I think that is a safe general rule of thumb to use, that if one goalie is .010 or better than another in save percentage, after adjusting for the team context they play in, and if both goalies have played the majority of games for their teams and you don't suspect there is any large differential in shot prevention between the two, then you can pretty safely say that the goalie with the higher save percentage is better. Even if one has played 70 games and the other has played 50.

Let's do a similar calculation for Tim Thomas this season to show that he was demonstrably better than the two other nominees, even though they played more minutes. Thomas played 829 fewer minutes than Backstrom, and 405 fewer minutes than Mason. If we assume that Thomas would face the same shot rate against if we has to play those extra minutes, we can figure out what stats he would need in that extra playing time to match Backstrom's and Mason's numbers.

To match Backstrom: 3.26 GAA, .884 save %
To match Mason: 3.85 GAA, .778 save %

Tim Thomas this season: 2.10 GAA, .933 save %
Tim Thomas, career: 2.62 GAA, .918 save %

How's this for a stat: If Tim Thomas played in 11 extra games and faced his usual shot rate, he could have allowed 5 goals against and lost every single one of them, and he still would have a better winning percentage and save percentage than Steve Mason.

Decide for yourself how likely it is that fatigue or any other factor involved would drop Thomas below those other guys.

Those are unadjusted numbers, of course, so take that into account. Shot quality measures suggest that Thomas faced tougher shots than Backstrom, but easier shots than Mason. We can calculate the shot quality factors that would be necessary for Thomas to have equivalent performance:

To match Backstrom: Bruins 13% easier SQA than Wild
To match Mason: Bruins 20% easier SQA than Blue Jackets

The typical spread from best to worst in the entire league is about 20%, so to argue that Mason's team-adjusted performance was better than Thomas' you would have to demonstrate that Boston was the best team in the league at shutting down opposing scoring chances, while Columbus was the worst. Even then the two of them would be virtually tied, so it would be a coinflip as to who wins.

Tim Thomas should win the 2009 Vezina Trophy. He should also be the First Team All-Star. I'm pretty confident Thomas takes the Vezina, as I don't even see the argument for either Backstrom or Mason to finish ahead of him, but I'll be interested to see what the writers do. They tend to put more weighting on things like wins and games played, so they might throw everyone for a loop and go with Mason, or even Evgeni Nabokov again.


Tom said...

Over his next 11 games (in the playoffs, against better-than-average competition), Thomas went 7-4 with a 1.85 and .935.

So yeah... fatigue was not an issue.

overpass said...

Good point on the playoff numbers, Tom.

While playoff numbers aren't a consideration for the Vezina, they should count to some degree when considering who the best goalie of 2008-09 was. As one of the few goalies who had good numbers in the regular season and playoffs, Thomas was almost certainly had the best individual performance of any goalie in this past year.

The Contrarian Goaltender said...

Overpass: That's an interesting thought, to look at combined regular season and playoffs to determine the best goalie. I don't think it would affect anything over the last couple of seasons, since none of the top goalies have matched a top-notch postseason with a top-notch regular season and the guys who went deep weren't even close during the season, but I'm sure it would affect some of the races in prior years.

Often the best goalie race is pretty close, so if one guy has a bad playoffs and the other guy has a great one it would probably make the difference. I think it would particularly help in the case of a goalie who put up a high save percentage but played fewer games and therefore didn't put up the win totals that attract attention, yet continued their play deep into the playoffs. Ed Belfour in 2000 and Miikka Kiprusoff in 2004 are two guys that I think would have easily won that year's best goalie award if you included playoff results. Other times two goalies have very similar numbers, and the playoffs could be seen as a tiebreaker (Luongo/Brodeur in 2007, for example).

overpass said...

Regarding the games played issue, I'd also rather not put a lot of weight on it. Every goalie's in a different situation with regard to backup strength, the coach's view on playing backups, etc. All they can do to increase their own playing time is play well.

If, like Thomas, a goalie begins the year as the starter, plays very well all year, and still only plays in 54 games...what else can he do?

Of course sample size is an issue, and you're more likely to put up great rate stats in fewer games by random variation. For that reason, I think the best way to rank a goalie's season is a z-score type system, where you ask the question "What is the probability that an average goalie could achieve this result?", and the most unlikely positive result ranks first. From that perspective, Thomas certainly had the best season (assuming team effects are not major), and your analysis basically says the same thing, in a less mathy way.

Ed Belfour in 2000 and Miikka Kiprusoff in 2004 are two guys that I think would have easily won that year's best goalie award if you included playoff results.

Kiprusoff's 2004 is the big winner here, I think. He got a lot of recognition considering he only played 38 games, but his 64 combined games made for a year that would fit right into the middle of Hasek's peak. 39 wins, 1.76 GAA, 0.931 SV%, in almost 4000 minutes.

And yet, as great as that was, it still doesn't match up to Hasek's 1999 - 43 wins, 1.85 GAA, 0.937 SV% in over 5000 minutes!

Sure, it's not entirely fair to use combined stats in this way. Goalies on bad teams don't get a chance to have a big playoff run, and tough playoff competition can kill rate stats. It's also the case that Thomas didn't play in 28 regular season games for his team - 11 more playoff games don't change that. Still, I think the playoff results have to count for something when looking for the best single season performances. It's something that an award-centric approach will miss, as the awards don't consider the playoffs.

Anonymous said...

I agree with your pick, Contrarian, and am glad that you are not advocating that Luongo, who was alright at best outside of November after factoring in team effects, win the Vezina.

Tom said...

I wouldn't go so far as to include playoff performances in award consideration, or even necessarily in my overall estimation of a goalie's season. Some goalies get lit up in the playoffs because they're simply up against an impossible task.

But I do think that playoff #s are valuable in proving whether a guy like Thomas benefitted from playing fewer games. He started their last 12 in a row, and showed no dropoff. That (more or less) proves that his numbers were not the result of a low GP.

Navin Vaswani (@eyebleaf) said...

I can't believe Tim freaking Thomas is going to win the Vezina.

The Contrarian Goaltender said...

"Kiprusoff's 2004 is the big winner here, I think. He got a lot of recognition considering he only played 38 games, but his 64 combined games made for a year that would fit right into the middle of Hasek's peak. 39 wins, 1.76 GAA, 0.931 SV%, in almost 4000 minutes. "

Luongo's 2004 was pretty good as well - .931 in over 4000 minutes. That would make for an interesting debate - Luongo vs. Kiprusoff in 2004, playoffs included. I think that the writers/GMs would certainly have voted Kipper in that scenario, because of the wins and GAA, but in my eyes it's a tossup between him and Luongo.

"Some goalies get lit up in the playoffs because they're simply up against an impossible task."

This is undeniably true. Simply adding the numbers together wouldn't be the way to go, as you'd have to heavily adjust some of the playoff results for strength of opposition.

"I can't believe Tim freaking Thomas is going to win the Vezina."

I'm probably not as surprised as you are, Thomas has been pretty good for a while now, but I certainly wasn't expecting this type of season from him either. Just once again proves that style doesn't matter, and it's all about whether you stop the puck or not.

Jonathan said...

Thomas' GP doesn't matter, because is save percentage was significantly higher than

Sometimes GP matters marginally. Example: Ryan Miller. Miller got injured during the season and only started 58 games. His save% was .918. The teams sv% was .914, so if Miller had started 70 games instead, his team's sv% would have bumped up by 2 points. The difference between Luongo's career save percentage and Osgoos' career save % is 13 points.

GP is probably wayyy overrated. It's not negligible. It is overrated. As CG said, GP is an important factor to take into account when two goalies are about ever.

Jonathan said...

*about even