Saturday, August 8, 2009

Making "Key Saves" 72% of the Time

Here's an interesting fact from Gabe Desjardins of Puck Prospectus:

"72% of all regulation play during the 2008-09 season was spent tied or at a one-goal differential."

This is why I think people are simply talking nonsense nearly all of the time when they start going on about goalies who make the big saves or who make saves at the key times. Most of the game is "key time", because there just aren't that many goals in NHL games. "Key" implies an important or pivotal time in the game, and it seems to me a bit improper to use that word to refer to a situation that includes nearly three-quarters of the saves a goalie might face.

Let's say we had an extreme case of a goalie who puts up a .920 save percentage in one-goal or tie games, but then completely tunes out and stops caring when the margin gets to 2 or higher (.890). If the shots against are distributed with the same 72/28 split, his overall save percentage is going to be about .912. Given what we know about how teams play to the score (shot rates tend to drop in blowouts and tend to rise in close games), that assumption is probably false, which means that there will be even less of a save percentage difference. In addition, that goalie might save a few more wins for his team in close games, but he'll also make it much harder for his team to come back occasionally from a 2 goal deficit, and he would also frequently put his team at risk of giving up 2 goal leads. As a result, I'm not sure there would be much of an increase in the team's win totals.

This shows that even an impossibly extreme split leads to an overall save percentage difference of just .008. Even if there are actually some goalies who consistently either raise their play in tight situations or lose focus a bit in blowouts, that is likely to have a very small effect over the course of an entire season. Let's say in the above example the goalie had a .910 save percentage in the blowout scenario, for a drop of .010 (which is about the difference between an average goalie and Roberto Luongo). That would leave the goalie's overall save percentage at .917, very close to the original .920.

The converse is also true, that a goalie who plays poorly during the 72% of the game that is "key time" is likely to have poor overall stats as well. It is not possible that a goalie can be so good in non-pressure situations that he makes up for having awful stats in close games. Let's change the example to a goalie who can't handle the pressure and can only stop 89% of the shots against when the score is tied or his team is within one goal. What save percentage would he need to have in all other situations to end up at the same .912 as in the first example? The answer is .970, and it obviously goes without saying that nobody is going to consistently stop pucks at that rate in any game situation.

It is not plausible to me that a goalie can make up a large statistical gap through timing his saves. Hockey is a fairly low scoring game, and as a result most of the time the score is close. Goalies who make a lot of saves also make a lot of key saves, and goalies who allegedly bear down when the game is close will spend about 3 minutes playing with extra focus and competitive fire for every 1 minute they are supposedly goofing off in a blowout. If one goalie has clearly better numbers than another goalie but a lower win total, then that is almost certainly not because of "making the key saves", but rather because of differences in the number of goals the two teams scored and the number of shots against each team allowed.

Note: Even though I've never seen much evidence to support it, I'm not ruling out differences in clutch play between goalies. I just don't think the difference would be very large and therefore we would need a huge sample size to be able to identify it. Even if we find some differences between goalies over a multi-season sample, we might still not be sure whether we are observing a difference of luck or a difference of skill. This uncertainty makes it something that is probably not really worth worrying about for the most part.

5 comments:

Lawrence said...

Can it be possible that the 72/28 split is shifted by the fact that it takes a large percentage of game time for a game to naturally progress out of a tied - one goal scenario. Even in a large blowout, let's say 6-1 or 8-1, if you average out the goals, for ease, it could take 3 - 20 minutes+ to move out of this 72% time period, even though the game isn't close. Whereas, the play, possession, corsi etc. can be completely dominating for the other team.

An interesting example of a previous conversation I was having was that Miikka Kiprusoff had an 'elite' goaltending record (GP'd with an over .920sv%, which coincidentally you have used as well) of 27-1-1. 16 of these games ended as 1 goal games. 15 were a .945% or higher. All of this sounds impressive but we know that he ended the year with a average .903% which is not good at all. I looked at the results and similar to what you have written, the Flames got hammered in a number of games last year. I singled out seven to show the huge effects of blowouts, both positive and negative on sv%. Those seven games make the difference between a year-end sv% of .903 and .914. A huge effect. I found this interesting. Of course not to exclude them, or make them trivial, but to show the fragility of the average sv% stat. Let's say 29 "elite" games, where he 'won the game' a series of average and 7 bad...reflects a disastrous season.

Thoughts?

The Contrarian Goaltender said...

I agree with you that a few bad games can have a disproportionate impact on a goalie's stats. I also agree that had some effect on Calgary and Kiprusoff in 2008-09. Kiprusoff had some games where he got shellacked but the team was probably going to lose anyway, and Calgary outperformed their goal differential both for this reason and because they did well in one-goal games (21-9-6).

However, Kiprusoff had more than his share of bad outings, and that cost his team in the standings. The Flames had two 8-6 losses where Kiprusoff got pulled, for example, so yes those games killed Kipper's stats but they also directly cost his team two games since scoring 6 goals in regulation is almost always a win.

That Kipper's record was so good when his save percentage was .920+ just confirms to me that Calgary was a strong outshooting and outscoring team last year (and they probably also got a bit lucky in close games). I'd suspect that most goalies on strong teams would have a very good record when they put up a high save percentage, because when outshooting teams also have the percentages behind them they become very difficult to beat. The median save percentage is .912, so .920 isn't really what I'd call "elite". The difference between .920 and .912 is the difference between 23/25 and 21/23, or 34/37 and 30/33.

A better clutch argument for Kiprusoff would be to split out his save percentage by situation, e.g. save percentage while the game was tied or within one goal.
He probably did play somewhat better in tight games, but given his overall record and his number of disastrous outings I still doubt that Kiprusoff's overall contribution to his team in terms of winning and losing games was a positive one in 2009.

Lawrence said...

CG: I agree with you. I use .920 as the cut-off for 'elite', admittedly somewhat sloppily, because that was the sv% of the 5th best goalie last year. It's crude, but you could say that those were the 08/09 elite goalies - Thomas, Vokoun, Anderson, Backstrom and Luongo.

I do disagree with this though: "I still doubt that Kiprusoff's overall contribution to his team in terms of winning and losing games was a positive one in 2009." as he only lost 24 games, which, I am sure a few must fall under the "where he got shellacked but the team was probably going to lose anyway" statement you made. Therefore, even conservatively if only 4 of those games qualify that's 20 losses.

As you pointed out, the Flames won 21 one goal games, none by McElhinney, 15 with a .945sv% and 15 more .920% or better. I would say being conservative Kipper likely won as many games for the Flames as he lost, but quite frankly, that still isn't good enough.

The Contrarian Goaltender said...

The cutoff of .920 for elite is not a bad cutoff if you are talking about seasonal results. But on a game-by-game basis I'd argue for a higher number. The reason is that on a per-game basis the median (.912) is higher than the average (.908). As you observed, bad games can have a big negative downside.

For example, a goalie can play 4 games at .930 and then one at .875, and he'll end up around .920 overall. If you are looking for elite games, I'd say .925 or .930 would be a better line in the sand.

I figured I'd look a bit deeper at Kipper's win totals. Kiprusoff faced an average number of shots against and the Flames took an average number of penalties, so if we assume they had average shot quality at EV and on the PK we can pretty much evaluate him on his goals against since we don't have to adjust for anything.

I have some win probability estimates for allowing a particular number of goals against, e.g. the win probability is something like 88% for a goalie that allows one goal against, so for every game he allowed one goal against we can count 0.88 expected wins. I quickly pulled Kipper's Yahoo game log, plugged in the expected win totals for each start, and figured that based on the goals against the Flames should have picked up about 80 standings points in Kipper's 76 starts. League average these days is actually .557 so an average team with an average goalie would put up 85 points in the same number of games, i.e. Calgary would be expected to win less often than average with Kiprusoff in net.

That's assuming Calgary had an average offence, which they did not. In Kiprusoff's starts the team actually picked up 95 points, 15 more than expected, because they had an above average offence and because they won the close ones.

Kipper might have had some small effect in helping win a few of the close ones, but I think on the whole he was below average and his teammates and a high GP number helped him to a high win total.

Corey Pronman said...

I agree CG, we see this argument faulted all the time in basebal when people bring up the clutch factor. Thing is in baseball over a large span we do see "some" clutch ability, but it so small its hardly worth glorifying.

In hockey with a much smaller scoring context, I'd imagine if we could truly neutralize it, the clutch factor would be even more nonexistent.