## September 10, 2007

### Player Save Percentage

I'm a big believer in "shot quality" as many readers have likely noticed. If players have an impact on the quality of shots a goalie sees, then they should affect how many pucks the goalie stops. If a goalie generally stops 90% of shots then a player who makes shots twice as easy to save should have a save percentage of 95%. Of course players have a much smaller impact of save percentage than goalies do.

In order to see if players can affect save percentage I did a season vs. season regression: I check if the players who out performed in 2005-2006 continued to outperform in 2008-2007. If there is a significant relationship between the two seasons then obviously players impact save percentage.

The regression does come out significant, which agrees with the theory that players can affect shot quality.
The regression is reasonably simple:
Difference in team's save percentage in 2006 vs. 2007 = D
Player's Save Percentage in 2006 = Save%
So:
Player's Save Percentage in 2006 = 0.7 + 0.235*Save% + 0.618*D
Basically the above equation is a regression towards the mean: it takes extreme values and brings them closer to the average (0.920).

Note: this data only includes even strength shots.

However, this data is worthless for predicting how a player will do next year due to the large amounts of variability. For example Crosby had a n excellent save percentage this season of 0.930, this system says he'll likely perform at 0.919 given identical goaltending as last season, however he could easily do as poorly as 0.890 with bad luck or as well as 0.950 with good luck. A 2% difference in save percentage on 600 shots works out to about 12 goals over the course of a season. So this large range of possible save percentages for players can have a large impact of a player's plus-minus, often due to just luck.

Results
Any way most people will want to see the results

David Johnson said...

Is this all shots, or just even strength shots? You should probably just consider even strength shots.

JavaGeek said...

just even strength shots?

I forgot to mention it at first, but it's in there now.

Vic Ferrari said...

Terrific stuff. Do you plan to post shooting percentage numbers for while players were on the ice?

Also, I have a script somewhere that generates these numbers from the shift charts, which aren't as accurate of course, but they do go back a few years. If you post just the 5v5 numbers I'll compare and see how accurate it is.

oilswell said...

Nice job doing a regression test, but what evidence do you have that the players themselves are more significant factors than the profile of opposition quality the players see?