THE BOOK--Playing The Percentages In Baseball

Great piece by Hawerchuk.

Tango, could your finding be selective sampling?  If a team has pressure on, and they score on the first shot, it looks like the goalie was cold, with an .000 save percentage that minute.  If they have sustained pressure, and score on the fifth shot, he has an .800 save percentage that minute.

Power plays could help cause this.  They are more likely to lead to extended shot flurries, which end as soon as the goal is scored.

I’m not sure what to think of this in light of the recent blog post about NHL shot record accuracy.

I think Hawerchuck makes a good point about trailing teams taking more low-quality shots, and I would guess that more shots in general means more easier save opportunities while fewer shots tends to mean more selective shooting leading to a lower expected save percentage.

BrianK/#2 also brings up a good point.  If shot recording is inconsistent, then that’s going to lead to a higher save percentage correlating to more shots.  Goals are always going to be shots, so the differences in recording shots on marginal on-goal plays are going to have an effect on save percentage.  When a guy dumps the puck on net (or one-times a slap-shot aimed at the 5 hole like in the video linked in the article on shot record accuracy), the scorer who counts that as a shot will be increasing both the number of shots and the save percentage of the goalie compared to the scorer who does not count that as a shot.  Any time you see more shots recorded in light of this inconsistency, you would probably expect a higher save percentage even if the goalie didn’t perform any differently.

Sort of off-topic, but does anyone know if there are any plans for Sportsvision to go back to the puck-tracking technology from the glowing puck days and get puck f/x data for this type of thing?  It seems like that would be a major breakthrough for addressing this type of question.

Hawerchuk’s point is reasonable, which is why I added:

“then further break it up by tied, up 1, down 1, up 2+, down 2+.”

If a team takes a lot of shots, they’re either (a) better, or (b) taking/forcing a lot of shots.  So I can see why a goalie’s save% is higher when facing a lot of shots… I’d guess it’s because a team is taking a lot of shots whenever they can, to try to make something happen (get a rebound, lucky bounce, etc).

If you like you can further break it down by quality of a team’s offense, and quality of a team overall.

These parameters you guys are bringing up, while legitimate, are not so arcane that they can’t be controlled.

As long as you are going by the official shots, wouldn’t you still have to also get around the bias created by inconsistencies in the recording, or at least be aware of it in your conclusions?  Every time there is a subjective call on a play on net that could go either way (shot on goal or no shot), the rise in shot total (scorer decides it’s a shot) will coincide with a rise in save percentage, so there will be a bias that when more shots are credited, the goalie will tend to have a higher save percentage by virtue of being credited with more saves on grey-area calls.  Even if you control for official scorer by breaking down the data by arena (assuming they have the same official scorer each game), you will still have the problem to some extent.

I’m not saying it can’t be accounted for or that you can’t do the study because of this, just that there might still be bias in the data even if you account for the score and the teams, and that’s something else that should be considered if you (or someone else) do conduct a study.

In theory, a team scoring easily (either through bad goaltending or a run of excellent/lucky shots) won’t tend to take as many shots - they cash in too quickly on their opportunities to amass a lot of putbacks and flurries.  But I’m not sure how one could identify a trend.  A team that gets blanked may have only 18 shots, or 40.  A team could score three times on 20 shots, or on 61 as we saw Toronto do this year.

I think that the trend would be most pronounced for higher-scoring games, where one team gets five or more goals.  Unless the other team is also scoring, a team enjoying that kind of a night will often not take as many shots, consciously playing back while the other team puts everything they can on net to try to climb back into it.

Of course, I grabbed Cam Ward’s 08-09 season as a quick example, remembering that he had a 57-save game against the Isles last year (facing 60 shots).  The ‘Canes had a two-goal lead for a total of 33:30 in that game, so the theory would hold; but he had a game late in the year, also against the Isles, when he shut them out on just 12 shots… and that ended 9-0 for Carolina, who took 57 shots.

Sorting his game logs for shots faced and goals permitted reveal the two clearest trends.  In the 14 games where he faced 35 or more shots, he stopped better than 94% of them 11 times.  So, do more shots mean better work?  Not necessarily.  Only twice did he permit more than three goals in those games - but when you look at the other 14 times he gave up 4+ goals, you see only one game where he stopped even 90% - stopping 38 of 42 Maple Leaf shots on Nov 4, 2008.

Those two trends seem to suggest that stopping a high percentage of shots is what pushes the totals higher, not the reverse.  In his four worst save % performances of the year, Ward was yanked every time, meaning that he didn’t play well enough to be allowed the chance to face a lot of shots - in much the same way as a pitcher getting smacked around doesn’t get a lot of IPs.  (In only one of those games did Carolina eventually outshoot their opponent - by 32-30.  Another game was 25-27 against, and the ‘Canes were outshoot by 19 in each of the other two.)

The next step, I suppose, is to go through more goalies’ game logs to see if the trends hold, and factor in the quality of the backup and the team.