Wednesday, January 12, 2011

The Minnesota Wild Penalty-Kill

Above: Dellow sending his minions (I kid :)
There is a bit of a debate going on over at mc79hockey with basically me on one side and a bunch of Dellow's Flying Monkeys on the other. The basic premise is that Dan Tencer wrote an article that attributed the Oilers recent success on the penalty kill to an improved coaching strategy. Tyler Dellow, Robin to my Batman, took issue with that idea, pointing out the Oilers 4v5 save percentage was a staggering .930 over that period. In his mind:
Without getting too technical, the Oilers continue to take a hellacious beating while they’re on the penalty kill. It’s baffling to me that the response to the alterations and well timed aggression has been to bleed shots against more quickly, but so it goes. The goaltending “improvement” is nothing more than 15 games of bounces. The analysis of the penalty kill and why it’s doing better is complete gibberish.
I suggested that perhaps the entire story is not being told in his snap judgment of "15 games of bounces". To my eye the PK had been better in two primary things: reducing the extra high quality shots against and the goaltending. As Tyler pointed out in the start of the article, it certainly wasn't the shot rate that had changed (64.5, actually up from the Oilers average this season). I responded with:
By your own numbers, then, the Oilers were a team that was far BELOW average to start the season. If you consider they are .852 on 4v5 for the season, even if they are playing way above that recently, this indicates they were even farther below their season average earlier.
Considering both of our goalies are playing better recently, it stands to reason that a lot of the improvement is a correction of Sv pct towards average, rather than woe-is-us “15 games of bounces”.
So your analysis, I think, also fails.
I probably shouldn't have used the word fails. It has such a negative connotation these days, and in fairness I don't think his analysis really failed as more was incomplete. This sparked off a debate that attacked my estimation that the shot quality against had gone down, my understanding of regression, I'm a big fat stupid, etc. All fine and dandy, it is his home turf. Tyler, to his credit, conceded the point that surely some of the recent improvement was goaltending, but maintaining it was not the only piece of the overall puzzle:
I’m happy to concede that a big part of their correction is regression towards the mean in save percentage. I don’t dispute that. There’s a reason I explicitly referenced league average and not the Oilers save percentage overall or prior to the hot streak.
I’m not entirely sure how that means my analysis fails. They’re the worst team in the league at preventing shots at 5v4 and have gotten worse during their hot streak. The expectation is that they’ll have the worst penalty killing in the league
In general, since hockey is a fast-paced sport with a bouncy puck played on a slippery surface, every shot against is 'bad', so to speak. So ultimately I'd think the best strategies are generally the strategies that yield the fewest shots against. With that said, there are definitely examples of team strategies that allow shots at a decent rate while attempting to minimize the damage each shot does.

Eventually a user named Schitzo posted the following:
Right, but why isn’t [the inflated save percentage] sustainable? If the entire premise is “keeps shots to the outside, let the other team fire away”, that should be reproducible. Design the entire system around that strategy. Spend 82 games on it.

But if the result isn’t a 0.930 save percentage at the end of the year, I would suggest that “quality of chances” is not actually the relevant factor.

So I decided to take a look. I immediately thought of the Minnesota Wild, a team I knew that through the Lemaire years had not actually been a great team at preventing shots against, but I was pretty sure they were up there defensively. I have listed some of their 4v5 stats available on (change the year in the URL to look at different seasons):

YearSA/60SA/60 RankSvPctSvPct RankGA/60 GA/60 Rank

The First two teams were coached by Jacques Lemaire, and the 09-10 version by Todd Richards (indicating a likely change in overall strategy in 09-10). All of the teams were backed by Niklas Backstrom (playing 58, 71, 60 games respectively). What is interesting to note is that under Richards direction, they were one of the best teams in the league against preventing shots against, yet dropped from the best PKing team in the league to right in the middle of the pack (almost double the goals against per 60). Now it's possible that Backstrom had an off... season (his SvPct dropped 23 places on the PK), but what is clear is that at least in some cases the shots against indicator does not provide ample information to determine whether a PK was good or not (indicated by the GA/60 rank).

In terms of the overall team, the 08-09 Wild were 89 points 40 wins, and the 09-10 Wild were 84 points with 38 wins. In other words the difference on the PK was likely a factor in the point discrepancy, but otherwise they were similar performing teams overall.

So what do we take from this? First .930 is a sustainable number over a single season (since it has happened at least once), but as far as I can tell it is the absolute top of the mountain over the last three seasons (some teams can maintain over 910). In other words it's highly unlikely to be sustained. Second, shots against per 60 minutes clearly do not tell the whole story, at least not in the Minnesota Wild's case.

There are others that are similar. The 2008-2009 Calgary Flames for instance, who were 25th overall in SA/60 4v5 at 55.9 but sported the 7th overall GA/60 4v5 at 5.8. They had the third best SvPct at .897. The following year Ottawa had the 25th overall SA/60 4v5 and the 10th best GA/60 4v5. A .900 Sv Pct in this case was 5th best in the league.

It seems like at least in some cases it is possible to have higher shot rates and still maintain a good 4v5 Sv Pct and therefore a good GA/60 4v5. Whether or not this is team strategy or simply good goaltending, I'm not sure. That would probably be a topic for a longer, more statistically intense post.


  1. Way to disprove a claim that nobody was making.

  2. It was actually the last comment by Schitzo that asked if there was an 82 game system that had a high save percentage with a high shots against that was effective. It turns out, there is. The Minnesota Wild.