Can We Accurately Predict Which PK Units Will Score Shorthanded?

Embed from Getty Images

Last week I went on Montreal radio and talked about how dangerous the Ottawa Senators’ penalty kill units are. Led by speedy forwards like Curtis Lazar, Jean-Gabriel Pageau and Mark Stone, and with help from puck moving genius Erik Karlsson, the team has feasted on opposing power plays this year to the tune of the highest GF/60 minutes shorthanded in the league since at least 2007-2008. When considering the team’s league worst GA/60 — mixed with a little bit of film — it becomes clear that the Senators yield chance against in exchange for opportunities for on the break. It may not have been intentional at first, but once the team started capitalizing on its rushes, it seems likely coach Dave Cameron gave his players the green light to go, to try and come out on top on aggregate. The result? While being last in GA/60 shorthanded, the Senators are third in GF%. The problem with GF% when it comes to special teams though is that volume matters more when the ice is tilted. Two goals for and Eight goals against isn’t the same as Four goals for and 16 goals against. So goal differential per 60 is a more accurate measure of success on special teams. The Sens are 30th in GD/60 shorthanded, so it’s hard to say the strategy has been that much of a positive for the team (unless, say they’re down a goal and shorthanded near the end of a game).

Regardless, following my radio appearance, the Senators came out and scored three shorthanded goals against a Canadiens unit that was lacking in foot speed and seemed somewhat unaware of this threat. Yesterday, Matt Cane had a great piece that suggested that aggressiveness on the penalty kill is a repeatable skill and that playing more aggressively may result in a better goal differential on aggregate.

Seeing as I haven’t delved too deeply into the penalty kill thus far, and considering it’s now topical, I decided to look into offensive performance on the penalty kill. Is it repeatable? And maybe more importantly, can we predict it based on past results?

Let’s start with repeatability. We can check the correlation between odd and even games for each team from 2007-2008 to last season to get an idea of whether each metric represents a repeatable skill. I will use War-On-Ice’s definition of Expected Goals, as I did in my original piece on power play predictive metrics.

Screen Shot 2016-03-23 at 11.58.07 AM

There is some repeatability here, especially with the larger samples, but nothing that comes close to matching Cane’s 0.77 for his aggressiveness rating, which is interesting. It does makes some sense that using SH offensive zone entry attempts provides an even bigger sample from which to draw conclusions. Now what about predictive ability? Can one look at the first X games from a season and accurately predict how many goals a PK unit will score the rest of the season?

SHGoalPredictiveMetrics

I left the y-axis equivalent to how I had it for the chart in my PP piece to show just how little we can predict about shorthanded scoring. The sample sizes are small enough that past metrics can explain only up to about three percent of future scoring. If one was to try to look ahead, though, it seems that shot attempts for are the most predictive for the first 15 games, at which point scoring chances are best until 25 games are accumulated. From that point forward, looking at the GF/60 column of the shorthanded stat sheet may in fact be the most accurate (if accurate is a fair word) way to predict future goal scoring.

It makes sense, therefore, that the Senators be seen as a dangerous unit shorthanded. They are only 13th in shot attempts for per 60 on the PK, but as mentioned have scored by far the most goals.

So what about year-to-year? Can we use the results from this season to predict whether teams will be dangerous shorthanded next year?

Screen Shot 2016-03-23 at 12.10.45 PM

The larger sample sizes do provide marginally more clarity, but still not enough to be confident in any kind of prediction. It’s interesting that past shot attempts are more predictive than past goals considering the results above, but overall the conclusion has to be that special teams should be treated with care. We still know very little about them.

As just one example, the New York Islanders were last year’s Ottawa Senators in terms of the penalty kill. This year they are one of the top teams in goal prevention down a man and are only middle of pack in goal scoring. Granted, their shot generation in such situations has been good both years, but it’s difficult to bank on a certain goal differential on special teams if the results are that prone to variance.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s