Scoring Chances, Nathan MacKinnon, and You

Recently the stats community has taken on scoring chances. Seemed like it might be illuminating to look at those numbers for the Avalanche and see what they might be able to tell us. In particular, after acquainting ourselves with what the number means and does, we'll apply it to Nathan MacKinnon's slump and see if anything can be learned.

What is a scoring chance?

Typically in the past, scoring chances have been tracked by basement-dwellers re-watching games and counting by hand. It's highly subjective, making the data suspect, and painful as hell to gather in the first place. The numbers are difficult to apply across teams or situations because teams tend to have one-ish blogger who tracks them, so criteria, even if stated to be the same, may well not be.

That's been changing. Thanks to work from David Johnson of HockeyAnalysis/Puckalytics and the dudes behind War on Ice, they have used their event-crawling scripts to come up with an objective, calculable definition of a scoring chance, lain out here on War on Ice's blog:

• In the low danger zone, unblocked rebounds and rush shots only.
• In the medium danger zone, all unblocked shots.
• In the high danger zone, all shot attempts (since blocked shots taken here may be more representative of more "wide-open nets", though we don’t know this for sure.)

These definitions are flexible but we feel they’re a reasonable starting point given all the data we have available. We’re open to changing it if we have sufficient numerical evidence.

Rush Shots and Rebounds are defined based on where and when the last event occurred, and also what it was. Those definitions are available and/or linked to in the blog, which is short so you have no excuse not to click it. Kenny Loggins Zones are defined here.

How to use it

So now that we've determined what a Scoring Chance is, we need to figure out what we can, or can't, do with it. For this I've relied on Steve Burtch, who for some reason prefers to dump all this information on Twitter at once rather than write an easily-linkable piece about it, so we'll get our stuff in chunks.

Using data for All-Situations the r^2 of SCF% to GF% is 0.395... FF% is the best we had previously at 0.358

— Stephen Burtch (@SteveBurtch) January 5, 2015

so that's almost a 4% improvement... which is very significant

— Stephen Burtch (@SteveBurtch) January 5, 2015

What this means is that percentage of scoring chances for correlates with percentage of goals for better than anything else we have so far. However this is a descriptive stat, not a particularly predictive one. That is to say, it explains the past much better than it predicts the future.

CF% r^2 = .396, FF% r^2 = .342, SF% r^2 = .324, SCF% r^2 = .318... I excluded the lockout shortened 11-12 and this season.

— Stephen Burtch (@SteveBurtch) January 5, 2015

But if you're trying to predict future outcomes? scoring chances are probably the 2nd least useful thing we have (only goals are worse)

— Stephen Burtch (@SteveBurtch) January 5, 2015

The reason for this is probably scoring chances having a relatively small sample size compared to our possession metrics. So, what's an analyst to do? Scoring chances work well in explaining the past, but not so much in explaining the future. If that sounds kind of familiar to you, it did to me too: It sounds a lot like shooting percentage.

Scoring chances were designed to tease out how well a player does in generating shot attempts that are more likely to go in the net. It makes sense that a player who generates more scoring chances would have a higher shooting percentage. We've seen that scoring chances aren't exactly repeatable year to year. But if your other underlying metrics are about the same, and your scoring chances are about the same, we can be more confident that changes in your shooting percentage are just variance. There will be more uses for this number as we get more work on it, but this is probably the easiest, and it applies really well to the Avalanche right now.

I'm not worried about Nathan MacKinnon

Everything in this section is at even strength. Quick explanation of abbreviations:

• SC - on-ice scoring chances
• CF - on-ice Corsi for
• FF - on-ice Fenwick for
• iSC - individual scoring chances and that's the same for the other two as well
  

To kind of piggyback on what Andi wrote yesterday, let's give an example of using scoring chances in this way to examine Nathan MacKinnon's goal output drop this season. There's a perception out there that his SH% is so low because he's firing low-percentage stuff from a mile away, as opposed to last year when he had a ton of rush shots. With scoring chances we can test that theory.

To control for changes in raw output, it's useful to take scoring chances as a ratio of your overall shot attempts. For that reason you're going to see me in the future using "iSC/iCF" a lot. A high ratio means more of your shot attempts are scoring chances. Easy. It's not that repeatable (Burtch tells me the R-squared is something like 30%), but as a descriptor, it's nice to have.

So let's see where Nathan MacKinnon's offensive output has been struggling by asking all four questions now at his disposal.

1. Is his possession suffering?
2. Is he generating fewer shot attempts?
3. Is he generating fewer scoring chances?
4. Is this just variance, something we should expect to correct on its own with time?

MacKinnon is actually doing slightly, probably insignificantly, better in every area. His on-ice possession is better, he is creating more shot attempts and as a result more scoring chances (that bump in iSC/iCF is probably insignificant [1]), but his EV SH% is in shit creek.

This will correct back upwards with time. He's not suddenly doing the wrong things. It just seems that way because the right things aren't going in like we got accustomed to. We, and the team too, need to be patient with MacKinnon, and give the shots time to go in. They will. In the meantime let's not nerf his assists by putting him out there on a third line with guys who aren't scoring threats, okay? Please?

I'm pretty excited about the way the research into scoring chances is going. It's a #fancystat I like a lot and will be using in the future. It may not be great for predicting future scoring chances; however, it gives us more tools to explain year-to-year changes in a player's output. Right now, it's telling us to chill about Nathan MacKinnon.

[1] In my own population study, I found the average iSC/iCF for forwards (2005-2014) to be 0.618121, with a standard deviation of .083991. MacKinnon is above average, but not enough to enter the fourth quartile.