On the scoring chance project, and why it must continue

Cam Charron
June 28 2012 02:29PM

 

 

Whatever tendency certain players might have for driving their team to get more scoring chances than a simple shot differential predicts is small and swamped by random noise. This suggests tracking scoring chances isn't adding much information to the readily available shot differential numbers.

That's from Eric on this very blog earlier this week. I'm willing to believe that, too. Over the course of the season, when I was tracking chances for Vancouver, I found that when the Canucks out-chanced a team, they were really not in a position to win the game all the time.

Looking at just even strength numbers, ahead. Since I can't get game-by-game scoring chance and shot totals without spending my entire summer hunched over my computer, I went back through all 82 Vancouver Canucks games that myself and Thomas Drance tracked this year.

Side-by-side, they produced some interesting results.

And, yes, the team's overall Fenwick number and scoring chance number formed a certain bond by the end of the year. The Canucks finished the season with a 51.7% Fenwick rate and a 50.3% scoring chance rate. That's a sizeable-enough difference, but if you look at the graph, both red and blue lines come together at about 34 or so:

That's over the course of the season.

I'm pretty well convinced that, after 82 games, judging players by scoring chance numbers versus possession numbers such as Corsi or Fenwick doesn't make a damn lick of difference.

However I split some of these games up. Since I was only looking at even strength, there's a possibility that there could be ties. Some games ended 2-2 or 1-1 in even strength goals, and I was looking for the team's win, loss or tie record at even strength to determine whether scoring chance statistics are better predictive than Fenwick.

And...

  Wins Losses Ties Win%
Shots 50% + 18 16 9 0.529
Chances 50% + 23 10 7 0.697
Both 50% + 13 8 7 0.619

The Canucks did much better in games where they out-chanced the opposition rather than simply out-shot them. For sure, the out-shot record is variant on score effects, but it doesn't seem that scoring chances were affected so much by this.

What about games where the Canucks out-shot but didn't out-chance the opposition? They were 3-5-2 in those games:

  Wins Losses Ties Win%
Shots, not Chances 3 5 2 0.375
Chances, not Shots 9 1 0 0.900

Shot metrics, at least those that haven't been adjusted for score, are clearly not as good over small game samples for determining victories.

Another interesting thing to note is that the Canucks save percentage in games where they out-chanced their opponents' was .931. Their save percentage in games where they were out-chanced was .930. On offence, there was a bigger difference, 8.8% versus 7.4%, so I'm more inclined to think that there's some form of shot quality that exists on offence, but not so much on defence.

There's still some use to counting chances. Maybe they shouldn't be analyzed with as much rigour at the end of the season, since corsi will do the trick, but, again, on a single-game basis, at least for the Canucks, scoring chances seem score effect-neutral so you have more events with which to judge a player's performance.

63811cbf517d2d685ea09e103488ea3a
Cam Charron is a BC hockey fan that writes about hockey on many different websites including this one.
Avatar
#1 Sam Page
June 28 2012, 02:34PM
Trash it!
0
trashes
+1
0
props

I don't think these results are especially surprising.

Now the question becomes: is all this work still worth it, just to have better metrics for small sample sizes that might not be indicative of true talent anyway?

Avatar
#2 Josh L.
June 28 2012, 02:43PM
Trash it!
0
trashes
+1
0
props

@Sam Page

In fairness, it isn't really much work if you're watching the game anyway.

Avatar
#3 Tach
June 28 2012, 04:52PM
Trash it!
0
trashes
+1
0
props

Isn't the finding that generating more scoring chances more highly correlates to winning than generating more shot attempts/possession somewhat tautological since a scoring chance is basically just a shot attempt with a higher likelihood of going in?

If a shot attempt went in 99% of the time, then having more shot attempts would virtually guarantee you a victory. At 1%, you would need many, many more attempts for an increased number of shot attempts to substantially influence outcomes.

Shot attempts go in at something like a 3-6% rate and scoring chances at about 25%, so finding that having more scoring chances correlates more closely to winning is just proving that people were fairly recording scoring chances.

The million dollar question remains whether a team can repeatably generate scoring chances at a rate disproportionate to their shot attempts. All evidence remains to the contrary.

Avatar
#4 Kent Wilson
June 28 2012, 06:41PM
Trash it!
1
trashes
+1
1
props

Shot attempts go in at something like a 3-6% rate and scoring chances at about 25%

All true except the scoring chance success rate seems to settle around 15% - that's including missed and the like of course.

Avatar
#5 Jared Lunsford
June 28 2012, 08:45PM
Trash it!
0
trashes
+1
0
props

Why not use goals, by this logic?

Avatar
#6 Ralph
June 28 2012, 10:57PM
Trash it!
0
trashes
+1
0
props
Jared Lunsford wrote:

Why not use goals, by this logic?

In the super long run goals tell the story. Going from goals-->chances-->shots decreases meaningfulness but increases sample, and the trick is to balance the meaning of each data point while not cutting back the sample too much so that your results are swamped by variance.

Comments are closed for this article.