SF/SA Ratios in Different Goal States
by Tyler Dellow • March 31, 2009 • Uncategorized • 10 Comments
This has come up in the comments to one of Dennis’ scoring chance posts and I’ve actually had someone email and ask me what I thought about it, so I thought I’d throw up the chart now.
What you’re looking at is each team’s ratio of ES SF/SA for 2003-2008, based on the game state at the time. As you’ll see, there’s some pretty solid evidence that teams sit back when they’re leading. I’ll do a lengthier post when I have more time but enough people have asked about it that I wanted to get it up. Of note: I am not at all convinced that it’s a good strategy.
“
Fascinating stuff… that Detroit line explains a lot.
it’d be helpful to see each team’s overall ratio, so that we can see how they deviate from their own average at each game state.
comparing between teams seems less useful. i.e. – comparing Detroit with a one goal lead to, say, Minnesota with a one goal lead isn’t quite fair because Detroit is more likely to outshoot at any game state than Minnesota is.
Also, as for you not being convinced, I definitely understand your skepticism, but I think there might actually be some logic behind the strategy.
First off, there’s the obvious shot quality implications.
Secondly, we are still at such an infancy with even making sense of our large-scale data glut in hockey that it will probably be a while before we get into the nitty gritty of win probability, which is really actually the bottom line, isn’t it? And I think that’s what is at the heart of this. More specifically, there are times when a team should essenitally be willing to give up goal equity to curb variance.
For example, in the long run would you rather have a team full of 5gf60/3ga60 guys or .8gf60/.9ga60 guys? Of course the former. But now let’s say your team is leading by two with ten minutes left in the game. I imagine, at this point, you’d rather have the latter guys because giving up a goal hurts your win probability more than scoring a goal helps it.
That’s the thing… hockey games have a binary outcome, and that sometimes gets lost in the averages.
What I find interesting is that the relationship appears to be linear, and that it (generally) holds for even the extreme values (+5/-5).
Granted, sample size issues are relevant for blowout games at the team level, so the relationship only emerges in aggregate.
I’m not sure if anyone here will find this interesting, but I’ll post it regardless.
I ran the Oilers scoring chance numbers (as compiled by Dennis, who’s done some awesome work here, btw). I cut and pasted the information from every game played since 12/13/08 (against Vancouver) into excel and wrote a couple formulas in order to analyze the data. Specifically, I was interested in the effect of leading/trailing on scoring chances.
Here are the results for the Oilers:
OVERALL
When Trailing:
263 Chances For — 308 Chances Against
When Tied:
320 Chances For — 366 Chances Against
When Leading:
336 Chances For — 368 Chances Against
AT EVEN STRENGTH ONLY
When Trailing:
200 Chances For — 229 Chances Against
When Tied:
262 Chances For — 278 Chances Against
When Leading:
237 Chances For — 269 Chances Against
The results may not be strictly accurate, but I’ve checked the results for a few of the individual games without finding anything wrong.
Not counting the ENG by Ana – for obvious reasons – I had the Oilers outchancing the Ducks 7-1 in tonight’s third period with the Oilers trailing heading into that frame.
So, just looking at third periods, that’s four of the last six games where the trailing team’s held the advantage in chances for/against.
d09ed181d182d0b0d0b2d18cd182d0b5 d181d0b2d0bed0b9 d0bad0bed0bcd0bcd0b5d0bdd182d0b0d180d0b8d0b9 d092d18b d0bcd0bed0b6d0b5d182d0b5 d0b8d181d0bfd0bed0bbd18cd0b7d0bed0b2d0b0d182d18c d181d0bbd0b5d0b4d183d18ed189d0b8d0b5 HTML d182d18dd0b3d0b8:
Perhaps another factor to take into account, is that the trailing team is much more likely to attack aggressively when down by two or more goals.
It may not always be the team leading by two hanging back as much as it is the trailing team throwing caution to the wind, and gaining more shots on net, at the expense of giving up a higher quality of shots against.
Likely a mix of both.
Even if the strategy of sitting back after leading isn’t the best possible, it still appears to be very effective.
29 out of 30 teams this season are over .500 when scoring first. About 2/3′s of teams have around a .750 winning percentage when leading after two, and only 8 teams have 3 or more losses when leading after two periods.
Bank Shot:
29 out of 30 teams this season are over .500 when scoring first. About 2/3’s of teams have around a .750 winning percentage when leading after two, and only 8 teams have 3 or more losses when leading after two periods.
True, but how much of that would still hold if every team scored/allowed goals at constant rates independent of the game state?
MC: Thanks for this. Not sure how to connect the dots between A —> Z, but surely there are implications on all individual metrics like shots differential, Fenwick and Corsi.
It looks like my original numbers are wrong — there was in fact an error in my formula.
Corrected Numbers for the Oilers (as of the last Anaheim game)
===============================================
OVERALL
TRAILING: 285 FOR — 266 AGAINST
TIED: 348 FOR — 365 AGAINST
LEADING: 268 FOR — 401 AGAINST
EVEN STRENGTH
TRAILING: 218 FOR — 201 AGAINST
TIED: 280 FOR — 285 AGAINST
LEADING: 201 FOR — 290 AGAINST
===============================================
So Dennis’ perception that the team playing from behind tends to win the scoring chance battle appears to be more or less correct.