Sunday, September 20, 2009

Season Preview: Colorado Avalanche


Colorado was a terrible team last season in terms of goal differential and their placement in the standings. However, to what extent they were bad, and to what extent they were unlucky, remains unclear.

At first glance, the latter appears to be true. The Avalanche were absolutely screwed by the percentages in 08-09. At even strength, they ranked 27th in shooting percentage and 26th in save percentage. On special teams, they ranked 25th in shooting percentage and 24th in save percentage. Overall, they ranked 29th in shooting percentage and 28th in save percentage. That’s undeniably rough.

However, the above data doesn’t quite tell the entire story. For one, it’s been demonstrated that shot ratio and shot differential at EV varies according to goal state. Whereas playing from behind tends to increase shot ratio, playing with the lead tends to decrease it. Therefore, shot differential per se may not be indicative of the balance of play for a team that plays a disproportionate amount of time with the lead or while trailing in any given season.

Colorado qualifies as one such team. No team played from behind more than the Avalanche did in 08-09. Not surprisingly, this is reflected in the team’s underlying numbers. Whereas the Avalanche ranked 16th in the league in terms of overall EV shot differential, they were 24th in terms of EV shot differential when the score was tied.

Secondly, while the Avalanche may have been respectable in terms of shot differential itself, they fared quite poorly in terms of Fenwick and Corsi. For reasons that remain unknown, the Avalanche tended to surrender many more missed and blocked shots than their opponents last season. In other words, the Avalanche had many more shots directed at their own goal than they directed at the opposition’s net.

Admittedly, I’m not quite sure what to make of the fact that the Avalanche fared so much worse in terms of Fenwick and Corsi relative to their shot differential. No other team in the league exhibited such an unusual discrepancy. On the one hand, it may suggest that Colorado tended to spend much more time in their end of the rink than in the other team’s end at EV. Such is the mark of a poor even strength team. If the Avalanche were in fact territorially dominated at EV last season, their results may have less to with poor luck and more to do with poor play.

Interestingly, the 07-08 Avalanche were characterized by the exact same profile. Like the 07-08 team, their EV shot differential was much better than their Fenwick and Corsi. Unlike the 08-09 team, however, the 07-08 Avalanche performed quite well at even strength, outscoring the opposition 160 to 143. Unless the 07-08 Avalanche were simply lucky – which I think is unlikely, it’s possible that shot differential is more indicative of Colorado’s EV ability than are Fenwick or Corsi.

Additionally, even if the Avalanche are evaluated on the basis of Fenwick and Corsi, and even if one controls for the fact that the Avalanche tended to disproportionately play from behind last season, it remains apparent that they were an unlucky team. To illustrate this, I’ve prepared a chart below, based on data kindly supplied to me by frequent poster Sunny Mehta several months ago, that shows how each team performed last season in terms of Corsi when the score was tied at EV.


In this chart, I've included two columns that show each team’s shooting and save percentage in terms of shots directed at the net with the score tied. Thus, shooting percentage is calculated by taking the number of goals that a team scored at EV with the score tied, and dividing it by the number of total shots directed at the other team's net. Save percentage is calculated in the exact same fashion, except with goals against and shots against. As depicted in the chart, the Avalanche ranked 22nd in shooting percentage and 24th in save percentage. The point is this: the Avalanche were an unlucky team any way you look at it, regardless of the particular approach or method of analysis.

All things considered, the Avalanche were likely a below average team that also happened to be subjected to an inordinate amount of bad luck. I suspect that they’ll be less awful this year just on account of the fact that they couldn’t possibly have a worse year with the percentages than they did last season. For the exact same reason, they’ll likely do better than most people expect them to. That said, the team lost one of its best forwards in Ryan Smyth in the offseason and had its second line centre retire. They’re worse on paper than they were at the beginning of last season and I’d be surprised if they made the playoffs.

10 comments:

Sunny Mehta said...

I like these season previews you're doing.

Couple things about Colorado...

First off, it's possible they had below-average talent in terms of shooters and goaltending. Nevertheless, in looking at previous seasons (and using a bit of common sense), I'd be pretty confident predicting a regression. They also added Craig Anderson in net for this season.

Secondly, the ES-tied data I sent you was sorted by Corsi ratio (i.e. - shots directed divided by shots directed against). If you look a couple columns over, you'll see what i dubbed "Zone Ratio" which is [shifts ending in offensive zone faceoffs] divided by [shifts ending in defensive zone faceoffs]. Sort by that column. Notice anything?

While Colorado was dead last in Corsi ratio with the score tied, they were pretty close to average in Zone ratio. Very weird. Corsi says they were dominated territorially, but Zone ratio says otherwise. Of course, both are really only proxies for "territory", so I'm not sure what the truth is.

The interesting thing is, if the Avs were a team that did in fact control territory better than their corsi indicates, you'd think that would actually HELP their percentages. (I.e. - they were in the offensive zone plenty but chose to be more selective about their shots, etc.)

JLikens said...

First off, it's possible they had below-average talent in terms of shooters and goaltending. Nevertheless, in looking at previous seasons (and using a bit of common sense), I'd be pretty confident predicting a regression. They also added Craig Anderson in net for this season.

While I'm inclined to agree with you about the goaltending, keep in mind that deviations in EV shooting percentage when the score is tied are entirely random. There's no skill component at the team level (or, as Vic described it, there are no 'real effects').

Colorado shot 0.065% at EV last year when the score was tied. Given what we know about EV SH% when the score is tied -- namely, that it's randomly distributed -- that can be attributed solely to bad luck, rather than any ineptitude on the part of Avalanche shooters.

Of course, in terms of their overall EV SH%, part of the reason why it was so low was due to the fact that they were chronically playing from behind.

And I agree that Anderson should help them in goal (provided that he gets sufficient playing time).

"Secondly, the ES-tied data I sent you was sorted by Corsi ratio (i.e. - shots directed divided by shots directed against). If you look a couple columns over, you'll see what i dubbed "Zone Ratio" which is [shifts ending in offensive zone faceoffs] divided by [shifts ending in defensive zone faceoffs]. Sort by that column. Notice anything?

While Colorado was dead last in Corsi ratio with the score tied, they were pretty close to average in Zone ratio. Very weird. Corsi says they were dominated territorially, but Zone ratio says otherwise. Of course, both are really only proxies for "territory", so I'm not sure what the truth is."


That's very interesting about their zone ratio with the score tied being essentially average.

Taken together with the shots data, it tends to support the idea that the Avalanche weren't, as you said, that bad of an even strength team, that corsi and fenwick are misleading with respect to their ability as a team at EV, and that the Avalanche were substantially unlucky as opposed to merely moderately unlucky.


The interesting thing is, if the Avs were a team that did in fact control territory better than their corsi indicates, you'd think that would actually HELP their percentages. (I.e. - they were in the offensive zone plenty but chose to be more selective about their shots, etc.)

As intuitive as that may be, it's also true that teams that block more shots tend to have lower save percentages (see, for example, this post by the Contrarian Goaltender). So if there is such an effect, it may be cancelled out and/or overridden by other factors.

Anonymous said...

That is interesting to see that teams trailing shoot more. Do you think shot quality is lower though, as teams playing from behind will more frantically put pucks on net in desperation?

JLikens said...

"That is interesting to see that teams trailing shoot more. Do you think shot quality is lower though, as teams playing from behind will more frantically put pucks on net in desperation?"

That appears to the case, yes.

See, for example, this article written by Tom Awad at PuckProspectus, which shows that the increase in shot ratio experienced by the trailing team is met with a corresponding decrease in shooting and save percentage.

Sunny Mehta said...

JL,

Imo Vic's research simply showed that one season is not a big enough sample size to conclusively identify shooting percentage skill in the wash of randomness. I don't think it means that every team is identical in their "true" shooting percentage w/score tied, but just that we can pretty much never know.

If every team played a thousand, or a million games this season, would every team have a S% of 7.5 with the score tied? Maybe. I doubt it. If someone offered me a bet right now of who will have a higher ES-tied S% this coming season, the Penguins or Islanders, but I had to lay 49-to-51, I'd bet the Pens in a heartbeat.

Back on point, do I think it's likely that the Avs are a "true" below-average S% team? No. There's basically no evidence for it, and I definitely would not bet on it.

It's interesting, we're dealing with SUCH small samples here it's ridiculous. Even over the whole season, the average shots per team at ES w/game tied was something like in the 600-700 range. Every team is within the 95%-confidence intervals of league average S%.

Sunny Mehta said...

oh also, i don't understand your reference to the blocked shot thing. what am i missing?

JLikens said...

It's possible that you're right. I mean, from an intuitive standpoint, I'm inclined to agree with you that not all team's have the exact same 'true' EV SH% with the score tied.

However, it's difficult to disagree with what's been demonstrated thus far, namely that:

1. 100% of the variance in team EV SH% with the score tied in 07-08 was due to randomness.

2. 100% of the variance in team EV SH% with the score tied in 08-09 was due to randomness.

3. The correlation between team EV SH% with the score tied in 07-08 and the team EV SH% with the score tied in 08-09 was 0.004.

And on top of that, there's Vic's analysis, which looked at the average correlation between pairs of 38 randomly selected games (for each team in 08-09) over 1000 simulations (and found none).

While more analysis may be required, my suspicion is that the data is correct and that there are no real effects.

JLikens said...

RE: Blocked Shots

Sorry, I should have been more clear about that.

I suppose my point -- and I was merely speculating -- was that because there's a negative correlation between shotblocking and save percentage, a team that consistently blocks a higher proportion of shots than its opponent (such as the Avalanche) may not experience any advantage to its percentages, even it is more selective than the opposition in terms of shot selection.

Bruce said...

One odd thing about Colorado's 5v5 shot data in 2008-09 that I noticed partway through the season. Their shot differential was relatively small, just -37, yet their Fenwick was over five times greater (-202) and their Corsi more than 12 times greater (-445).

Put another way, here are the ratios for/against:

Shots: 1827/1864 = 98.0%
Missed shots: 645/810 = 79.6%
Blocked shots: 780/1023 = 76.2%

I would have thought that these ratios would be fairly close. Wrong in Colorado's case. Given that, I would expect that GF/GA results would track much more closely to shots on goal as opposed to shots that didn't make it through. Wrong again, as the goal ratio of 131/171 = 76.6%, fairly similar to the missed and blocked shots ratios. In fact, Colorado's opponents made more saves than Colorado did, despite the fact that the Avalanche attempted 445 fewer shots than their opponents.

I have no plausible explanation, just thought this data was weird enough to mention. Sorry it's so late.

JLikens said...

Yeah, it's definitely quite puzzling.

I think that with Scott counting scoring chances for the Avalanche this year, it'll interesting to see how Colorado's EV scoring chance ratio compares to its corsi ratio and its EV shot ratio.