Friday, May 14, 2010

Shot Quality and the Montreal Canadiens

I fooled around with tracking shot quality in last year's playoffs, using shot charts to try to get a sense of where teams were shooting from. One of the most interesting sets of numbers that I happened across was at CBS Sportsline, where they track where a shot was targeted on goal (e.g. low glove, low blocker, five-hole, high glove or high blocker).

This data showed that high shots were much more likely to go in than low shots. The problem was that it was very unevenly done. Some games had no high shots recorded at all, others had just a few, while others seemed like they had a more reasonable number. I ended up just filing it away as something to potentially check on if I was trying to assess the shot quality of a particular team or series of games.

A lot of people are trying to explain why the Montreal Canadiens are having so much success this postseason, and one of the factors they typically point to is that the team is forcing their opponents to shoot from the outside. Often people talk about shot quality when they want to try to justify why one team is running hot with the percentages. In the long run teams tend to converge to the average. However, I think it is entirely possible that shot quality could be a factor when one team is playing against a single opponent over the course of a playoff series.

I thought to look at the high/low CBS numbers for Montreal's series against both Pittsburgh and Washington to see what they tell us. If Montreal's skaters are legitimately doing something to impact the other team's shots, then we would expect to see their opponents have a higher percentage of low shots than the Habs. If this is a bunch of talking head garbage, then the percentages are going to be similar for both teams and Montreal was just lucking out.

What do the data tell us?

Montreal vs. Pittsburgh:

High shots: Pittsburgh 45, Montreal 45
Low shots: Pittsburgh 180, Montreal 128

Montreal vs. Washington:

High shots: Washington 73, Montreal 62
Low shots: Washington 216, Montreal 131

Wow. I'd say that at least warrants a closer look.

It's important to note that we have to be careful with this data. It still seems like whoever tracks this stuff sometimes just falls asleep for a period here and an entire game there. For example, in game one of the Pittsburgh series just 2 out of 55 shots were recorded as being high, and in game three against Washington just 3 of 77 shots were marked down as high shots. Several other games were suspect (games 6 and 7 against Washington, games 3 and 5 against Pittsburgh), all of whom had a total of 8 high shots or fewer for both teams.

I'm inclined to throw those games out, and rerun the results.

Montreal vs. Pittsburgh (G2, G4, G6, G7 only):

High shots: Pittsburgh 37, Montreal 39
Low shots: Pittsburgh 116, Montreal 53

Montreal vs. Washington (G1, G2, G3, G5 only):

High shots: Washington 60, Montreal 56
Low shots: Washington 98, Montreal 61

Montreal won 6 out of these 8 games, so these games may not be exactly representative of the two series. However, it is a good sample to try to figure out whether the Habs' defensive tactics were effective. This evidence certainly suggests that they might have been, assuming that this data is good. Whether it was because they had more pressure on the puck, or because they were clogging the shooting lanes, or because they had more counterattack opportunities or odd-man rushes, these numbers are evidence that Montreal's shots may have been of higher quality than those of their opposition, particularly against the Penguins.

Even if that was true, that of course does not mean there was no luck involved. The Habs have benefitted from a healthy dose of good fortune, especially against the Caps. Today's goalies are very good at taking away the bottom of the net, but low shots aren't by any means harmless, especially ones taken close to the net. Think of Halak's pad save on Evgeni Malkin in the third period of game 7, for example. The Habs have survived a bunch of similar chances in the playoffs.

Montreal is not likely to sustain either their shooting percentage or save percentage numbers so far (13.7% on high shots and 10.5% on low shots in the second sample, compared to 9.3% and 6.1% respectively for their opponents). I'd say they should still be the underdog in the Eastern Conference Finals against whoever wins tonight. Yet maybe there is at least something to Jacques Martin's madness after all.

23 comments:

Derick said...

Do you know where I could get information on even strength shots and even strength goals against, specifically for the Canucks series? Thanks.

R O said...

It sounds a lot like "shooters didn't make their shots", especially when paired with the scoring chance data.

YMMV on that.

Topham said...

Nice work. Good eyes to remember CBS sportsline.

The Contrarian Goaltender said...

Derick: NHL.com has it. Go to individual stats, goalies, special teams.

The Contrarian Goaltender said...

Shooters not making their shots is surely some of it. But I'm not sure it accounts for all of it, particularly against Pittsburgh.

RO: Did Gabe or anyone else ever run shot quality data on the Pittsburgh series? Shot distance, expected goals, that type of thing? To me against Washington it looked like the Habs were consistently lucky to keep the puck out of the net. Against Pittsburgh the Penguins didn't seem nearly that dangerous, even when they were piling up scoring chances with the Habs holding onto the lead. But we all know that memories are faulty things, so I don't know.

Sunny Mehta said...

Even if it's empirically true that there were more low shots taken than normal, that doesn't necessarily say anything about Montreal's ability to induce low shots. For that you'd have to show that low-shots against percentage has any kind of predictive value.

In other words, Montreal's "luck" may be partly encompassed in the fact that they gave up a shit ton of shots that happened to go low.

overpass said...

Chris Boersma at HockeyNumbers ran expected goals for all playoff series so far.

http://hockeynumbers.blogspot.com/2010/05/pit-vs-mtl.html

Olivier said...

Overpass: Nice catch.

Hawerchuk said...
This comment has been removed by the author.
Hawerchuk said...

Phil,

Much as I like the CBS database, I'm very skeptical of its accuracy. I've only gone through Game 1 of the MTL-PIT series, but I think the scorer is on crack.

First shot of the game:

NHL: Staal - 1:26, X=11, Y=30
CBS: Staal - 1:20, X=21, Y=17

At 1:20, the Pens still have the puck at their blue line. Six seconds later, Staal is down the ice, and he takes a very bad angle shot from the right wing. In other words, the NHL feed is correct and the CBS feed is wrong in every imaginable way. (Although they did record that it was a low shot, which was correct.)

I just don't know what the hell the CBS guys are doing, because they literally agree on only 10 shot times over the course of the entire game with the NHL/ESPN feed.

I was going to match up the scoring chances with shot location (X-Y and 1-5 holes) but CBS is just a disaster.

The Contrarian Goaltender said...

Gabe: I agree with you about CBS, I put some disclaimers in the post but still should have probably emphasized even more that we're dealing with a very questionable source. This is certainly not conclusive proof of anything.

I was actually expecting the high/low shots to fall pretty much in line with the scoring chances, but they didn't so I thought it was worth throwing it out there to see if anyone had any comments or could make something of it.

Hockey Numbers had the expected goals in the Washington series at 28.9-17.0 and the Pittsburgh series just 15.5-13.1. I think it is absolutely fair to say that the Habs' win over Washington was a fluke and very lucky, but I still am not convinced that scoring chances are fully representative against Pittsburgh. Olivier had it at 158-89, and I don't think that truly reflects the play. Again not that the Habs were better or more deserving of the win, just that they weren't too far behind the Penguins and their tactics had a big impact on the shot numbers.

The Contrarian Goaltender said...

Re: Game 1 of Pens-Habs, I agree that scorer wasn't doing his job. He had 53 out of 55 shots marked down as low shots, that's one of the games I threw out. I ran into other issues like their shot charts didn't always match the shot totals for the game, which didn't always match the NHL's totals, and so on.

I'm not sure whether they get it right some of the time or if it's all screwy. My sense is that it looks like the CBS intern or whoever it is knows what he is doing on some nights and some nights he's studying for an exam or something. If you are going to look in some more depth, I'd recommend looking at some sample of the eight games that I picked out above.

JLikens said...

Interesting stuff.

And just as a general point, the expected goals figures from hockeynumbers don't take goal state into account (i.e. which team was leading or trailing at the time).

That might explain part of the discrepancy between the numbers from the Pittsburgh and Washington series.

In the former, the Habs trailed about as often as they led. In the latter, they played with the lead for a large part of the series and rarely trailed.

Hawerchuk said...

I'll check out some of the other games to see how well they match up. I'd love to add shot target location to the shooting model if it has some predictive value.

One thing to keep in mind is that a big chunk of the shots CBS records as 'high' are taken by defensemen from the blueline. So the real question might be whether guys got their shots up from close in.

I have the expected goals in the MTL-PIT series at 21.5-13.3 for PIT. This is one of the worst parts of the shot quality models - they don't agree in the least.

Tom Awad said...

Hawerchuk, we're actually pretty close. Chris had them after game 6 at 15.5 - 13.1 PIT; I had 15.2 to 12.9. In game 7, the Pens dominated expected goals, so after game 7 my totals are 19.1 to 14.7. Note that I don't count empty-netters.

JLikens said...

Do either of your models take game score into account?

I don't think that it would affect things that much in the case of your totals for the series; I'm just interested.

The Contrarian Goaltender said...

Thanks for that Tom, I didn't notice that Chris' totals so far only ran through game 6.

Hawerchuk said...

hah, I too did not notice that Chris didn't have game 7.

My model does not include score effects - they might have a significant impact on game 7. It also doesn't incorporate giveaways - Pittsburgh had more really bad defensive zone ones, though they may not be in the PBP.

I also removed empty-netters.

Bruce said...

And just as a general point, the expected goals figures from hockeynumbers don't take goal state into account (i.e. which team was leading or trailing at the time).

That is an extremely important point, one that seems to affect everything from shots to Corsi to scoring chances to, presumably, expected goals.

I note that Javageek had Montreal with more expected goals than Pittsburgh in two games, both of which were Montreal losses. In the four games Montreal was "expected" to lose, the Habs actually won three. Presumably Game 7 was more of the same, given the 18-3 shots count in the third period alone. Of course those shots were driven by what I call "game state" or JLikens calls "goal state". It seems to be a gigantic factor in all of these things, to the point that if the model doesn't take it into account, the model is probably wrong.

Tom Awad said...

My shot quality model takes game score into account, but Montreal's defensive strategy is so extreme that it doesn't change that much: when you're getting outchanced 2-1, an extra 10% on your expected goals ain't gonna cut it.

Montreal Canadiens said...

With the new coach of the Montreal Canadiens I think they will make this time.Jacques Martin was a perfect coach.

yanmaneee said...

pg 1
yeezy
nike sb dunks
golden goose sale
yeezy boost 350
kyrie shoes
golden goose
pg 4
yeezy boost 350
yeezy 350

Unknown said...

h9s90z8a10 p6z27g8s63 j7t26l8q98 p9z74w8u92 r0f91l9b90 l2a57k6t90