Washington-Montreal Game-By-Game By the Numbers
by Hawerchuk on May 4, 2010 9:00 AM EDT
Email
PrintWelcome to our next installment of "all Capitals, all the time." Commenter TMS wondered if Washington did better over the course of the series but did poorly in game 7. If we look purely at shot location, it doesn't look that way:
| MTL | MTL | MTL | WSH | WSH | WSH | |
| SF | EXG | ACG | SF | EXG | ACG | |
| 415 | 58 | 4.22 | 3 | 66 | 5.44 | 2 |
| 417 | 31 | 2.20 | 5 | 54 | 3.74 | 6 |
| 419 | 32 | 1.95 | 1 | 46 | 3.25 | 5 |
| 421 | 42 | 2.62 | 3 | 49 | 2.97 | 6 |
| 423 | 38 | 2.26 | 2 | 46 | 3.40 | 1 |
| 426 | 35 | 2.10 | 4 | 69 | 4.47 | 1 |
| 428 | 27 | 2.10 | 2 | 52 | 3.91 | 1 |
| 263 | 17.45 | 20 | 382 | 27.18 | 22 |
EXG refers to the expected number of goals based on where shots were taken; ACG is the actual number of goals scored. I did not include blocked shots in the shot total - things would look even more lopsided if I did.
Washington got more chances in every single game, and had an average expected shooting percentage of 7.1% to 6.6% for Montreal. If you haven't been over to the best Nordiques blogue on the internet, you should check it out, because it lists observed scoring chances for every Habs game. Washington, of course, won the series there too:
| TOT | 5v5 | 5v4 | |
| WSH | 189 | 130 | 48 |
| MTL | 126 | 89 | 28 |
Washington also recorded a scoring chance on 49.5% of its shots, while Montreal was at 47.9%. So there goes the notion of Montreal having better scoring opportunities on average. Washington also won the scoring chances in every single game:
| WSH | MTL | WSH | MTL | |
| 5v5 | 5v5 | 5v4 | 5v4 | |
| 1 | 23 | 18 | 11 | 1 |
| 2 | 19 | 13 | 4 | 3 |
| 3 | 18 | 9 | 4 | 5 |
| 4 | 17 | 13 | 4 | 6 |
| 5 | 13 | 10 | 8 | 6 |
| 6 | 17 | 15 | 13 | 4 |
| 7 | 23 | 11 | 4 | 3 |
But Gabe, you say, what about rebounds? Surely Don Cherry was right that the Caps weren't getting a guy in front of the net! Here are rebounds by game by how closely they followed the previous shot:
| MTL | 0s | 1s | 2s | 3s | 4s | WSH | 0s | 1s | 2s | 3s | 4s | |
| 415 | 0 | 1 | 0 | 0 | 4 | 415 | 1 | 0 | 2 | 1 | 3 | |
| 417 | 1 | 1 | 0 | 0 | 0 | 417 | 0 | 1 | 0 | 1 | 2 | |
| 419 | 0 | 1 | 0 | 0 | 0 | 419 | 0 | 2 | 0 | 1 | 0 | |
| 421 | 0 | 1 | 1 | 1 | 3 | 421 | 0 | 0 | 1 | 1 | 2 | |
| 423 | 0 | 1 | 0 | 0 | 1 | 423 | 0 | 1 | 1 | 0 | 1 | |
| 426 | 0 | 0 | 1 | 0 | 1 | 426 | 0 | 1 | 1 | 1 | 2 | |
| 428 | 0 | 1 | 0 | 0 | 1 | 428 | 0 | 0 | 3 | 5 | 1 | |
| Tot | 1 | 6 | 2 | 1 | 10 | Tot | 1 | 5 | 8 | 10 | 11 |
Washington won the series overall here too, but didn't win every game. For what it's worth, they did win game seven on rebounds.
To sum it all up - in game 7, Washington:
- out-shot Montreal 52-27 at even-strength and on the power-play
- took shots from better locations than Montreal and expected - on average - to win 3.9-2.1
- out-chanced Montreal 27-14 at 5v5 and 5v4
- out-rebounded Montreal 9-2
There's a reason why Jaroslav Halak was the first star in that game.
102 comments
|
1 recs |
Do you like this story?
Comments
Funny how all these stats just confirmed what everyone knew watching the game
Stats are nice but just watch the game and one can figure out what happened
just watch the game and one can figure out what happened
I think Gabe’s shown that that is not the case at all.
Pension Plan Puppets: A Toronto Maple Leafs blog and a group therapy session.
Like reading thoughts confined to 140 characters? I'm on Twitter too.
by PPP on May 4, 2010 9:06 AM EDT up reply actions 3 recs
actually there are plenty of people who are trying to see this as something that was a one-sided, dominated team getting bailed out by its goaltender and luck, and are instead trying to point to the Caps “system” or to the Capitals lack of “leaders” or other intangibles. BS. These stats have been showing that if this series were played the same way 10 times, 9 times the Caps would win.
Glen Sather is a Hockey Genius.
http://glensathersucks.com/
http://twitter.com/ThGeneralissimo
Actually it did tell me one thing I already knew: shot-based metrics are plainly not the be-all end-all measure of hockey goodness some people are making them out to be.
Wait, is that an ex post facto explanation or are you being serious?
Analyzing single-game performance is very difficult. This is the holy grail in the NHL – if you have a system, you’ll get the ear of every NHL team!
I’m sure I would. I don’t have the model, of course — but I do have a bit of a problem with presenting the shot-based model as if it were that. I imagine that’s not the real intent, but in a way, that’s the way it’s coming across.
I’m actually somewhat serious. I’ve been somewhat sketpical of shot-based metrics in the past. I’d started to come around as I started to understand more and more what they mean, but the way they’re liberally being spread out and made to say things they really can’t say, especially when everything else is being dismissed — one might even say “derided” — as “luck”… well, frankly, it loses credibility, as if the only tool we had was a hammer and we were trying to see everything as a nail. As a tool it has lots of value, especially on aggregate, but it doesn’t seem to me like it’s close to telling the whole story, and even less on a per-game basis. I’m convinced Corsi has value as a puck-possession proxy metric, but let’s not turn it into a religion either. There are other factors that affect winning games, and the value of shot-based might even differ from team to team because of strategic approaches (such as the propensity of taking shots by point men).
As an aside, the other question I’ve been asking myself is whether it might not easier, in the short term (say, a playoff series), to affect percentage metrics positively rather than shot metrics — and since you’re more likely to get mileage out of percentage improvements than shot improvements, that it might not be a rational approach for coaches to take.
At a team-season level, Corsi explains 40% of winning. Add to that goaltending, turnovers, penalties and shot location, and you might be able to explain 60% of what happened. So there’s a big hole there, which I have never denied exists.
The reason people believe luck plays such a large role is that if you simulate the game purely as a random event, the spread of outcomes is not much smaller than the true spread in many areas. Some “talents” don’t appear to persist from day-to-day. If you can’t control your own skills, does it really matter if it’s luck (ie – random skill variation) or a short-term deterministic issue? I mean, a hockey player has to execute – if he might show up for a game with less skill than expected, that’s not a point in favor of being able to affect certain metrics in the short-term.
That’s largely due to the scoring context of hockey is it not? Where such a small amount of events are the detereming factors in a game.
I don’t think I would even want to look at a game to game analysis system. At least on the individual level, because the sample would be so horribly skewed no matter what “skills” or “events” you’re trying to isolate even if you were able to isolate something like passing.
Question Gab, that 60% of “goaltending, turnovers, pentlaites and location”, did you just wing that or it’s actual but you don’t know how to divide it?
Strictly speaking, you’re right. The result of a coin toss isn’t driven by “luck.” If you could perfectly model the force on the coin, air resistance, the elasticity of the floor etc, a coin toss would be 100% predictable. In the real world, we can’t perfectly model a coin toss in the field. So although the result of a coin toss may not be “luck,” it is unpredictable. Because it is unpredictable, we say the ability to win a coin toss is luck.
NHL hockey games are weighted coin tosses.
Exactly. And then you see a person that tosses 60% heads (eg. wins) and then you “dismiss” it as luck — random variance.
And then you find out that there is such a thing as coin-tossing skill. http://www.cmaj.ca/cgi/content/full/181/12/E306
Thanks for the link – it makes the perfect case for weighted coin tosses.
If someone has the ability to repeatably win 60% of events, we call that skill. The fact that we don’t know whether the next event will fall within the 60% or the 40% is the random variance.
The existence of people “able to successfully manipulate the toss of a coin” is not evidence that luck does not exist. The Detroit Red Wings of coin tossers lost almost a third of the time. Luck is the part that decides which 1/3 are tails and which 2/3 are heads. If the Detroit Red Wings of coin tossers loses 4 of 7 tosses, we say he had bad luck. He had an 85% chance of winning but this time the 15% chance came through.
Some observers will step forward with a long list of character flaws to explain why he failed to get heads four times out of seven despite his usual 68% success rate. It’s becoming apparent that these observers will never change their minds.
Conversely, the fact that you don’t model something is not evidence that it is random. It could be, and from a modeling standpoint it makes sense to treat it as such, but in reality, we don’t know.
Part of my point, I guess, is that, because “luck” is a loaded term in hockey circles (fans bristle at it like at little else), maybe it’s not the ideal way to describe things that fall outside the model. Call it “variance” and it’ll sound all scientific and will be a lot more value-neutral. People understand that if you put a shooter in the slot and have him shoot pucks at a goalie, sometimes he’ll score and sometimes he won’t. What people don’t understand is why that should be called “luck”. It’s not luck in the sense of a random coin flip — it’s in large part an adversarial test of skill that leads to something that resembles a probability (if I were the one shooting, I’d never score).
The other thing is that I think there’s an assumption that teams cannot reliably affect percentages with skill or strategy and I’m not sure this assumption is correct over a small number of games. Certainly, the idea that save percentage can be affected by skill to some degree, especially over a small sample, is (mostly) uncontroversial.
I would disagree. The idea that save percentage can be affected over a LARGE sample is uncontroversial. The idea that it can be affected over a small sample is generally held to be false, although any evidence to the contrary would be interesting!
What ‘everyone knew watching the game’ was that shot blocking was key and Hal Gil was a monster and that Montreal’s bionic defensive scheme was the be-all end-all of winning playoff hockey. That’s what the media, and ‘everyone’, knew.
What Gabe just demonstrated is that all those factors had pretty much no impact, that Washington dominated on pretty much any metric imaginable, and that Halak stoned them. Not sure that’s what everyone was thinking.
Well, considering the sheer number of shot blocks, I would certainly argue that it helped the Canadiens’ cause, and that Gill was a bit part of that. That being said, anyone who understands how the game’s played realizes that spending two or three times as much time in the defensive zone as the offensive zone is a shitty way to win hockey games, and that while all those blocks are nice, they’d probably have been much better off if they hadn’t had to block 40-odd shots in Game 7 in the first place.
SNN Sports - A theoretical Oilers blog (i.e. theoretically, I write stuff there). Link now 100% less broken.
Although because Montreal led most of Game 7 (and indeed much of the series), those values are tilted a bit by score effects, too.
In a Sportsnet piece last night, they were trying to say that Montreal’s success proved that size didn’t matter. It was a bizarre piece… they equated being small with being skilled, which almost inferred that they were somehow more skilled than Washington or Pittsburgh. Also, lack of stature doesn’t account for the size of one’s heart.
So there you go: Washington was too tall and clumsy, Montreal has bigger hearts.
In all seriousness, the only thing I was really curious about is whether conceding the blue line is a better strategy than being more aggressive in the offensive/neutral zone (you know, having a forecheck), particularily for teams that lack skill. These were obviously extreme results, but I was curious if that strategy had any merit whatsoever (gives an inferior team a slightly greater chance) or if it’s just playing with fire.
Hockey blogging can't get any flatter.
In fairness to Habs fans, when the Oilers upset the Wings in the ‘06 playoffs, despite being territorially dominated … well over 99% of Oiler fans claimed that it was because the Oilers were keeping the shots to the outside, limiting the shot quality, frustrating the Wings with their ingenious system, etc. All of which were a complete load of shit. And people were spewing this stuff RIGHT AFTER the game. Jebus, we’d just watched it.
The ‘06 Oilers weren’t chopped liver, but the ’06 Wings were a terrific team. They rang a lot of pucks off the iron, had a shitload of great scoring chances, got a tonne of deflected shots at net and a bunch of random rebounds caroming through the low slot. They were the better team by some distance in that series, every one of those guys could pass the puck and receive a pass brilliantly, it seemed. They outplayed, outchanced, and outshot the Oilers by a wide margin. They were unlucky, simple as that.
The other narrative was that the Oilers wanted it more. Stauffer was advocating that Ken Holland should blow up that Wings team and start anew, they just didn’t have the intangibles to win playoff hockey (they had been eliminated in the first round by CGY the previous hockey season as well). I thought Stauffer was off his nut, but every single caller agreed with him.
And though empirical evidence has limited value in arguments with fools … Olivier’s scoring chance numbers, and all of your analysis here, at least it serves to harden the resolve of the rational.
But clearly when that puck bounced off Letang’s skate right to Cammalleri in full stride on a breakaway, that was the skill of the Habs’ small forwards, not luck. :)
Hockey blogging can't get any flatter.
There are at least two sports cliches to counter that!!
1. You got to be in the right place at the right time.
2. You make your own luck.
I’m sure there are more. And if all else fails blame the refs, Bettman and the ice.
And also remember luck is not called luck it is the whims of the Hockey Gods. May they always smile upon your team.
“Make your own luck” has a basis in fact though. Even with the assumption that getting a goal out of a shot were a purely random event, generating more shots will give you better odds of “getting lucky” at least once. Getting a 6 on a die is luck, but the more dice you roll, the more you’ll get “lucky” and get sixes.
So does right place right time. Cliches are cliches because they “work”. Like stats they are generalizations. Like stats they are filled with assumptions. While in Stats you’re supposed to state your assumptions they often are left as implied.
The hardest part about stats is that luck and not fully specified models are all wrapped up into one big rather messy error term.
Take for example the fitting of linear models. Linear models often fit well enough for most applications especially when you are looking at only a small portion of the actual curve. Now when the model doesn’t match the data how much of the error is luck and how much is due to the wrong model (It fits good enough but it is not the function that generates the underlying data). Add to this that we often do not even know all of the predictor variables either because we don’t have them recorded or we choose not to include them for simplicity or simply overlooking them or even ignorance. This cause the model once again not to match the underlying function. This is wrapped into the error. The error is often viewed as luck. It is not all luck.
You can model your error term with mixed effects models (I do like to harp!). And this can help but still won’t fully eliminate the issue.
Another thing… I’m a little perplexed by the focus on the Habs’ goaltending and that people, stats folk and otherwise, look at a series where the Habs had .918 goaltending throughout and call it a series that was stolen by the goalie. Halak is a .919 career goaltender, and over the course of the series the Habs’ goaltending was no better or worse than it’s been all year (.919). So there’s no reason to act as if the Habs’ goaltending prowess was an beyond-the-pale outlier. The Habs have good goaltending, and have had it for years.
Yes, they had fantastically high save percentages on the power play. They also were fantastically low percentages against shorthanded shots.
There might well be a story about how the game-by-game variances in performance by a goaltender matter here (and how the Habs concentrating all their "bad goaltending’ games in games 2-3-4 might have helped them win the series). Games 5-6-7 stick in the mind, but over the series It certainly was not the .930+ goaltending that kept the Avalanche afloat.
I think that if you want to demonstrate “luck” you’ll want to look at the other end of the ice, where the Habs shot 9.8% which is higher than what they had all year (8.6%). But the story before the series was that Washington had somewhat weak goaltending… .908 over the course of the year, certainly not up to Montreal’s standard. If there’s a discrepancy, it’s there… and even then, on such a small sample, the difference between the observed save percentage (.902) and the Caps’ year-long average (.908) is barely more than a single goal.
The reality was that over the course of series both goaltending tandems really performed according to their season average, give or take about one goal against. Montreal parlayed a -1 series-long goal differential (excluding empty-netters) into a 4-3 record and while that’s certainly fortunate for them, that’s certainly no awesome run of luck either (their tendancy to sit and defend leads, whereas the Caps did not, may have been a factor here).
Games 5-6-7 stick in the mind because they were dramatic. But over the series, Montreal got performances from their goalies that matched their over-year average. Goaltending stealing the series might just be the wrong narrative, too…
It’s the story of the season in a sense, when Halak was off during the year, he was bad (.886 SV% in losses) and in wins he had a .948.
It obviously was an outlier/small sample of games, but a .970 is just so eye-popping that it causes this kind of debate. Especially with the quantity of shots, chances and high quality shots faced.
It in essence was “lucky”, because it’s not very often that a goalie plays like Halak did. However it’s not lucky to the degree of a divine blessing or anything.
I wonder, if you broke down other goaltenders’ save percentages by wins versus losses, whether you’d see a similar distribution?
facepalms at self again Gyaaaah. Nevermind, I’m just embarassing myself.
Montreal’s goaltending has been .918 to this point in the playoffs, which includes the two Penguins games. I need to be looking at the Caps’ SA and GA stats to isolate that series, leading to a .931 save percentage (20/290), so I guess it was the same .930 goaltending that’s kept the Avalanche afloat. That’ll teach me to get my data from newspaper articles.
Mea culpa, mea maxima culpa. Still, it’s not a completely beyond the pale result for Montreal’s goaltending to achieve over seven games; however it’s a difference of 3.5 goals over what they’d get from season-average goaltending — half a goal per game, which is a lot.
So yeah, Montreal’s goaltending was a big factor.
Mathman:
Regarding luck, I think you are confusing the methodologies of sabermetricians (and a big chunk of hockey analysts) with guys like JLikens, Sunny Mehta and occasionally Tyler Dellow.
The latter are determining the luck (chance variation) distribution first, at the full exclusion of ability, or any non-luck factors. It’s a random effects model that determines what we would expect to see through the effects of chance variation alone, and compares it too the actual distribution of results.
ANOVA tests, like the one on EVsave% below, have built-in models. In that particular case the “luck” associated with each player’s results is equal for each player, regardless of whether they faced 500 shots against or 3000. That’s obvious wrong. Also assumptions are made about the distribution of talent, the nature of censorship and survival bias, and effects of rule changes are neutralized. And even the final conclusion about % luck is entirely sample size dependent, and a bizarre way of looking at things. I struggle to see the value in it, even it it were correct.
Sunny’s treatment of the same issue in the comments of a previous thread is solid, and the criticisms of his methodology in that thread are baseless IMO. There are survival and censorship issues left unaddressed, but it’s terrific stuff. The next logical step is to built an ability distribution. If your the joint probability distribution of ability and luck results in the observed spread of results, repeatability and predictive capacity … you’re done. That’s some ungodly math, because in that particular case the conjugate prior (beta form) doesn’t work. (This is rarely the case in MLB by the way, which is why the shorthand math of guys like Robert Woods works so well). Presumably that’s why Sunny doesn’t take it any further.
Thank you Vic for saying much better than I could have what I wanted to say. We don’t attribute everything to luck, nor everything “unexplained” to luck. The correct methodology is to calculate how much luck should be observed, based on your model. Then you try and extract modeling factors, and you’re inevitably left with remaining unexplained variance, which may be either “talent” or factors missing from your model.
ANOVA tests applied blindly are close to useless. Sunny’s methodology was 100% valid, however by including small samples he somewhat set up the conclusion he wanted, i.e. it’s all luck. As for the ability distribution, the problem here is that, for example for goaltenders, ability is highly correlated with playing time, such that there is selection bias.
Nevertheless we’re making progress.
An ANOVA model and a random effect model are all based on the same set of models. They have sightly different assumptions since you are modeling the G vs Z matrix but the are very very similar. Can you point me to the Sunny’s stuff so I can better understand what is being talked about.
My point isn’t that ANOVA isn’t valid. My point is that people use it without understanding what they’re doing. For example, taking all goaltenders in a single season and throwing their save percentages at ANOVA is not valid mathematically: save percentages over a different number of shots are NOT distributed normally, therefore the model breaks down.
The best discussion I remember with Sunny was on the messages of the NJ/PHI previews he wrote. I don’t know if there’s more.
ANOVA is not they way I would go about it. It might be my first cut because once I have the data it takes me one line of code but it wouldn’t take much more to fit a generalized linear model or even a GAM. The assumption is not that the data is normally distributed but that the error is. That being said, a standard ANOVA does implicitly imply an assumption about how the error is generated and it does not perform well with percentages especially when they are not near 50%. Mind you random effects models make the same assumptions at their most basic.
I’ll check out Sunny’s comments so I understand what he is doing better. From what you and Vic are describing it sounds like a better option. Thanks for the pointer.
Your referring to his Monte Carlo study? I might try to run a permutation study as well. His Monte Carlo methods seems more than reasonable but I think a permutation test might provide a better test of that particular question. I suspect it will provide the same result but I’m the sort of guy that like to try things for himself. My problem is always raw data. So don’t hold your breath…besides my boss is expecting an analysis before the end of the week.
Boy, the guy doing the ANOVA must be a complete idiot!
The model assumes the errors are normal. From a practical, computational standpoint, the errors just have to be “normal enough”. With at most 12 observations for each goalie, it’s going to be hard to show they aren’t normal. I’ll look at it and post it.
I can run a Monte Carlo on the data, linking each goalie to a save percentage randomly. I think we can all predict the results of that.
by DoctorMyBrainHurts on May 4, 2010 8:34 PM EDT up reply actions
DMBH
I wasn’t implying that you were a fool, far from it. I just don’t think that it is he right horse for the course. Though my knowledge of frequentist statistics is not great (I have no idea what David Mogen’s permutation test will entail). Having said that, I would have probably reacted the same as you.
Another important aspect of Sunny’s study was the use of road data only. That’s very wise imo. Using road data with the score close would be preferable (as Gabe showed recently, within one goal over the first two periods, or tied in the third/fourth).
And linking each goalie to a save percentage randomly wouldn’t make sense, nobody is suggesting that. Sunny is using Bernoulli trials, so he’s assuming binomial distribution of luck. That’s not likely perfect, but it’s a reasonable assumption based on the relationship of scoring chances to the shots metrics at evens, esp with the score close, and wrt streakiness patterns (I suspect that the latter is the greater influence in his case, since Jim Albert seems to be his mentor/influence for this stuff).
So his initial prior assumption is that all goalies have the same natural ability to stop pucks in the same circumstances. i.e That the distribution of ability in the population, plotted out, is a vertical line. Of course the ability (or non-luck) distribution is the Holy Grail for anyone interested specifically in predictive value in general, and wagering in particular.
The problem is that when he ran that trial he came up with a shitload of simulated seasons that yielded results that were spread out WIDER than the actual, observed season. And the sample is small, it’s a big problem, there just aren’t that many NHL goalies to study. And because his model accounts for frame size there is no inherent problem in including guys who didn’t play much, but we don’t necessarily give a toss about 3rd string goalies.
So really, it becomes a bit of a nonsense to try and build an ability distribution from one year’s worth of data. Other than noting that “everyone is the same” is a surprisingly close guess.
Another smaller problem is that we don’t know the overall average EVsave% ability of the goalies in question, only the observed.
I think that using the same methodology over several other seasons would be sensible. Because gauging by Sunny’s results any one season may indicate, by chance alone, that the ability distribution is made of antimatter, which is of course a nonsense. And though it’s a slightly different population from year to year, it’s largely the same people making goalie decisions in the league.
The five seasons since the lockout should be enough to get a good idea.
Or so I think.
I think in general, if you and Tom think something is a bad idea, it’s a bad idea. However, I don’t think my approach is baseless, and I’m trying to build a more complete model a step at a time, hence the title.
I had missed the last half of Sunny’s thread until earlier today. I think Monte Carlo is a great approach. It makes perfect sense (and amuses me a little) that a poker player would turn to Monte Carlo. And his null hypothesis, that all goalies are the same, is my null hypothesis too. I just expect to prove it wrong. He expects to uphold it.
My suggestion of linking goalies randomly is essentially what mogen_david is suggesting (I think)(it’s a form of Monte Carlo). If save percentage really isn’t linked to individual, we’ll get the same result. We won’t. The demonstrated result has a probablitiy of 1 in 10,000 of coming from chance. As far as the GLM versus ANOVA, GLM reduces to ANOVA. Rerunning the data with lm or glm instead of aov gives the same result.
by DoctorMyBrainHurts on May 4, 2010 10:23 PM EDT up reply actions
Permutation tests take the original data set and reorder it. Over and over again.
So under the null that all goalie are the same if I take the entire population of goalies and randomly assign the observed save percentages for each observation (game, season, you can even go down as far a shot just takes more cpu time and doesn’t matter that much) to a goalie the same number of times as I have observations. You then repeat a large number of times. If an individual goalies are the same and difference are due solely to random chance regardless of the underlying function the result should be the same as the observed data.
I could try to bootstrap as well but it would give the same result in the end.
Bootstrapping is exactly what I need to do. I realized this last night. This will show the link between individual and performance. Which is exactly where the coin flip model breaks down.
Imagine a “season” of 30 guys flipping fair coins two thousand times each. There’s about a 5% chance that one of the guys will have 1065 or more heads. But do a second season. The probability that you again have someone at 1065 or better is now 2.5 in 10,000. And the probability that the same guy is number one is one in 30. The probability that the top 10 the first time are the top ten again is tiny. Tiny times 0.00025 is essentially zero. If that’s not close enough to zero, do a third season.
Finally, I think what Sunny showed is that the distribution of save percentages in a single season is binomially distributed. It may not be discrete of me to say this (pun intended), but that suggests to me that the data is roughly normal (probably another pun there).
by DoctorMyBrainHurts on May 5, 2010 10:04 AM EDT up reply actions
Saves have binomial distribution by their nature. It is how they are generated. The problem isn’t their discrete nature since we have lots of events. The problem is that percentages are bounded at 0 and 100. This isn’t a problem if the mean is near 50. However when you get near the margins than things get funky. The logit link in a GLM can be thought of as transformation (it isn’t quite that) take the odds of success and takes the natural log→ (ln(mu/1-mu)). This essentially makes the percentages normally distributed. You would want to check for over or under dispersion. If this is the case, in mixed model you can model that into the error term (G matrix). There is a R package to deal with Zero Inflated Poisson distributions (saves and shots in a game are generated as a Poisson although) and I believe it includes the ability to deal with dispersion problems in binomial models.
As for Sunny’s analysis what it is showing is that the distribution of save percentages within the league is not different from what would be generated if they had been produced from a single goalie with the average save percentage. Thus the null can’t be rejected.
Are you Bayesian then? I’ve never had a chance to really get my head around Bayesian stats other than wonder about those non-informative priors.
No I’m Welsh. And I have no priors. (smiley)
My background was Biostatistics. A little Bayes. Definitely NOT an area of expertise. More small scale clinical trials.
by DoctorMyBrainHurts on May 5, 2010 10:07 AM EDT up reply actions
I’m an ecologist/data analyst I turn trees, grass, fire, cows and water into numbers and look for pattern. Currently getting my head around Structural Equation Modeling (it would be fun to apply to total defence completely useless in predictive power, IMO). I do a lot of experimental design with what ever technique best fits the data, GAM, GLM (mixed usually due to the experimental design and clumpy nature of nature), Survival analysis, computer intensive methods, spatial stats although limited. Ecology always has the issue of low sample size. Used to do a bit of Multidimensional stuff (NMDS, CCA, PCA, etc) but hat has lost favour in ecology.
you should check out the book “Bayesian Inference: With Ecological Applications.” i haven’t read it yet, but it came highly recommended from a good source.
by sunnymehta.com on May 5, 2010 11:54 AM EDT up reply actions
I’ve attended one Bayesian work shop as well as a Meta-analysis workshop that was Baysian i its approach. That text is on my to read list. I have to get up on SEM before I get my head fully around Bayes. I also read in a seminar course about 7 years ago (damn, I’m getting old) the Ecological Detective and that is Bayesian in it’s approach.
Have you ever dealt with Tom Fleming, DMBH?
Many years ago a close friend worked in management for a company that brought Tom in to consult. His opinion was at odds with the in house biostatisticians, and he broke out a baseball metaphor to make his point. Oddly enough, his baseball metaphor would have put him at odds with the vast majority of the sabermetrosphere, but I’m sure he was right. As far as I know, he has never used sports metaphors in his writings, but this was spot on. I know a bit about sports and numbers, that’s probably why buddy brought it with me over a beer.
Any road, he strikes me as one of the best guys to follow in terms of linear patterns wrt censorship and survival bias issues. I’m not at all familiar with martingales, or any of the counting process methods, though. And the shorthand these cats use is otherworldly.
Is this an area that you have some expertise in, DMBH.
On an unrelated note, you need a shorter ineternet pseudonym.
No. I’ll have to look into his work.
Doc works, or Grant.
by DoctorMyBrainHurts on May 5, 2010 2:30 PM EDT up reply actions
Vic,
I think Gabe noted that league average Corsi drops to like 48% when up by one in even the first two periods. That seemed like kind of a lot to me, so I think my preference would be to stick with tied data if we had numerous seasons. What say you?
Also, not sure if you guys saw the end of my exchange with Tom in the NJ/Phi thread, but I reran the sims at the team level using two seasons worth of data:
http://www.behindthenethockey.com/2010/4/13/1419530/new-jersey-devils-vs-philadelphia#35066508
My biggest problem with ignoring backup goaltenders is the huge selection/survival bias effect I think might be in play in the NHL. There is no objective criteria for what constitutes “starter” versus “backup” except playing time. And the problem is that playing time is doled out by teams and is almost always correlated to results. So, essentially teams play a goalie, and if he runs good he gets to keep playing, if he runs bad he sits. And teams usually have a pretty short leash. They rarely play a guy who runs bad for a prolonged period of time, even an “established” goalie. Look at TIm Thomas this season.
My solution is to look at a few EV-road-tied seasons combined, but look at it at the team level. After all, teams have to play SOMEONE in net, right? So even if Boston is flip flopping goalies based on who is running good, if we just look at the total output from Boston’s goaltender position, now we can get a fair sample size over a few seasons. And if we take like 4-6 seasons worth of data and find basically no difference in goaltending results between teams, i don’t know, that seems pretty damning to me.
Vic, maybe you have a better solution? You know way more about this than me. I only found Jim Albert’s work because of the suggestions on your blog.
by sunnymehta.com on May 5, 2010 11:48 AM EDT up reply actions
One thought I had had was to look at team effects through looking at threshold goalies. Your thoughts on that?
by DoctorMyBrainHurts on May 5, 2010 1:08 PM EDT up reply actions
The tidy thing about your model is that whether the true ability of goalies is distributed in normal fashion, beta, vertical line (your initial prior), or a perfect square, or like the gas chromatograph of ether … in every case the posterior will generally test well for a normal distribution (you can check by building those models and running them, then checking the result with on online test for normality, an Anderson-Darling calculator or some such. That’s cool, because it efs up the market.
Also, in every case the sum of variances of the prior and the chance distribution (which varies from sim to sim, at the caprice of the Gods of chance) equals that of the posterior.
Of course variance is only one aspect, the universe will distribute ability precisely as it wants to do so. And humans will decide who gets the ice time.
Really what you are looking at is a variance likelihood distribution.
Also, and this is getting into fine brushstrokes, but I would use road data only, score close, and Fenwick. With Fenwick begetting chances begetting goals.
In any case, you general point, that there is a fraction of the difference between goalies that the NHL GM market and wagering market believes … that’s obviously true. The real effects of either your metric or mine, they are piss poor. You could construct a wager involving future EV road save%, basing it on preivous results, or better yet, trends. And 99% of hockey fans would intuitively sense that you were wrong, many would criticize your thinking and wallop you senseless with a big anecdote hammer, but very few would have the stones to bet against you. That’s the real measure of efficacy in our game, methinks.
For the record, I also was not implying you’re an idiot, and I respect the work you’re doing. In this case, like Vic, I disagree with the method, because errors on save percentages won’t be normal if the goaltenders aren’t facing the same number of shots. If the sample was limited, say goalies facing between 1000 and 2000 shots, it would be good enough.
I apologize if I came off harshly. The overarching point is that doing statistical analysis is hard, the data is unreliable, and the null hypothesis is often attractive.
Thanks Tom. I should have put a smiley or something, as my feelings really aren’t hurt. This is meant to be step one. If the ANOVA showed no effect, I would move on. It looks to me like there is an effect. Next step is pulling out confounding factors and refining the model.
by DoctorMyBrainHurts on May 4, 2010 10:32 PM EDT up reply actions
I’m glad there should have been a smiley since I surely don’t think your an idiot. When you move on to the next step I’d run a GLM with a logit or log link ( always have to look that up) I can never remember which one for percent. I do know it would logit if you used the raw data shots instead of percentages. This will basically cut the discussion off at the pass about normality (well okay it doesn’t but it does assume the correct underlying function).
Not to mention with goalie ability, we’re missing several factors such as Shot Quality (with high accuracy, not just shot location), shot prevention (Brodeur is a Fraud) and others to isolate pure goalie ability, which usually needs larger samples than we have access to.
does there have to be only one narrative? we’ve got an unlikely result and three plausible explanations:
1) luck
2) strategy
3) goaltending
now, personally, if i have to pick one i go with #3, for extremely complex reasons that i am as yet unable to demonstrate but hopefully will be if i continue to obsess about this for another few days. but for everyone else, why does must it be 1 vs. 2 in a battle to the death? i am persuaded both that the canadiens had a strategy that they executed well, but which still needed a fair bit of luck to work. i doubt all the fortune in the world would have helped them if they hadn’t blocked all those shots, or if they’d let ovechkin/semin/backstrom get cozier with halak, but then again, all the capitals would have needed was a slightly better-than-abominable shooting % on the pp to get the few extra goals necessary to change the result. point is, canadiens get less luck, they lose. canadiens play a different strategy, or execute their strategy poorly, they lose.
hard work + serendipity is better than either alone.
There’s no reason there has to be one narrative. If Halak doesn’t play out of his mind, they lose. If the bounces don’t go their way, they lose. Heck, maybe if Green doesn’t take that cross-checking penalty, they lose. As humans, we like to find THE reason, but there never is just one. You’re absolutely correct. The problem is that quantifying “Halak skill” versus “Halak luck” is hard/impossible.
One thing that doesn’t get talked about nearly enough in these situations is that the "luck"may not be in that the puck hit Halak in a way that it wouldn’t in most days, but the luck may be that Halak just had a good day.
There’s a certain amount of variation in the skill a player shows from game to game that can be attributed to pure randomness, the way the puck bounces on the ice, hitting the post, but I think people miss that over the course of a season a player has days in which their maximum skill shines through and days in which it doesn’t.
Halak’s skill in Game 7 was clearly at or near his maximum, and the Habs got lucky that Halak felt good that day. No harm in that.
The New Improved Avalanche. Now with Real Coaches!
Jibblescribbits: C'mon over and waste some time
by Jibblescribbits on May 4, 2010 1:46 PM EDT up reply actions
The more likely and more reasonable explanation is that the shooters just didn’t make their shots.
I mean just by looking at what scoring chances represent (shots from places where the goalie doesn’t have time to react, he can only guess), well to give a goalie too much credit for guessing correctly… it’s like giving too much credit for winning in rock-paper-scissors. Obviously some goalies can guess better, read opposing shooters better, just like some rock-paper-scissors guys can (probably) spot trends in their oppositon.
But it’s a big fat load overflowing heap of luck covering that up.
by R O on May 4, 2010 2:45 PM EDT up reply actions
The more likely and more reasonable explanation is that the shooters just didn’t make their shots.
Why?
A human body is a treasure trove of randomness, starting with the most random instrument in the solar system: the human brain. Muscle twitch, reaction times, or just the ability to read a play vary significantly from day to day, hell it’s a giant load of randomness. I think saying the random factor is the shot, and not how the human reacts to it is putting the cart before the horse a little bit.
This goes back to an argument you and I had started after a particularly heated game. Why did you call Anderson’s great performance one night “lucky” and Iginla’s skill. Iginla clearly has wide variation in his performance from night to night, as does Anderson. The average level is high, but there are nights Iginla sucks sometimes (not enough though). I don’t see how that can’t be random, but a goalie’s performance is.
The New Improved Avalanche. Now with Real Coaches!
Jibblescribbits: C'mon over and waste some time
by Jibblescribbits on May 4, 2010 3:02 PM EDT up reply actions
I say it’s the shooters because it’s the nature of hockey, Jibble. Randomness lies a ton with the shooter, and sometimes it’s not even something he can control. Puck’s on edge, rut on the ice, puck just rolls off the wrong way, he’s juuuuust not ready for the shot because it took an unexpected path to him.
And the goalie, well I suppose he could also have his moments when he just doesn’t have it, maybe in the guesswork part of scoring chances, especially glove high (although I think Gabe showed once that SV% on the high parts of the net is really low, like maybe sub-0.800? Which highly suggests that goalies aren’t that great guesswork, and the shooters have the control).
Anyways, the goalie’s job is to just block as much nets as possible. I mean these guys aren’t superhuman the way we make them out to be, they’re just big moving cushions. And they’re subject to the vagaries of chance, they literally have no control where the puck is going and even the most in-zone goalie is just rolling dice in his head when the puck’s coming off the shooter’s stick from the hashmarks.
As for skaters, well the chance-converting aspect of their game is similarly subject to huge variations in chance, that often have little to do with how well they’re actually executing their game.
But other stuff like outchancing and territorial advantage, well these are pretty tightly linked to concepts like being hard on the puck and keeping the play alive. Which is just something that hockey skaters exert a lot of control over, it’s the nature of the game.
by R O on May 4, 2010 7:48 PM EDT up reply actions
it’s interesting to consider that, for whatever reason, the canadiens’ experience of the season predisposes them to believe halak can win in games like these (i.e. where they sit back and protect a narrow lead by allowing a whole mess of shots). over the regular season, he was nearly undefeated when facing +40 shots (one overtime loss), and his S% gets better as his SA goes up. whether that’s just randomness due to small sample size or the nature of halak as a goalie, i couldn’t say, but martin does have reason to believe he’ll perform in this apparently-awful scenario.
Puck’s on edge, rut on the ice, puck just rolls off the wrong way, he’s juuuuust not ready for the shot because it took an unexpected path to him.
I just have a hard time believing that ruts in the ice, taping the stick improperly, or whatever have nearly as much impact as the shooter’s mental state, and his physical state (muscles sore, etc). I mean you could get rid of all those exterior effects (Freshly Zamboni the ice, stationary target, use the exact same rink, same stick, same puck etc) and just put a shooter against targets, and you’ll still get a wide variation on the rate at which he hits the targets.
But other stuff like outchancing and territorial advantage, well these are pretty tightly linked to concepts like being hard on the puck and keeping the play alive.
That’s just false. That would mean that hard workers would be top liners and the best scorers. But Matt Hendricks, who works as hard as anyone, isn’t a 1st liner for a reason, and that’s skill. And skill is such a precision tool that there’s no way it’s going to vary from day to day (or hour to hour).
And that’s not even allowing the fact that not only is it hard work, but it’s smart work. Positioning is a major role in out chancing and territorial advantage, and the mental controls that govern a person’s ability to make the snap judgement on where to position himself is also subject to randomness. Years of training helps make people make smarter decisions, but players still make split second decisions and will make the right or wrong one, sometimes based solely on randomness in their brain. Kevin Bieksa is a fine defenseman, but he still made a mistake on CHI’s GWG last night. Put in that exact same position he may not make the same decision.
And that gets to goaltending, which is ~60% positioning. Some nights goaltenders will have better positioning than others.
The New Improved Avalanche. Now with Real Coaches!
Jibblescribbits: C'mon over and waste some time
by Jibblescribbits on May 4, 2010 9:23 PM EDT up reply actions
I just have a hard time believing that ruts in the ice, taping the stick improperly, or whatever have nearly as much impact as the shooter’s mental state, and his physical state (muscles sore, etc). I mean you could get rid of all those exterior effects (Freshly Zamboni the ice, stationary target, use the exact same rink, same stick, same puck etc) and just put a shooter against targets, and you’ll still get a wide variation on the rate at which he hits the targets.
What can I say, the game is fast. And scoring chances… they appear and disappear like that. As much as we’d like to think that shooters have all day to pick corners, they just don’t.
That’s just false. That would mean that hard workers would be top liners and the best scorers. But Matt Hendricks, who works as hard as anyone, isn’t a 1st liner for a reason, and that’s skill. And skill is such a precision tool that there’s no way it’s going to vary from day to day (or hour to hour).
I wish I knew what you meant by this. Being hard on the puck is a skill. Some guys get by working hard at it but some piss-offs will just never get it (see: Jokinen, Olli; Malkin, Evgeni; Kovalchuk, Ilya; need I go on?) and some poor fuckers seemingly forever live by the adage “if at first you don’t succeed…” (see: Kotalik, Ales).
And of course keeping the play alive in all three zones, esp. against tough competition… that’s a supreme skill.
So I’m not sure what you’re getting at. These two concepts (amongst others) are highly skill driven and also highly repeatable
And yet Jokinen and Kovalchuk still are 1st/second liners, despite not giving their best effort every shift. And there’s 4th liners who give their best effort every single shift. Saying that all it takes is being hard on the puck and keeping the play alive is the main skill in puck possession seems to trivialize the actual skill of recognizing passes, and having the skill to make the passes (and receive the passes).
I could have saved a lot of words had I just said that it’s far more likely that instead of some outside version of bad luck, (bad hop, etc) it’s far more likely that a shooter misses because he simply mis-timed the pass/shot. That has a lot more to do with internal randomness of the body than any external randomness.
The New Improved Avalanche. Now with Real Coaches!
Jibblescribbits: C'mon over and waste some time
by Jibblescribbits on May 5, 2010 9:23 AM EDT up reply actions
Jokinen is anything but a “first liner” by any . Coaches are afraid to use him against good players because he sucks against good players.
Kovalchuk’s finish is what has kept him from not getting terribly outscored. He has talent but he does not play the game against the best very well at all.
by R O on May 5, 2010 1:11 PM EDT up reply actions
Yeah, taking and making a pass, that’s a skill that players exert lots of control over too.
by R O on May 5, 2010 1:13 PM EDT up reply actions
it is, but since passes have to be so precise, there’s a lot of randomness involved from pass to pass, both externally and internally.
The New Improved Avalanche. Now with Real Coaches!
Jibblescribbits: C'mon over and waste some time
by Jibblescribbits on May 5, 2010 3:55 PM EDT up reply actions
Passes are rarely precise. I mean missing a pass is easy to do, or having it not quite in your wheelhouse when you’re making a shot.
Still, recovering from a missed pass is easier if you’re already a good player, you can chase the puck and win the subsequent battle, or pressure the opposing puck carrier if they recover it instead.
by R O on May 5, 2010 3:58 PM EDT up reply actions
I’m not arguing against that at all. All I’m saying is that the reason passes are missed often is because the randomness involved inside a person at any one time. Some days guys make good reads and good passes, other days they make poor reads and poor passes. It’s rather random.
So a skaters good day (relative to their own mean) would be just as much a function of randomness as a goaltenders
The New Improved Avalanche. Now with Real Coaches!
Jibblescribbits: C'mon over and waste some time
by Jibblescribbits on May 5, 2010 4:14 PM EDT up reply actions
And that gets to goaltending, which is ~60% positioning. Some nights goaltenders will have better positioning than others.
Well the really bad goalies will just lose their nets, but most goalies will have fine positioning unless there are extenuating circumstances.
I mean when was the last time you saw a competent starter not go square against a shot he saw all the way? It’ll be too soon yet, in my opinion.
Problem is that extenuating circumstances happen all the time, nothing the goalies can do. Sometimes they get screened or the puck deflects the wrong way on them and they find themselves not square to the eventual shooter.
And really, even though the best goalies are better at reading these things, you’re going to have to clear your calendar and sit with me all day before you can convince me that luck, outside of the goaltender’s sphere of influence, doesn’t have a bigger impact on that.
I’m not saying that it isn’t luck, and certainly external luck plays a big factor in that, but there’s a big deal of internal luck involved as well.
Why does a puck squeeze through a goalie’s pads on night A, but he saves it on night B. A large part could be due to differences in the shot itself, but it’s just as feasible that the goalie was “feeling better” that day (for lack of a better term) and his muscle memory, or his recognition or whatever allowed him to cover his five hole, or squeeze the puck. That’s a certain element of luck too.
The New Improved Avalanche. Now with Real Coaches!
Jibblescribbits: C'mon over and waste some time
by Jibblescribbits on May 5, 2010 9:25 AM EDT up reply actions
Caps fan's perspective
The inherent flaw in these supposedly objective stats is they rely on a subjective measure of “scoring chances.” I’m sorry, but two shots from the same spot are not 100% equal in terms of “quality.” On the Caps’ stagnant power play, for example, it was super easy for Halak to make saves of shots from supposed high scoring areas because they came from stationary shooters. Where is the statistical measure for “shots taken on a goalie who has to make lateral movements” versus “shots taken on a goalie who is square to the shooter for 4 seconds before he faces the shot”?
That was the problem with the Caps this series. It has nothing to do with luck, I’m sorry. They took a lot of shots from supposed “high scoring” areas, but they were all easy for Halak to save. So he saved them.
Dismissing the Canadiens’ strategy and gameplan which forced these meek scoring chances is hogwash.
The keyboard is mightier.
I can’t win, can I?
I use objective stats and people tell me that I’m missing subjective data. Olivier watches every game in slo-mo and scores the games subjectively, and you say he should divide the chances into a whole bunch of different subjective categories.
Sorry, that approach doesn’t work – it’s what they do inside the league, but they can’t define what’s a really good chance and what’s merely good, so they’ve thrown their hands up. Olivier’s method is as good as if not better than the one you suggest.
“They took a lot of shots from supposed "high scoring" areas, but they were all easy for Halak to save. So he saved them.”
This is great. This is exactly what I was looking for when I asked people for b.s. after-the-fact explanations.
“Dismissing the Canadiens’ strategy and gameplan which forced these meek scoring chances”
Letting the other team come in to your end and load up on shots is not a good strategy. Play this series 10 more times, and the Caps win 7.
So what you’re saying, was that whenever the Caps had a glorious chance from the slot, that Halak was “able” to react because the Caps were throwing a 20 kph muffin at him?
Or that before they took their shot, that they pointed out which corner they were going to pick, so that Halak could guess correctly?
I mean unless the Caps were intentionally trying to lose the game, your assertion is baseless. And it’s laughable that you think the Habs actually had anything to do with this, unless you think they bribed the Caps to throw the game.
by R O on May 5, 2010 1:15 PM EDT up reply actions
I don’t see why it’s hard to understand the concept that two shots from the same spot are not always equal. They weren’t throwing 20 kph muffins, and it’s true they were unlucky. But why isn’t there room for the possibility that the majority of these shots were taken from stagnant positions (every power play), off of the back foot (semin), right into Halak’s stomach, etc. that made his job relatively simple? If you watched Caps games throughout the season, you would have never seen them take crappy shots on such a consistent basis as they did against the Habs.
I’m not saying there is a better way to accurately measure. I think it’s a great way to measure. But you have to acknowledge that as this current measure stands, it treats every shot from the same spot as equally likely to go in. And that is objectively not the case.
The keyboard is mightier.
For fuck’s sake. If you can’t even see the inherent significance of a scoring chance to the game of hockey then there’s really no point.
by R O on May 5, 2010 3:38 PM EDT up reply actions
If I understand the stat correctly, it treats all shots from the same spot as the same quality of scoring chance. If I’m wrong then the premise of my argument is wrong.
If I am right, though, and you can’t see that a one-timer from spot X after a pass across the zone is better than Semin standing in spot X with the puck for 5 seconds and then firing off a shot, then there’s really no point.
The keyboard is mightier.
You don’t understand anything correctly.
We’re recording shots from places on the ice where the goalie exerts the least influence over the puck going in.
I mean a shot from the puck, that the goalie sees all the way… well he’ll whiff now and then but he sees it and reacts from it.
A shot from the hashmarks, even if it’s not from Kovalchuk… well the goalie can have all fucking day to set up for it but he’s not going to stop that puck, ever, if the shooter makes a good shot. The goalie gets big and tries to guess and that’s that.
So really, nobody cares whether or not we’re talking about one-timers or 5 seconds of puck control or whatever, we’re just looking at places where the shooter has the puck and view of large parts of the net and a real and significant chance to score if he makes a good shot.
Aside: the fact that you think players ever have five seconds with the puck in the same spot before firing off a scoring chance… it betrays your complete, total and utter misunderstanding and lack of knowledge of the game of hockey
by R O on May 5, 2010 3:47 PM EDT up reply actions
So really, nobody cares whether or not we’re talking about one-timers or 5 seconds of puck control or whatever, we’re just looking at places where the shooter has the puck and view of large parts of the net and a real and significant chance to score if he makes a good shot.
An addendum: we look at these things because the ability to access those scoring areas and prevent your opponents from accessing them… they depend on a large number of real and sustainable hockey skills. Quibbling about whether the shot is a one-timer or a wind-up with all day to make the shot… it’s missing the forest for the trees.
by R O on May 5, 2010 3:49 PM EDT up reply actions
he fact that you think players ever have five seconds with the puck
I was exaggerating to make the point, but if you watched the Caps power play you might say I’m hardly exaggerating.
We’re recording shots from places on the ice where the goalie exerts the least influence over the puck going in.
That’s commendable, and it’s definitely better than looking at total shots.
So really, nobody cares whether or not we’re talking about one-timers or 5 seconds of puck control or whatever, we’re just looking at places where the shooter has the puck and view of large parts of the net and a real and significant chance to score if he makes a good shot.
This is where we diverge. I understand why methodology forces you to draw a line somewhere. But even before I saw this post, I felt that the Caps were systematically having trouble getting off quick shots, especially shots where he had to move side to side. I can’t even think of a save Halak had to make where he didn’t have plenty of time to square up.
The ability to access scoring areas is great. But if you’ve ever played, and I would like to assume you have, you would understand that yes, if you make a perfect shot from point X, the goalie can’t save it, but you significantly decrease your chances of having an imperfect shot go in if he is square.
When people say that the Caps’ power play was stagnant, it’s not an abstract term, it’s a concrete problem.
The keyboard is mightier.
Oh good, we are making “time to getting square” the newest bullshit explanation for the loss.
Time is an advantage for both goalie and shooter. Goalies have time to get square, shooters have time to settle the puck down and make the perfect shot.
I mean most chances are pass-or-rebound, on-the-stick, off-the-stick, players always have a sense of where the net it but the game is so fast that they just can’t aim their shots all the time.
So we talk about these insane concepts of “more time with the puck” being somehow better for the goalie and that’s just unbelievably dumb, and stupid, and misguided.
Last thought: you ask a goalie whether he’d like to face a bunch of 5on5 shots or play in the shootout where they have literally all fucking day to get square… then get back to me.
by R O on May 5, 2010 4:03 PM EDT up reply actions
Time is an advantage for both goalie and shooter. Goalies have time to get square, shooters have time to settle the puck down and make the perfect shot.
No, it is not an advantage for both. It is a significantly greater advantage for a goalie. Have you ever actually played sports?
you ask a goalie whether he’d like to face a bunch of 5on5 shots or play in the shootout where they have literally all fucking day to get square… then get back to me.
This analogy is completely unrelated to my argument, and if anything supports it: a goalie that has to move (laterally or in and out) while making a save is at a significant disadvantage than a goalie just sitting there anticipating a shot.
It’s not unbelievably dumb, or stupid, or misguided to suggest that the Caps did not have the quality scoring chances the data suggests. This argument has run its course I guess, but thanks for being a condascending douche all the way through.
The keyboard is mightier.
No, it is not an advantage for both. It is a significantly greater advantage for a goalie. Have you ever actually played sports?
Oh boy! I’d love to see you back this one up. What, do goalies need time to blow up the inflatable bags in their pads to get bigger? Are they psychic, but need a couple of seconds to read the shooter’s mind?
I mean I can get behind the general idea that rebound shots are more dangerous, the Caps had a ton of those though! But the idea that a goalie can just get square-er and square-er as time goes on, and the shooter can’t possibly benefit from being able to settle the puck down and pick his shot… yeah. You go on living in the fantasy-land of hockey thinking.
It’s not unbelievably dumb, or stupid, or misguided to suggest that the Caps did not have the quality scoring chances the data suggests. This argument has run its course I guess, but thanks for being a condascending douche all the way through.
Stupid is as stupid does, and you’re welcome.
by R O on May 5, 2010 4:15 PM EDT up reply actions
Oh boy! I’d love to see you back this one up.
Take some time with your slo-mo replays and I’m sure it will back up my assertion. If stats are all you believe then there’s no point in me saying that I know from experience and playing both hockey and lacrosse, similar sports where a shooter is at a greater advantage over the goalie if he gets a shot off quickly. It’s actually a pretty basic concept.
You don’t need to get so defensive. I’m not saying this data is not valuable. But to suggest it tells the whole story is asinine.
The keyboard is mightier.
No offense (well, that’s a lie), but this…
If stats are all you believe then there’s no point in me saying that I know from experience and playing both hockey and lacrosse, similar sports where a shooter is at a greater advantage over the goalie if he gets a shot off quickly. It’s actually a pretty basic concept.
Nobody cares that you’ve played hockey, it doesn’t matter and it’s irrelevant. You still don’t have a damn clue about how the game works.
by R O on May 5, 2010 4:21 PM EDT up reply actions
And you do because you crunch numbers? Where does your infinite trove of hockey wisdom come from?
I’m not saying I don’t buy that objective stats are great and tell much of the story. Get over yourself.
The keyboard is mightier.
If I may intervene in this conversation… I’m the guy recording MTL (and thus WSH) scoring chances.
To put it simply: yes, your understanding of what I record as a scoring chance is correct. Basically, I try to get, through those numbers, an understanding of how good or bad team and players are at playing hockey. Location is the key for me because:
a) It is easily recognizable (well, with HD slo-mo, that is ;) )
b) This is a terribly fast and tough sport; unbeilivably talented individuals misses all the freaking time. I mean all the time. So success, it seems to me, is largely about getting the most exposure possible to favourable conditions. Ovechkin’s dominance is in part due to his ability to shoot 500+ pucks at the net in a given year. Having watched various “snipers” labor to get to 200 on a given year, I insist: getting in position is a huge part of the whole deal.
c) The results are pretty much inline with what we know overall: Ovechkin is a beast, Marc-André Bergeron is a turd (at even strength that is), that Shaone Morrisson dude didn’t look very good though, dunno about him.
Now, I don’t claim I’m “objective”. But I think I’m pretty fair. What can I say; Semin spent the last 3 games giving me (a habs fan) heart attacks; dude couldn’t hit the broad side of a barn from 15 feets, but he had an immense quantity of very, very nice scoring chances. Trust me, a guy like him has 3 seconds to set up in the slot, you can be sure the opposing coach will yell at the guys who let that happen.
I made this chart to show how much Semin’s line Scoring Chances output shot up in the last three games of the series.
you're making some wild assumptions
All these numbers fail to take into account the position of the defense. A shot taken when the D is in perfect position is not as likely to score as a shot taken after Mike Green takes himself out of the play with poor decision-making. Montreal did a better job of defending. That is what made the difference in the series. So, maybe now McPhee will go out and get a decent 1-2 defenseman? Probably not.
A shot taken when the D is in perfect position is not as likely to score as a shot taken after Mike Green takes himself out of the play with poor decision-making.
What in the hell? The new obsession is perfect positioning?
We’re playing hockey here, not chess.
by R O on May 5, 2010 1:50 PM EDT up reply actions
I tought Green wasn’t that bad, except he stayd back too much, it helped the habs a bunch. He’s getting stripped for that last Moore goal, but the real faulty player here was the other D who simply didn’t bother backchecking and left Green to fend of 2 players all by himself.
Greene is fine and the team in general is fine. The habs shut down the bottom of the lineup and Semin went cold. It sucks, but it happens.
Yeah, I mean if that’s a one on one play then Green played it perfectly, and the fact that it was a bit of a vertical two on one… what’s he gonna do, block the pass to the trailer, and leave the puck carrier with a clear break to the net?
In any case I just don’t know why the new excuses keep popping up. The first was shots to the outside, then lack of breakaways, then scoring chances from the slot are just not that good anymore, then it’s perfect positioning…
It’s not a Habs fan thing either, even the Caps fans and other teams’ fans… it’s just a stupid people thing, I think.
by R O on May 5, 2010 2:12 PM EDT up reply actions
Moving the goalposts fallacy…. kind of literal in this case.
by rsm on May 5, 2010 9:56 PM EDT up reply actions
I think real test for this stuff would be to put up a poll with this hypothetical question:
The NHL has decided for whatever reason that the first round series between the Capitals and Canadiens has been invalidated and must be replayed. Who do you put your money on to win the re-match?
I know who I’d bet on.
It’s funny, the most flagrant idiots who will ever disregard the impact of luck and the importance of territorial advantage and scoring chances and context of icetime… their most common attack is that we’re too “into the numbers”.
When really, what is a win but the number of goals on one side being more than the number of goals on the other side? And these disciples of The Win, well they are also students of the Point and Goal Totals, and embrace shot quality like it came to Newton via a falling apple.
Comments For This Post Are Closed
Secondary Sidebar
User Tools
FanPosts
Community blog posts and discussion.
Certain photos copyright © 2012 by Associated Press or Getty Images. Any commercial use or distribution without the express written consent of Associated Press and Getty Images is strictly prohibited.
