Advanced Stats

[tigervixxxen]

On top of the other 5000 reasons I want Montreal to beat Boston would be hilarious to see the stat crowd try to explain how a possession team worse than ours beat one of the possession kings.

PDO is the worst stat. At least Corsi is founded on some semblance of logic but why someone decided that shooting % plus save % should equal 1000 is beyond me. They are stats completely independent of each other.

If people want to say the Avs were fortunate and overachieved, fine. I don't think anyone is going to disagree with that. So to trumpet MATH to prove that point seems a bit silly. So if they finish with anything less than 112 points next season, the "I told you so" crowd is going to come out. That's not analysis. What would be useful is a discussion on how much was maturation of young stars and how much was statistical anomaly and how much each aspect are projected to contribute next year. That's the part where the whole argument becomes impossible. Even the stat crowd can't deny the Avs have one of the best groups of young talent.

 

[BB1133]

All these stats and math that prove why the Avs lost are completely pointless. People pick and choose what they want to use them for. They try to use them in the regular season for the Avs, and they don't work. Then you try to use them in the playoffs and all of a sudden there's the proof.

You don't lose in the playoffs because of a stat, you lose because you made mistakes, and also didn't react the right way in "key" moments. There's no stat for those things, and many of them are mental, not what you're doing on the ice that leads to it.

The mistakes being the defense failing to get the puck out, and losing all the battles down low because the Wild studied them during a long series and knew their weaknesses. The key moments being their special teams failing, and being too passive after scoring goals, and letting the Wild respond with goals of their own rather than having the killer instinct to keep pounding on them.

There's a multitude of things that lead to a win or a loss in hockey, and trying to boil that down to a stat is ridiculous. They're fun to look at for some people, but they have very little importance IMO unless you're using them with the sample of a whole career.

What most of us said last summer would be the Avs problem in the playoffs, was the Avs problem in the playoffs. The defense. Injuries didn't help, and Varly wasn't able to cover for the mistakes in every single game.

You don't need inaccurate advanced stats to pick and choose from, when all you have to do is look at the team, and how it's playing to see it's weaknesses.

If you think analytics guys chalk everything up to "a stat", you really haven't taken any time to look at what they're saying.

 

[Bender]

Oh, and please quit with this "moral victory" garbage. Call the advanced stats crowd smug if you must (certainly some are) but none have ever used that term, nor have they ever implied it.

Fancy stats can't predict games, but they can somewhat accurately predict trends. They were spot-on regarding the Leafs, for instance. Chances are Toronto would've missed the postseason last year as well had it been 82 games.

LOL I didn't needs any advanced stats to tell me that...it was WAY too obvious they relied on Bernier and at times Reimer to keep them in games all season long.

Vary was good for us this year but he didn't steal games nightly. It wasn't like in 2009 when Craig Anderson was just about the only reason we won games. The team played well in front of Varly and kept it up for 82 games. Sure, he was outstanding when he needed to be but very rarely the only reason we were winning games. That's the biggest difference.

People that NEED 'advanced stats' to make these kinds of determination are either:

a) too stupid when watching games and can't make the realization on their own
b) too lazy to stay up late to watch the games where they can make that determination

Just for fun, I'd like to be an NHL coach and I'd tell my roster this:

- Whenever you get the puck, I want you to get a shot on net. I don't care from what angle and how soft it is. If you can manage to get several rebounds, that'd be even better. I don't care if you score or how dangerous the shot, I just want to make sure we out shoot the opposition. I don't care if we lose the game, I just want to outshoot everyone.

The dumb thing about advanced stats is that it goes off of the premise that each team is as talented as the next and made up of all the same type of players. Which is horribly inaccurate. If I have a team full of grinders that cycles the puck down-low to manufacture their scoring chances that's apparently better than if I have a team full of game breakers that can score off the rush and go from one end of the ice to the other in 6 seconds and score a goal? The cycling is much better 'cuz it took longer??

Gimme a break.

Advanced stats are not completely useless, it can give an indication of what's going on but if you don't dig deeper to try and figure out why a certain stat is what it is, then it IS useless because it doesn't always tell the REAL story.

Put it this way, we're there ANY Avs fans that were nervous at all during the regular season that the 'wheels were going to fall off' as the advanced stats crowd suggested and that we weren't going to make the playoffs? I never doubted it for a mili-second because I watched the games.

 

[Pokecheque]

On top of the other 5000 reasons I want Montreal to beat Boston would be hilarious to see the stat crowd try to explain how a possession team worse than ours beat one of the possession kings.

PDO is the worst stat. At least Corsi is founded on some semblance of logic but why someone decided that shooting % plus save % should equal 1000 is beyond me. They are stats completely independent of each other.

If people want to say the Avs were fortunate and overachieved, fine. I don't think anyone is going to disagree with that. So to trumpet MATH to prove that point seems a bit silly. So if they finish with anything less than 112 points next season, the "I told you so" crowd is going to come out. That's not analysis. What would be useful is a discussion on how much was maturation of young stars and how much was statistical anomaly and how much each aspect are projected to contribute next year. That's the part where the whole argument becomes impossible. Even the stat crowd can't deny the Avs have one of the best groups of young talent.

PDO is simply a way to measure a team's relative "luck," i.e., a hot goalie and team shooting percentage. But by it's very nature it's a very unreliable stat. It doesn't take into account that some teams with very good scoring talent can actually sustain an above-average shooting percentage.

The only thing that's annoying me to hell about the stats crowd (notably Tyler Dellow and Ryan Lambert) are how smug they're being about it and how they keep lumping the Avs organization in with the Leafs, the latter of whom openly snubbed the stats crowd, while the former repeatedly admitted they were having huge problems with gap control and limiting shots on their own net. Does anyone think the Avs are truly looking at their current roster and saying "Yeah, we'll be fine with the exact same lineup next season"?

 

[BB1133]

On top of the other 5000 reasons I want Montreal to beat Boston would be hilarious to see the stat crowd try to explain how a possession team worse than ours beat one of the possession kings.

PDO is the worst stat. At least Corsi is founded on some semblance of logic but why someone decided that shooting % plus save % should equal 1000 is beyond me. They are stats completely independent of each other.

If people want to say the Avs were fortunate and overachieved, fine. I don't think anyone is going to disagree with that. So to trumpet MATH to prove that point seems a bit silly. So if they finish with anything less than 112 points next season, the "I told you so" crowd is going to come out. That's not analysis. What would be useful is a discussion on how much was maturation of young stars and how much was statistical anomaly and how much each aspect are projected to contribute next year. That's the part where the whole argument becomes impossible. Even the stat crowd can't deny the Avs have one of the best groups of young talent.

The Avs were significantly worse than Montreal in possession this year. The three "fraud" teams that would really give them problems are the Leafs, Avs, and Ducks. Next year they project the Avs to finish around 95 points (or worse).

 

[tigervixxxen]

The Avs were significantly worse than Montreal in possession this year. The three "fraud" teams that would really give them problems are the Leafs, Avs, and Ducks. Next year they project the Avs to finish around 95 points (or worse).

That's not true. The Avs were ranked 25th in Corsi, the Habs 26th.

The other thing I'd love to see explained is how a terrible team ended up with 95 and 88 point years among the tank years. The Oilers for example haven't seen 88 points in a long time.

 

[BB1133]

PDO is simply a way to measure a team's relative "luck," i.e., a hot goalie and team shooting percentage. But by it's very nature it's a very unreliable stat. It doesn't take into account that some teams with very good scoring talent can actually sustain an above-average shooting percentage.

The only thing that's annoying me to hell about the stats crowd (notably Tyler Dellow and Ryan Lambert) are how smug they're being about it and how they keep lumping the Avs organization in with the Leafs, the latter of whom openly snubbed the stats crowd, while the former repeatedly admitted they were having huge problems with gap control and limiting shots on their own net. Does anyone think the Avs are truly looking at their current roster and saying "Yeah, we'll be fine with the exact same lineup next season"?

Totally agree about Tyler Dellow. I agree with most of his analysis, but when he compares the Avs to the Leafs I just shake my head.

 

[Bender]

That's not true. The Avs were ranked 25th in Corsi, the Habs 26th.

The other thing I'd love to see explained is how a terrible team ended up with 95 and 88 point years among the tank years. The Oilers for example haven't seen 88 points in a long time.

There is no stat, regular or advanced that can measure the inept coaching impact of the creature called joe sacco.

 

[Foppa2118]

If you think analytics guys chalk everything up to "a stat", you really haven't taken any time to look at what they're saying.

I think they break everything down so much that it renders it pretty irrelevant in terms of trying to draw any conclusion from.

Most of these stats need a huge sample size to be considered accurate and even then they represent such a small factor in terms of whether a team or player is good or not.

Therefore they're much more misleading than they are helpful if you're trying to draw any sort of meaningful conclusion from them.

 

[BB1133]

That's not true. The Avs were ranked 25th in Corsi, the Habs 26th.

The other thing I'd love to see explained is how a terrible team ended up with 95 and 88 point years among the tank years. The Oilers for example haven't seen 88 points in a long time.

They primarily combine Fenwick and PDO. Also look at the season long trend for the Avs. They were consistently plummeting as the year went on (and it continued into the playoffs). I agree that they don't look at Montreal as a contender, but they certainly weren't considered worse than the Avs.

 

[tigervixxxen]

There is no stat, regular or advanced that can measure the inept coaching impact of the creature called joe sacco.

Then how do you explain he wrung 88 and 95 points out of them? Seriously though, the most astonishing to me is how each of the four years he was here the team played something like .375 hockey or worse at the end of the season. Even in decent years. How does a team do that? Four years in a row? That's exceptional talent.

I think the idea of looking at shooting % and save % is useful at the player or team level but to compare teams is where it loses it's usefulness. Like the Avs don't take a lot of shots but were 4th in goals so obviously their shooting % would be high. None of the stars outside of Stastny and maybe O'Reilly had an inflated shooting % though.

 

[BB1133]

I think they break everything down so much that it renders it pretty irrelevant in terms of trying to draw any conclusion from.

Most of these stats need a huge sample size to be considered accurate and even then they represent such a small factor in terms of whether a team or player is good or not.

Therefore they're much more misleading than they are helpful if you're trying to draw any sort of meaningful conclusion from them.

They have about 7 years of data. I'd say that's a pretty large sample size. The only team during that time to make an outlier playoff run were the 2010 Canadiens when Halak morphed into Patrick Roy.

As a predictive model, it works. Most people who rail against it are emotionally charged fans that are taking it personally.

 

[zxcvnm]

I didn't really want to bump this thread but I didn't really want to make a new one. Wish we had a statistic thread or something.

Anyway, found this link interesting. It shows where all the shots came from this season. Interesting to compare it to other teams or last season.

http://legion.stat.cmu.edu:3838/plotsberry-teams/

This data is the beginning of the advanced stats crowd being able to actually start looking a bit beyond what they have. It's absolutely shocking to me that they haven't yet incorporated shot location information into their corsi/fenwick and PDO data.

Here's a simple example. Right now, corsi/fenwick treats all shot attempts identically. For the most part, it's okay, but it does ignore second order effects like shot location. For some teams, those second order effects can be meaningful.

Here's a simple thing they should do, but I'm flabbergasted about why they haven't. It's not clear it will be meaningful, but it should be ruled out first before being ignored. They should weight shot attempts (for corsi/fenwick) and shooting/save percentage (for PDO) based on shot location. Take the percentages in each area and divide by the average. Suddenly you have a weighting scheme. If the league average is 7% but the shot is from a 14% location, then it's worth two average shots and should be counted as such. Using the reweighted data, they could then look at correlation, etc. and see if anything improves.

Will it? I don't know, but it's patently ridiculous that I haven't even seen this attempted yet.




As for how any of this applies to the Avs? Looking at the Minnesota series, game 6 was a revelation compared to games 3 and 4. It's a tiny sample size and there are many other variables, so you have to be careful reading to it too much, but what I took away was how much a single talented player can change the game so dramatically. We were dominated, possession-wise, in games 3 and 4. When Duchene came back, our lines balanced out, and game 6 was much more even. That line suddenly had someone who can create scoring opportunities. Jamie McGinn was moved from a top 6 role where he is arguably overmatched to a 3rd line role where is one of the best. Duchene's return improved the quality of two entire lines.

Similarly, one more top pairing defenseman will allow someone like Hejda to play a more appropriate role as a middle-pairing D. It will push someone like Holden, Benoit, or Guenin into a more appropriate role as well. With a single player, a very poor group of defenseman can become better than average overnight. Acquiring such a player won't be easy, but I do believe that possession statistics can be dramatically improved with a single acquisition. At least, that's what I took away from Duchene's return to the lineup against Minnesota.

 

[Foppa2118]

They have about 7 years of data. I'd say that's a pretty large sample size. The only team during that time to make an outlier playoff run were the 2010 Canadiens when Halak morphed into Patrick Roy.

As a predictive model, it works. Most people who rail against it are emotionally charged fans that are taking it personally.

Which stats have 7 years of data and what do you do with them?

If they're individual stats, how many players play on the same team, in the same role, and don't have any season to season variation in their play over seven years?

If they're team stats, which teams have the same group, the same coach, and don't have any season to season variation in their play?

You need a consistent environment and a much larger sample to account for the variation.

 

[Hasbro]

I swear to god if I see the word "sustainable" one more time...

 

[Pokecheque]

That's not true. The Avs were ranked 25th in Corsi, the Habs 26th.

The Habs were actually much better to start the year, then reverted to a pure dump-and-chase defensive scheme because Michel Therrien is an idiot.

And if you compare Corsi, yes, the Avs and Habs ended up comparable. If you measure Fenwick (Corsi minus blocked/missed shots) the Canadiens then pull away.

Most of the stats crowd will tell you that talent sometimes cancels out bad possession. Players like Phil Kessel or Matty Duchene are just good at putting the puck in net and therefore have an above-average shooting percentage. The "luck" thing when it comes to shot percentages usually refers to guys like Jamie McGinn and Alexander Steen, guys who aren't natural scorers who suddenly hit a "hot" streak and seemingly everything they throw at the net for a while goes in. And then for long stretches, it doesn't.

I think the Avs system helps in that regard as well. When you have your defensemen continually jumping into the play, that brings more weapons to bear, and as a result the team as a whole is just able to put the puck in the net more often.

 

[jsalz16]

Which stats have 7 years of data and what do you do with them?

If they're individual stats, how many players play on the same team, in the same role, and don't have any season to season variation in their play over seven years?

If they're team stats, which teams have the same group, the same coach, and don't have any season to season variation in their play?

You need a consistent environment and a much larger sample to account for the variation.

I believe, in 20 years, no team has won the cup with a neg goal differential in the reg season. 1 team (Pit) has won the cup with a neg SOG differential. That's pretty telling even if it ignores blocked or missed shots. Again, 20 years, how much data do you want?

Even if you don't subscribe to Corsi or Fenwick, no one can deny that being consistently outshot is bad for business.

Roy's strategy was to play man to man to try to limit quality scoring chances. When the man to man broke down, it was up to Varly all season to bail them out and he did most times. He didn't in game 7 by no fault of his own.

 

[Steerpike]

The stupidest thing about anyone saying that advanced stats predicted the Avs demise is that the whole point of looking at corsi and fenwik and shots is they offer a far larger sample size than goals alone. This measurement in a larger sample size should have less random variation than just goals.

However, while Minnesota and Colorado both went into the playoffs neck in neck in shot data during the regular season, the Wild massively outshot the Avalanche in the playoffs. In fact you could even argue by looking at score effects, the Avalanche were a slightly better possession team during the season.

This was supposed to predict that the Avalanche would take slightly more shots than the Wild during their series, while the PDO would be very unpredictable.

The truth was exactly the opposite. The Avs' PDO was substantially better than the Wild's, as was true during the regular season, but the shots were egregiously in favor of the Wild.

Nothing predicted this. While 48.3% shots for with the score close is probably fine (regular season), everyone knows that 37.5% (playoffs) is horrendous. It's the difference between getting outshot 29 31 and getting outshot 22.5 37.5.

 

[BB1133]

Which stats have 7 years of data and what do you do with them?

If they're individual stats, how many players play on the same team, in the same role, and don't have any season to season variation in their play over seven years?

If they're team stats, which teams have the same group, the same coach, and don't have any season to season variation in their play?

You need a consistent environment and a much larger sample to account for the variation.

The combination of PDO + Fenwick, used to predict a) making the playoffs and b) success once you get there.

In terms of season to season, a better question would be how many teams significantly change in one offseason? Everyone would love to add a Suter and Parise, but they can't. Most bottom dwellers would love to add a coach of the year, rookie of the year, and a Vezina goalie. How many do?

These types of changes from year to year are exceptions, not the norm. True shifts happen over time, but these stats can recognize those trends along the way, and help determine which are mirages and which aren't.

 

[jsalz16]

http://www.sportsnet.ca/hockey/nhl/more-nhl-teams-turning-to-advanced-analytics/

Brian Burke doesn't support advanced stats but apparently the Blackhawks do. Draw your own conclusions.

 

[tigervixxxen]

I agree shot location should be incorporated and other things that are not universally tracked or published such as zone entries/exits, zone time, scoring chances, actual time of possession. In baseball every single action on the field is recorded and quantified, in football every yard is broken down but in hockey nothing counts until the puck is directed at the goal or goes in. If someone actually sits at the game and counts hits and giveaways why can't they count scoring chances? Roy clearly does. If fans of certain teams break down all the zone entries/exits why can't the league? It's mind boggling.

 

[Foppa2118]

I believe, in 20 years, no team has won the cup with a neg goal differential in the reg season. 1 team (Pit) has won the cup with a neg SOG differential. That's pretty telling even if it ignores blocked or missed shots. Again, 20 years, how much data do you want?

Even if you don't subscribe to Corsi or Fenwick, no one can deny that being consistently outshot is bad for business.

Roy's strategy was to play man to man to try to limit quality scoring chances. When the man to man broke down, it was up to Varly all season to bail them out and he did most times. He didn't in game 7 by no fault of his own.

I'm not saying none of the stats can tell you a little piece of something. It's how much of the big picture do they tell, and what conclusions do you draw from them. As in what are the reasons for them, and what do you do to change it.

It doesn't have to be that complicated in hockey. The eye test, the roster, and any hockey knowledge you might have are overwhelmingly more useful.

Baseball is a game of plays, where only a few players are involved in the outcome at a time. That makes it much easier to use their actions to determine the outcome of the individual play, and the game.

Hockey has a much more symbiotic nature to it. What the defenseman does affects his partner big time. What the forwards do affect the defense. The goaltender letting in goals changes things. Little chance plays affect everyone. Play can go on for a long period of time without a whistle, and have ebbs and flows in momentum. It's an extremely quick game of reaction, and a game of mistakes.

It's not a game of stops and starts where how a player does at bat or the 2-3 players on defense involved in the play have a direct and instant impact on what happens.

My reaction to all these stats is usually who cares. Just assess your roster, how it played, and try to put together a good lineup, and keep it focused, motivated, and making as few mistakes as possible.

I don't need any of these stats to tell what's wrong with the Avs and why they lost, and if I do spend as much time focusing on them as their proponents do, it'd probably lead me towards having some complicated resolution to the problem or swapping players that won't really fix it.

There's no magic formula for success in hockey. You do what you can to address the problem areas, and you hope for the best. There's lots of good teams in the league, and just because you have a good one doesn't mean you'll win. It's like poker, you have to do certain things the right way, but it still takes a good amount of luck to win.

 

[PepsiCenterMagic]

It irks me, and probably others as well.

But there is a thread going on the GHD, and people are discussing the future of the Calgary Flames, and their playoff chances. Yet, when fans are discussing it, no one is bringing up any sort of advanced stats into the equation to help their answer.

And when Colorado is being discussed, it seems like everything Colorado will do has to do with such things. It doesn't matter that we have players X,Y,Z, in division Q. Yet, that's all everyone is using in their answers. If you asked them the exact same question involving Colorado, I bet 1/2 of them use Sabremetrics in their reasoning.

Colorado, unfortunately until they become a better "possession" team, or "regress" to expectations, will never be rid of this diseasing stigma.

 

[Pokecheque]

It irks me, and probably others as well.

But there is a thread going on the GHD, and people are discussing the future of the Calgary Flames, and their playoff chances. Yet, when fans are discussing it, no one is bringing up any sort of advanced stats into the equation to help their answer.

And when Colorado is being discussed, it seems like everything Colorado will do has to do with such things. It doesn't matter that we have players X,Y,Z, in division Q. Yet, that's all everyone is using in their answers. If you asked them the exact same question involving Colorado, I bet 1/2 of them use Sabremetrics in their reasoning.

Colorado, unfortunately until they become a better "possession" team, or "regress" to expectations, will never be rid of this diseasing stigma.

I agree 100%. But to me that's just a bunch of idiots who cling onto any excuse to justify their hate for a specific team. Most of those who jumped on the "Avs have bad Corsi" bandwagon likely have no real understanding of advanced stats, they're just chirping what they skimmed in a blog entry somewhere.

Yes, the Avs have to prove this isn't a fluke season, like 2009-10 proved to be. They can only do that by putting together another good season. And another one after that. People seem to forget the Bruins won the Cup with a team that gave up an ungodly number of shots and relied on their goaltender to bail them out. Lucky for them they had a goalie who, arguably, had the best season of any goaltender in NHL history. But now they're a very good possession team. The detractors just seem to think there is no foundation to build upon in Colorado, and that's just wrong. There's only one way of silencing most of them in any event.

But there will always be trolls, it's probably just best to ignore them.

 

[tigervixxxen]

Yes, the Avs have to prove this isn't a fluke season, like 2009-10 proved to be.

This leads to something that's always been one of my biggest questions., especially when people say a lottery team was not a legitimate division winner. It's not like the Avs had one fluke year in a sea of terrible, they also haven't had a linear trajectory of anything. How does one explain a team goes on this trajectory of points: 95, 69, 95, 68, 88, 67 (projected), 112. I seriously have no idea. Was the team ever as bad as it seemed? Everyone just wants to say they certainly were never that good in the good years.