Oilers analytic/advanced stat thread

[PerformanceMcOil]

cant disagree here. You invent a good metric you drive traffic to your website. The creator of GAR probably tripled website traffic/followers in past 4 months. I do believe they do have good intentions when making a metric and attempting to quantify the game. Id say the star power ranking fits into what you describe- more about shock value than credible model. Again, eye test has same issues as the "hot take" era is here to stay it seems. I see a ton of shock value articles written by MSM with not much evidence, but more about writting a piece that will generate buzz

I also agree with your last point. I think people do geniouly enjoy discussing hockey and when discussing players from different teams, stats are a great middle ground as so few people actually watch the suffficient amount of games to get a good view of both guys. Although this crosses the line at time when stats are held as the be all and end all.

Well, I don't think there is much more to say, since I think we agree. But, i do want to say I enjoyed this brief exchange with you, which has not usually been the case when I have waded into this issue, so thanks for that .

 

[Asiaoil]

Well, I don't think there is much more to say, since I think we agree. But, i do want to say I enjoyed this brief exchange with you, which has not usually been the case when I have waded into this issue, so thanks for that .

Agree that this was an unusually sane discussion and thanks PerformanceMcOil and Aceboogie. The available stats are, as you say, good for entertainment value. But that's the whole point and I also only get chufffed with people who are demeaning to others who don't share their perspective or who underplay factors they can't measure with their primitive tools.

I value GF% supported by lots of contextual information (ice time, zone starts, competition, teammates, WOWY etc etc). The whole point of the game is to outscore the opposition either through excellent offense, excellent defense or some combination of both and this gets to the point. REAL expert opinion by people who have a serious track record is also important. In the case of goalies which I spend more time looking at - a guy like Allaire's opinion mean's a lot more to me than a few boxcars. The stats guys can't sort out goalies so they just pretend it's unknowable voodoo - not really - guys like Allaire and other elite coaches can sort the wheat from the chaff darn well.

Anyway....thanks for the chat.

 

[nexttothemoon]

I value GF% supported by lots of contextual information (ice time, zone starts, competition, teammates, WOWY etc etc). The whole point of the game is to outscore the opposition either through excellent offense, excellent defense or some combination of both and this gets to the point.

That's what I think some/most don't get as well.

They throw out "advanced stats" entirely because they look at simple one-dimensional stats like corsi, zone entries, zone starts etc... and that's true, any one stat in a vacuum is relatively useless.

As you say when taking into account multiple stats and putting that into context with GF/GA relative to their teammates, stats tells a more complete story of what actually matters... goals for and against on the scoreboard.

 

[Zaddy]

No idea how this chart works, but Chiarelli is 8th. Pretty good, I assume.

 

[Aceboogie]

No idea how this chart works, but Chiarelli is 8th. Pretty good, I assume.

Breaks down gar additions by type of pick up. Mcdavid makes this for the Oilers but sekera and Lucic help

Although not sure how they get draft gar since only a limited number of guys have graduated

 

[Aceboogie]

That's what I think some/most don't get as well.

They throw out "advanced stats" entirely because they look at simple one-dimensional stats like corsi, zone entries, zone starts etc... and that's true, any one stat in a vacuum is relatively useless.

As you say when taking into account multiple stats and putting that into context with GF/GA relative to their teammates, stats tells a more complete story of what actually matters... goals for and against on the scoreboard.

Again as I showed before, GF is the least predictive of the stats

 

[nexttothemoon]

Again as I showed before, GF is the least predictive of the stats

I don't think you are understanding the stats then.

No one is saying use GF.

GF and GA has to be adjusted to take into account teammates, opposition, time on ice, zone usage, personal offensive stats etc.

You have to look at the correlative data that influences GF and GA and find out the relative GF and GA after taking into account all those factors that affect each player.

That study you linked doesn't take that into account. It's using raw GF/corsi etc... which is obviously flawed because it's too simplistic and doesn't take into account all the various underlying factors.

Those numbers are also on a team basis.. and individuals on a team are obviously going to be better/worse than a team average as a team is comprised of 23+ individuals who all influence GF and GA stats.

 

[Zaddy]

Breaks down gar additions by type of pick up. Mcdavid makes this for the Oilers but sekera and Lucic help

Although not sure how they get draft gar since only a limited number of guys have graduated

Doesn't Caggiula and Benning count into that FA category too?

I don't understand the draft thing either, maybe means players he has drafted in the past (meaning way back) that have gone on to play in the NHL in the last three years?

 

[Asiaoil]

I don't think you are understanding the stats then.

No one is saying use GF.

GF and GA has to be adjusted to take into account teammates, opposition, time on ice, zone usage, personal offensive stats etc.

You have to look at the correlative data that influences GF and GA and find out the relative GF and GA after taking into account all those factors that affect each player.

That study you linked doesn't take that into account. It's using raw GF/corsi etc... which is obviously flawed because it's too simplistic and doesn't take into account all the various underlying factors.

Those numbers are also on a team basis.. and individuals on a team are obviously going to be better/worse than a team average as a team is comprised of 23+ individuals who all influence GF and GA stats.

You made most of the key points nexttothemoon and well done. As you said, raw possession numbers and other secondary stats do not address the main point which is whether a player (in collaboration with his teammates) produces a positive goal impact while on the ice or a negative goal impact. Negative goal impacts can only be justified under very limited circumstances (eg killer zone starts, strong competition, weaker teammates etc ) with the objective of freeing up other lines to more than make up the deficit. But who cares if you have strong possession numbers or boxcars if you lack defensive commitment or make an endless series of weak plays that result in goals against and a crap GF%. Eberle was the posterboy for this until this last season where he finally started to take defense seriously after xmas. He generally had decent possession numbers, his boxcars were good, but he was a liability in spite of these. The Austins for all their flash simply dug a hole most nights and the other weaker lines only made it deeper. That's how you lose for 10 straight years.

Players are ultimately judged by their contribution to goal differentials and all kinds of factors play into that in a variety of ways. It's all about context and not all of the variables are easily quantifiable. People with a corsi or zone entry fetish are simplifying where simplification is not warranted, and indices out there like GAR are nothing but guesswork as their factor weighting appear to be entirely arbitrary. It's my understanding (from the few reports I've seen) that Chia values GF% supported by tons of secondary data and eyes (coaches, pro scouts etc). With this in mind it's not hard to understand this summer's moves as Pouliot and Hendricks were getting beat up by 4th liners. Eberle was traditionally weak in GF% as outlined above and how do you measure the impact of him bailing on contact in the playoffs versus Russell taking brutal shots and coming back for more? I think we can guess that the GM thinks this issue is important even if it's not precisely measurable. RNH had the worst ES GF% on the team but had killer competition so he gets some slack......but..... he's always had good linemates and he's been getting beat up by quality opposition for years. So the clock is ticking and you would like to see him at least break even doing that job @ $6 million per year.

The so called advanced stats (which are anything but) are actually very limited and mostly just provide context. Doesn't mean these are useless or wrong - but a lot of people have overplayed their significance to boost their site hits, get a job, or just feel smarter than others. I enjoy the numbers which help me evaluate players and the games, but in the end, it's who scored more / gave up less and why which is most important.

 

[Delicious Pancakes]

RIP https://puckalytics.com/
RIP stats.hockeyanalysis.com/

 

[McDNicks17]

Damn. I really liked puckalytic's Super WOWY.

 

[Aceboogie]

RIP https://puckalytics.com/
RIP stats.hockeyanalysis.com/

Dammit! Hockey Analysis was great, so easy to use and a ton of info. Great to see these guys get hired en masse the past 2 years but every time they do another great website/blog gets shutdown. Not even sure whats out there anymore (until Corsica goes back up).

Guess well just have to watch the games

 

[Aceboogie]

As far as that hiring goes, David Johnson is a total Burke hire. While Johnson has made 2 incredible websites, his views on analytics in hockey are.... controversial to say the least. Hes closer to a David Staples than he is to a Dellow/Yost when it comes to applications and interpretations of analytics. His viewpoints go directly against many well-researched theories. For example, he has a hard on for Russell because he believes individual players can drive on ice sv% (which has been proven to be luck based, not skill based)

For Burkie, he can now point to this hiring and tell fans/media "Look Im in the analytics game as well! Im hip right?!" But then have Johnson be a hire that wont question Burkes ways as their viewpoints mostly agree already

 

[Aceboogie]

Posting just for sharing sake.

How bout them Oilers. Plus for all the hype about Calgarys super D, youll have to scan near the bottom of the defense ranking to find them

 

[Aceboogie]

Did a bit of digging into Bennings underlying numbers. Man there could be a gem here. Using PuckIQ I gathered his numbers vs elite competition (since knock against him is that he was sheltered). He did play only 285 mins very elite competition, versus 649 for Suter (the leader) and 480 for Klefbom/larsson (leaders on the Oil). Here is where he ranks league wide

Dangerous Fenwick against/60: 36th in the NHL for D
DFF%: 42nd; 50.5%
CA/60: 20th
CF%: 25th; 50.5%
GF%: 24th; 61.1%
GA/60: 30th

Hes the best or 2nd best amongst Oiler D in this areas too

Like my god. Its amazing to see him top 30/45 across the board in these defensive stats. Only the really top D show up with that consistency, and almost no young D show up top 30 in so many areas. Even crazier considering hes a friggen rookie

Looks like hi splay is screaming for more challenging minutes because hes exceeding in minutes vs elite opposition

 

[McDNicks17]

As far as that hiring goes, David Johnson is a total Burke hire. While Johnson has made 2 incredible websites, his views on analytics in hockey are.... controversial to say the least. Hes closer to a David Staples than he is to a Dellow/Yost when it comes to applications and interpretations of analytics. His viewpoints go directly against many well-researched theories. For example, he has a hard on for Russell because he believes individual players can drive on ice sv% (which has been proven to be luck based, not skill based)

For Burkie, he can now point to this hiring and tell fans/media "Look Im in the analytics game as well! Im hip right?!" But then have Johnson be a hire that wont question Burkes ways as their viewpoints mostly agree already

I don't think that's entirely the case.

In most cases it's going to be luck, but I think a player like Russell actually does positively affect onSV%. It seems pretty logical for a player who allows lots of low danger chances to have a high onSV%.

Unfortunately that's just another thing that the public doesn't have enough data on to prove either way though. I assume NHL teams track stuff like that, but we barely even have a site with basics advanced stats anymore let alone stuff like that haha.

 

[Aceboogie]

I don't think that's entirely the case.

In most cases it's going to be luck, but I think a player like Russell actually does positively affect onSV%. It seems pretty logical for a player who allows lots of low danger chances to have a high onSV%.

Unfortunately that's just another thing that the public doesn't have enough data on to prove either way though. I assume NHL teams track stuff like that, but we barely even have a site with basics advanced stats anymore let alone stuff like that haha.

There is a couple macro studies that deal with this:

https://hockey-graphs.com/2014/07/07/defensemen-still-have-no-sustainable-control-over-save-percentage/

This study is great and probably the go to for any looking into this. Year over year On Ice sv% for D was almost never repeatable. The correlation was 0.026 which is extremely low and shows next to no correlation.

The 10% of players with the most extreme negative impact in their first two seasons end up having no impact on average over the next two seasons. In contrast the 10% of players with the most extreme positive impact in their first two seasons end up with a positive of about +0.005, less than 20% of the impact the previous two seasons.

theres a couple other great studies:
https://www.arcticicehockey.com/2014/6/26/5830516/save-percentage-defense-dustin-byfuglien-ondrej-pavelec-blamebuff
https://www.broadstreethockey.com/2013/7/4/4487304/save-percentage-variability-regression-defense
https://theleafsnation.com/2014/05/27/analytics-mailbag-save-percentages-pdo-and-repeatability/

The overwhelming conclusion being there is next to no evidence ON ice sv% is repeatable. Id say an issue with Johnson is that he still maintained this idea players had a major role in On ice save% when the overwhelming evidence was the opposite. It is all fine if you want to disagree, but if he considers himself an academic when it comes to hockey, he would have provided studies or evidence to the contrary. Instead, he relied on primarily on anecdotal evidence. Made a great website, but never actually relied on statisitcal evidence

In regards to Russell: He allows high danger chances at a rate similar to rest of the D core. Only Benning separates himself from the pack for letting fewer high danger chances, Nurse for allowing more HD chances. Russells ON ice save % has also varied wildly year over year, there is no resemblance of repeatability. There are 2 years of high on ice sv% but then years of really low on ic sv%. Theres also no evidence he allows less high danger chances than average in the past either. Hes always ranked around team average of High danger chances. Which is not bad but no way youd explain higher on ice save % as a result

 

[Bank Shot]

I don't think that's entirely the case.

In most cases it's going to be luck, but I think a player like Russell actually does positively affect onSV%. It seems pretty logical for a player who allows lots of low danger chances to have a high onSV%.

Unfortunately that's just another thing that the public doesn't have enough data on to prove either way though. I assume NHL teams track stuff like that, but we barely even have a site with basics advanced stats anymore let alone stuff like that haha.

I find stats guys are sometimes pretty quick to say they have proven something that they actually haven't proven at all.

I've seen a few stats/idea go through the matters-not matters-back to matters cycle. lol.

 

[Aceboogie]

I find stats guys are sometimes pretty quick to say they have proven something that they actually haven't proven at all.

I've seen a few stats/idea go through the matters-not matters-back to matters cycle. lol.

I dont think On ice save % or shooting % has even gone tru anything like that because it can be proven pretty much definitely. There is a clear way to measure it and the outcomes are pretty binary

Things like quality of comp/teammates/zone starts effect on possession have gone tru varying degrees of changes because there is a ton of variables that go into it and the outcome is not as clear. I also do not think anyone ever concluded they had definitive "proof". Its like global warming research. Sure you can find studies that say its worse than thought or not as bad, but the overwhelming majority show a correlation

As far as On Ice save % goes, the counter argument is pretty much always an isolated example of a player maintaining a unusually high/low sv % for 2 years. Ive yet to run across a study showing otherwise that actually uses a large sample size. But if you look an the entire sample size over multiple years youll see little to no correlation. Although the defense someone on the other side could use is that it can be repeatable over 2 years, which is very much relevant. All studies use a 3 year/5 year period to show its not a repeatable skill. But 3/5 years might be too long of a time frame for people to care. I mean if I said, Kris Russell will have a average on ice save % in 3 years and well probably all turn on him en masse, thatd probably be met with "meh what can you do"

A good anaology is that its been shown almost definitevely you can not beat the stock market over a long enough sample size. But still every year 1 fund manager pimps how much they beat the stock market, then the next year its a different one. So yes every year there is isolated incidents, but on the whole its pretty much impossible

 

[Bank Shot]

As far as On Ice save % goes, the counter argument is pretty much always an isolated example of a player maintaining a unusually high/low sv % for 2 years. Ive yet to run across a study showing otherwise that actually uses a large sample size. But if you look an the entire sample size over multiple years youll see little to no correlation. Although the defense someone on the other side could use is that it can be repeatable over 2 years, which is very much relevant. All studies use a 3 year/5 year period to show its not a repeatable skill. But 3/5 years might be too long of a time frame for people to care. I mean if I said, Kris Russell will have a average on ice save % in 3 years and well probably all turn on him en masse, thatd probably be met with "meh what can you do"

If Russel has a save percentage that is even half to one percent better than average over a 5 year span than that means something. Most of the margins in advanced stats are very small to begin with. Even a 1% difference matters.

Corsi isn't repeatable for A LOT of players from year to year but no one is throwing that idea out the window.

I've seen tonnes of players with corsi that bounces like yo-yos from season to season.

All the stats guys have proven is that they can't isolate a single player's effect on save percentage/shot quality.

No one has come even close to proving that the effect doesn't exist.

I think logically that everyone must realize that good defence and bad defence exist on some level, not just solely offence>defence.

 

[Zaddy]

 

[Aceboogie]

All the stats guys have proven is that they can't isolate a single player's effect on save percentage/shot quality.

No one has come even close to proving that the effect doesn't exist.

I think logically that everyone must realize that good defence and bad defence exist on some level, not just solely offence>defence.

This is true. But whats also true is that luck plays a major role in sports, specifically the NHL. This is important because people wants stats to explain everything, with the notion everything in hockey is 100% skill influenced with no luck influence. Here is a great video on the subject and theres also great studies into luck in sports (including NHL) and the big impact it has. From the study below, NHL actually led all sports in terms of contribution of luck with 53%.

Its a bit of a futile exercise to say "Well stats guys cant explain individual impact on save %" because 53% of the NHL is luck. So a majority of what drives save % is luck, pure luck. I think to ignore the role of luck is a bit nieve. This is a main reason that over 3 years, a perceived elite defensive D can have lower on ice sv% and a mediocre D can have a good on ice save %

Its kind of like going to a roulette table that just had red 18 come up 3 times in a row and saying "Well stats cant explain this, the casino hand can obviously influence the outcome"



Corsi also does vary each year 100%. But those stats use a much larger sample size (up to 10X more than goal stats), so the variability and luck influence is not as much as goal based stats

 

[Aceboogie]

This is a good way to help formulate lines. Kassian and Slepyshev are both good shooters with a shoot first mentality but not as good entering the zone. So both should be on different line, preferably with a good zone entry forward who likes to pass

 

[Aceboogie]

One area im changing my opinion on things is shots by D (or just benefit of raw shot generation in general). Before I held if a D is generating a ton of shots then he is doing his team a big benefit. However, when you have the puck you want your best shooters shooting as you really get a limited number of chances at a shot. For a D to blast one from the point is much less dangerous than a forward with a better shot taking one from in close. Klefbom and Nurse have amazing shot generation numbers, but only Klefbom has that plus shot. Nurse shooting more might actually be costing us more dangerous shots

I do not think this excuses low shot generation numbers all together, but it might re-adjust past thinking. Maybe a D with lower shots/60 and higher SF/60 (team stat) is helping the team more than a D who shoots more by himself

 

[Bjornar Moxnes]

So with my work on ranking every NHL team defensively, offensively, and overall completeness as a team, Edmonton is the 17th best defensive team when you include goals against metrics as well. I haven't done just purely shot suppression metrics. Edmonton is actually one of the better teams at preventing fenwick and shot against chances, and allowed the 8th least amount of goals. However Edmonton seems to allow a ton of dangerous chances and thus a lot of high danger goals. That's one major area Edmonton needs to improve.