The Analytics say Goaltending is the Reason for the Oilers Resurgence under Coffey, NOT improved D Play. An interesting case study on analytics.

[TheNumber4]

Saw an interesting video that compared the xGF% and SCF% of the Oilers D core before Coffey took over and after Coffey took over. Comparing those metrics before and after for every defenceman in the line up, the following conclusion was made (based on the numbers only):

"Play driving and possession has actually regressed under Paul Coffey with this D core, but the Defence PDO has started regressing to the mean, if we were to look at this from an analytical standpoint I would actually be concerned moving forward with the team but they are getting better goaltending, the goals against are starting to go down, but that is just because of better goaltending (and not cause of the defence)."

So do you agree or disagree with his conclusion?

Personally, I'm a strong disagree with that conclusion. While goaltending has improved, it's done so on the back of a much better defence.

You can watch the video, see the numbers used, and hear his analysis here:

 

[Cloned]

Analytics still has a long way to go in hockey.

It is one of the hardest sports to model using numbers.

The eye test is still generally more reliable, which can’t really be said for any other sport with advanced analytics.

Most knowledgeable hockey observers would tell you that the Oilers are playing better overall hockey under Knob and Coffey.

 

[mkatcherin00]

Weird. To me, goaltending is the product of what we are doing. Skinner is not being tested like crazy at all. We are just playing better overall, everywhere.

 

[foshizzle]

Analytics still has a long way to go in hockey.

It is one of the hardest sports to model using numbers.

The eye test is still generally more reliable, which can’t really be said for any other sport with advanced analytics.

Most knowledgeable hockey observers would tell you that the Oilers are playing better overall hockey under Knob and Coffey.

I respectfully and completely disagree on this. Analytics doesn’t have a long way to go, people’s understanding of it does. The eye test is not reliable at all. Eye tests produce biases and the sport moves way too fast to see everything that happens in the sport. The eye test had people trading Bouchard and an asset for Tyson Barrie 10 games ago.

This persons assessment is only partially correct. A shot heat map would show the Oilers are giving up way less slot shots and are contesting more shots. Puck retrievals, zone denials, zone exits, etc all play into this which have been largely ignored. Woodguy is a good guy to follow

 

[Cloned]

I respectfully and completely disagree on this. Analytics doesn’t have a long way to go, people’s understanding of it does. The eye test is not reliable at all. Eye tests produce biases and the sport moves way too fast to see everything that happens in the sport. The eye test had people trading Bouchard and an asset for Tyson Barrie 10 games ago.

This persons assessment is only partially correct. A shot heat map would show the Oilers are giving up way less slot shots and are contesting more shots. Puck retrievals, zone denials, zone exits, etc all play into this which have been largely ignored. Woodguy is a good guy to follow

Eye test isn’t perfect by any means but hockey has too many chaotic variables moving at a high speed to be properly modeled using the metrics that are available currently.

The analytics that were praising Bouchard early on despite his inconsistent play are the same analytics that are now downplaying the effect of the coaching change on the performance of the defense.

 

[Aerchon]

Saw an interesting video that compared the xGF% and SCF% of the Oilers D core before Coffey took over and after Coffey took over. Comparing those metrics before and after for every defenceman in the line up, the following conclusion was made (based on the numbers only):



So do you agree or disagree with his conclusion?

Personally, I'm a strong disagree with that conclusion. While goaltending has improved, it's done so on the back of a much better defence.

You can watch the video, see the numbers used, and hear his analysis here:

Analytics still has a long way to go in hockey.

It is one of the hardest sports to model using numbers.

The eye test is still generally more reliable, which can’t really be said for any other sport with advanced analytics.

Most knowledgeable hockey observers would tell you that the Oilers are playing better overall hockey under Knob and Coffey.

Cloned has it.

I would go so far to say any fool can see the Oiler net minders are not facing the quality of chances they were to start the year and that is primarily on the team defense. I've rarely... maybe never seen so many grade AAAAAAAA+ chances against where I feel like a broken record saying, "no chance on that shot", " no chance on that shot", "wtf what was the defense doing there?"

Blatant.

Its stuff like this that just makes NHL analytics look useless.

A shot heat map only touches on the quality of shots. Doesn't take into account on uncontested/misplayed 2 on 1's... which I "feel" we had a dozen of in the first ten games. Opportunities where NHL shooters would be ashamed if they missed the wide open net.

Edit: I am a big Bouchard fan. I wasnt thinking trade him but the eye test rightly observed his massive collapse in his defensive play. I joke but a midget player could play better defense than Bouchard did to start the year and if the Analytics cant make that determination. LOL that is very bad.

 

[TheNumber4]

Analytics still has a long way to go in hockey.

It is one of the hardest sports to model using numbers.

The eye test is still generally more reliable, which can’t really be said for any other sport with advanced analytics.

Most knowledgeable hockey observers would tell you that the Oilers are playing better overall hockey under Knob and Coffey.

I agree. I think also, it was Dustin Brown I was listening to go into detail about the difference between Twitterlytics and “real” analytics that NHL teams use.

Essentially his point was all internet analytics are based off publicly available data tracked by the NHL, so you know stuff like shot volume and shot location which essentially forms the basis of all Twitterlytics like Jfresh. Whereas NHL teams have way more data points to draw from. For example something that doesn’t seem to be tracked on Twitterlytics when they rate shot quality is whether a cross crease pass preceded the shot in question I think. Another person put it this way, the twitterlytics give you a decent idea but it doesn’t paint the complete picture, and NHL teams with their investments in specialized data firms do have access to those additional data points that can paint a fuller picture.

 

[Connor McConnor]

Hockey analytics can’t figure out goaltending. Improved defensive structure benefits goaltending stats (see any Nashville goalie ever)

 

[TheNumber4]

Weird. To me, goaltending is the product of what we are doing. Skinner is not being tested like crazy at all. We are just playing better overall, everywhere.

Weird indeed. Cause I agree, there has been a marked improvement on that front. Take yesterdays game for example, the only grade AAA chance they had just happened to go right into a wide open gaping net from the one defensive breakdown we had all game. The rest of the game were just your regular wristers, barely a one timer to be seen in sight or odd man rush. Twitterlytics would rate that one timer shot that went in as the same danger level as a player standing and just shooting from that exact same spot but with a stand still wrister. These shots are not the same danger level though and I think that’s where Twitterlytics currently fails.

 

[TheNumber4]

Cloned has it.

I would go so far to say any fool can see the Oiler net minders are not facing the quality of chances they were to start the year and that is primarily on the team defense. I've rarely... maybe never seen so many grade AAAAAAAA+ chances against where I feel like a broken record saying, "no chance on that shot", " no chance on that shot", "wtf what was the defense doing there?"

Blatant.

Its stuff like this that just makes NHL analytics look useless.

A shot heat map only touches on the quality of shots. Doesn't take into account on uncontested/misplayed 2 on 1's... which I "feel" we had a dozen of in the first ten games. Opportunities where NHL shooters would be ashamed if they missed the wide open net.

Edit: I am a big Bouchard fan. I wasnt thinking trade him but the eye test rightly observed his massive collapse in his defensive play. I joke but a midget player could play better defense than Bouchard did to start the year and if the Analytics cant make that determination. LOL that is very bad.

Yeh i agree on your second point. What it currently takes into account is shot location, but stuff like a 2 on 1 with a preceding cross crease pass are ignored currently with the shot data publicly available. There’s an old adage in the stats world, I learned in Stats 101, “Garbage In Garbage Out”. The quality of that data needs to be high if you are going to get relevant insights from it.

 

[Davo Ikinzom]

It's all a bunch of gobbledygook as far as I'm concerned.

 

[KCC]

Weird. To me, goaltending is the product of what we are doing. Skinner is not being tested like crazy at all. We are just playing better overall, everywhere.

And that's what, well not scares me but, doesn't make me trust Skinner still because when there are more danger chances will be be able to actually hold the fort and make those key saves when they matter? Currently pretty much every shot I could probably stop because the team is doing such a good job in front of him. He either plays like trash or hasn't had to really do anything. When the team has off nights can he actually steal some games? That would be nice and would make me trust him more. lol

That's why (hot take) I think goalie's like Brodeur were so overrated because his best years were during the dead puck era with no cap, an insane defense in front of him and the team trapped. He only had to stop like 20 shots a night if he was lucky and they were dribblers to the net. It's why I think goalies like Roy and especially Hasek were much better. They did way more with much less. But all that is for another day. lol.

 

[jukon]

Hockey analytics can’t figure out goaltending. Improved defensive structure benefits goaltending stats (see any Nashville goalie ever)

And for bad defense, see any Edmonton goalie ever.

 

[joestevens29]

Yeh i agree on your second point. What it currently takes into account is shot location, but stuff like a 2 on 1 with a preceding cross crease pass are ignored currently with the shot data publicly available. There’s an old adage in the stats world, I learned in Stats 101, “Garbage In Garbage Out”. The quality of that data needs to be high if you are going to get relevant insights from it.

It was kind of nice when a guy like Staples or even Gregor once and awhile calculates their own scoring chances.

Gregor's data was good on a game by game basis as it would tell you if we were winning the battle in the game.

Staples though like most his stats were only good to compare the Oilers against themselves.

Essentially for that data to ever be good you need a group of people that all have the same thought process and watch every single game.

This whole shot location means little to me. There is a lot more to it then where the puck was shot from

 

[McDNicks17]

I wouldn’t read too much into xGF% for them this year. Public models were doing a terrible job of gauging how dangerous chances were.

Something like heat maps show a massive improvement in D structure. Way less ten bell chances around the net.

 

[Dazed and Confused]

Only going by xGF% and scoring chances isn't really that deep of an analysis. Especially when you don't control for HDSC, which was one area the OIlers were getting killed in to start the year.

Plus how confident can you be in statistical conclusions drawn from ~10 game same sizes? There's way too many variables that can influence things.


As @McDNicks17 points out, the heat maps look way better. The Oil may be giving up a few more shots/ low grade chances, but they're from the outside vs. in the slot or off an odd man rush.

In short, not all scoring chances are created equal. I'm going to start calling this the Dallas Eakins Folly, for his brilliant mastermind idea to cheese for more shots, because that **must** equal more goals. Regardless of where they come from, because those things average out over the course of the season.

 

[mkatcherin00]

And that's what, well not scares me but, doesn't make me trust Skinner still because when there are more danger chances will be be able to actually hold the fort and make those key saves when they matter? Currently pretty much every shot I could probably stop because the team is doing such a good job in front of him. He either plays like trash or hasn't had to really do anything. When the team has off nights can he actually steal some games? That would be nice and would make me trust him more. lol

That's why (hot take) I think goalie's like Brodeur were so overrated because his best years were during the dead puck era with no cap, an insane defense in front of him and the team trapped. He only had to stop like 20 shots a night if he was lucky and they were dribblers to the net. It's why I think goalies like Roy and especially Hasek were much better. They did way more with much less. But all that is for another day. lol.

I agree. We still need a goalie to split games.

 

[McDoused]

Did they do a pearson or spearman correlation test to prove that it was goaltending? or they just making things up?

Edit: watched the video and yeah its someone who doesn't understanding analytics

 

[ZJuice]

What if I told you my brain processes 100 billion queries per second and can outperform any advanced stats graphs or reports? Same with anyone with a good head for hockey and a good pair of eyes.

Advanced stats are good for trying to identify diamonds in the dirt when you don't have time to watch a bunch of different players 7-30 minutes of ice time per game, but as we've seen with Puljujarvi, advanced stats don't mean shit if the player is dumb as a doorknob

 

[joestevens29]

What if I told you my brain processes 100 billion queries per second and can outperform any advanced stats graphs or reports? Same with anyone with a good head for hockey and a good pair of eyes.

Advanced stats are good for trying to identify diamonds in the dirt when you don't have time to watch a bunch of different players 7-30 minutes of ice time per game, but as we've seen with Puljujarvi, advanced stats don't mean shit if the player is dumb as a doorknob

I'd add it's also good to identify players usage within your own club. If a guy seems to be struggling a quick look at some stats might suggest that he's better in a different situation than the coach is using him. Granted even with that data you might have no choice.

Just thinking about Bernier as I was looking at him earlier. Let's just say the eye test shows he's playing similar to last year, is there something different with how he's being used this year to show why he's on pace for 20 less points?

 

[foshizzle]

Eye test isn’t perfect by any means but hockey has too many chaotic variables moving at a high speed to be properly modeled using the metrics that are available currently.

The analytics that were praising Bouchard early on despite his inconsistent play are the same analytics that are now downplaying the effect of the coaching change on the performance of the defense.

Thanks for your thoughts- analytics don’t praise or denounce anything- they just record what is happening. That is all. They don’t take into account of any feelings or biases. Just what is happening.

I disagree with hockey being too chaotic. It really isn’t. Every industry on the planet has KPIs. Every action that a human takes has probabilities and risk as measured by actuaries. Eye tests are inherently biased.

Good debate and good points made by you, however

 

[PositiveCashFlow]

Our goaltending is better because we’re defending as if we have an EBUG in net

 

[HockeyGuy1964]

Saw an interesting video that compared the xGF% and SCF% of the Oilers D core before Coffey took over and after Coffey took over. Comparing those metrics before and after for every defenceman in the line up, the following conclusion was made (based on the numbers only):



So do you agree or disagree with his conclusion?

Personally, I'm a strong disagree with that conclusion. While goaltending has improved, it's done so on the back of a much better defence.

You can watch the video, see the numbers used, and hear his analysis here:

Considering I've said this ad nauseum over the past little while in other threads, I'll be forced to agree.

It's not coaching just like it wasn't coaching when they fired Tippett & brought in Woodcroft.

 

[ZJuice]

Thanks for your thoughts- analytics don’t praise or denounce anything- they just record what is happening. That is all. They don’t take into account of any feelings or biases. Just what is happening.

I disagree with hockey being too chaotic. It really isn’t. Every industry on the planet has KPIs. Every action that a human takes has probabilities and risk as measured by actuaries. Eye tests are inherently biased.

Good debate and good points made by you, however

And who is responsible for determining a high quality shot versus a low quality shot?
Who determines what is a takeaway or a giveaway?
Just like how I know Gagner didn't actually score 8 points in that one game even though the stats guys gave it to him.

what I'm getting at.. Is the basis of your advanced stats based on the eye test? Some invisible boundaries in the ice that were determined by a human?

These stats are as flawed as an eye test!!!

Real life example.. Say a shipper only has a KPI of 50%. Upper management see this and fire him. Suddenly the rest of the shipping team drops from 120% to 80% of their targets. But someone who worked with the 50% guy knows he did all of the dirty complicated shipments. Now he's gone and everyone has to do the shit work. Advanced stats said the guy was slow though!!

Then there was our corsi under Eakins.. They were coaching to increase corsi. That introduced some bias into these "factual" stats

 

[Soundwave]

Skinner is OK if he can be square to the shooter but he's crap at lateral, scramble, and desperation style saves.

So he can look OK in stretches if you can let him just set up in one spot for shots against and not move too much, the problem is you'll hit a game or series where you need more and he can't provide that.

Our D has been better, but it would be hard to be worse than the Eakins-style D that was going on the first 10-12 games. And McDavid scoring at 2 ppg again the last 8 games helps everyone feel more confident.