Large enough sample sizes then yes you should be able to factor out a lot of variables and get reasonable stats to compare guys playing different roles. Maybe imperfect but things worth comparing especially if you look at multiple different measurements. You can run differences in differences and factor in a lot. Especially if a guy had a full year of data.

If ice time is drastically changed I would have a concern. A forward playing 12 min a night should be skating harder than a guy playing 18 min. Less time on ice tired, but any advance stat worth quoting should be factoring in zone starts and qoc. That really wouldn't be too hard to come up with. But I honestly don't which stats people quote factor in those things and what are more of a simple average.

There are times to get away from stats when you have a quality reason. Schultz obviously had some psychological issues coming from Edmonton. But plenty of other reasons to ignore stats such as coming off injury or maybe a guy had family issues. But some things should be showing up in quality stats.
- sditulli

Every game is different. Every environment is different, every opposition is different, every time a player plays he will play different etc etc etc. you want more differences in your data to provide you an accurate appraisal of a players ability?? Each year a player will play differently whether it be better or worse. He's the same player playing under so many different variables & not even able to reproduce the same effort due to so much change.

Sid played like garbage at the start of the season, then played like the champion he is. But the data in 2 years will show his numbers as an average & not as independent or broken down periods. It's not a true reflection of how he played. Now if Vermette for example was to completely roll over at the start of the season, he would get traded or waived (I would think), as a contender won't take the chance if they can replace him. If he shadows Sid's form from this season (at his level) how are you going to know by analytics? I truly believe for hockey analytics players should be measured in smaller sample sizes giving more acurate numbers with how they actually play not generalised overall stats. The later the data the better the chance you have of seeing how a player will play - 2 years ago player X may have been 10kg overweight & getting on the cans every night. Now he might be 10kg's lighter with a steady partner & living a healthy life - how does analytics show you that?

Well he was attempting to show the difference between two different years and if you read the fine print, it shows him in a 1st C role and a 2nd C role.

Vermette has always been a 3C, and maybe a borderline 2nd C earlier in his career. When you take a 3C and play him in a 1st C role, you're going to get the results you see in those HERO charts. Plus, when you take an aging player who's now more of a 3C/4C and play him in a 2nd C role, you're going to get those results you see.

As we've seen with Daley and Schultz, player usage is HUGE.

EDIT: Plus, comparing Cullen and Vermette's HERO charts would be inaccurate considering their roles and usage.
- Rinosaur

So show Vermette's HERO based on his time in CHI. Pretty sure he was not 1C or 2C there and he wouldn't be here. Even tho his time in CHI would be a small sample its comparing apples to apples. CHI system and his usage there would be a better comparison to PITT. Also, Vermette's role and usage in CHI would pretty identical to Cullens here, last year.

Any chart showing his analytics as a 1C or 2C in ARI is going to suck we can all agree on that.

Every game is different. Every environment is different, every opposition is different, every time a player plays he will play different etc etc etc. you want more differences in your data to provide you an accurate appraisal of a players ability?? Each year a player will play differently whether it be better or worse. He's the same player playing under so many different variables & not even able to reproduce the same effort due to so much change.

Sid played like garbage at the start of the season, then played like the champion he is. But the data in 2 years will show his numbers as an average & not as independent or broken down periods. It's not a true reflection of how he played. Now if Vermette for example was to completely roll over at the start of the season, he would get traded or waived (I would think), as a contender won't take the chance if they can replace him. If he shadows Sid's form from this season (at his level) how are you going to know by analytics? I truly believe for hockey analytics players should be measured in smaller sample sizes giving more acurate numbers with how they actually play not generalised overall stats. The later the data the better the chance you have of seeing how a player will play - 2 years ago player X may have been 10kg overweight & getting on the cans every night. Now he might be 10kg's lighter with a steady partner & living a healthy life - how does analytics show you that?
- Aussiepenguin

Smaller sample sizes are bad. Larger sample sizes are good. Would need a good reason to use a smaller sample size over a larger sample size (like signs of aging, signs of player development, was hurt for part of sample size, changed schemes, etc.). analytics aren't perfect, but they are a good place to start when evaluating a player.

So show Vermette's HERO based on his time in CHI. Pretty sure he was not 1C or 2C there and he wouldn't be here. Even tho his time in CHI would be a small sample its comparing apples to apples. CHI system and his usage there would be a better comparison to PITT. Also, Vermette's role and usage in CHI would pretty identical to Cullens here, last year.

Any chart showing his analytics as a 1C or 2C in ARI is going to suck we can all agree on that.
- stowerkraut

Vermette was actually used mostly as the 2nd line C. He wasn't really that good, but he definitely held his own. Scored some pretty clutch goals too. Given the depth of the team, he'd be fine as the 4th line C. There will be a time where 2 of our top 3 centers are injured, and vermette could hold down the fort as a 2nd or 3rd center for a short time if needed. Im also assuming he would sign a cheap 1 year contract at no more than 1 mil.

Smaller sample sizes are bad. Larger sample sizes are good. Would need a good reason to use a smaller sample size over a larger sample size (like signs of aging, signs of player development, was hurt for part of sample size, changed schemes, etc.). analytics aren't perfect, but they are a good place to start when evaluating a player.
- sditulli

You're (not you but analytics), trying to apply scientific strategies to a volatile entity that abides by no scientific principals. How does a players form from years previous show what he will do next season? As I said before Sid finished strong to produce a season of stats that would follow previous years. But how he got there is important but is 'generally' ignored by the statistics. They generalise & are basically an average. I would love for someone to provide the data on teams using analytics & aquiring players based on them & see how many succeed & how many fail. which point of view will people take - look at the comments on here alone for Vermette. What is he, a solid 3/4 C or a poop player due to his 'numbers' telling us that? (Numbers that probably don't tell us what he can do playing in a 3/4 C role)

I've seen it on here that the analytics can produce contradictory results for the same person, it's just which results you want to use to justify your prospective. That in itself should sound the alarm.

I realize I'm old and just too stupid for fancy chart reading, but I can read these numbers...14, 14, 9, 9, 9, 17, 19, 6, and 6. Those would be Vermette's faceoff % rankings for the past 9 years. I can't seem to find that info in the charts comparing Vermette to himself(????), so I guess winning a faceoff consistently no longer has any relevance in the new NHL and just doesn't count towards driving possession. Add to that the number 37 which is his lowest point total in the past decade, aside from the season he only played 48 games.

Does everyone really believe Sundqvist will come close to 37 points or a 56% on faceoffs as a rookie? Or Fehr will who only once has surpassed 33 points and whose career faceoff % is 48.7? I see absolutely no reason to think Vermette would not be a suitable 4th line center for this season...besides the past his due date reason offered in the blog??? I never realized that hockey players came stamped with a specific best by date like a gallon of milk I think we could do far worse replacing Cullen for this year if need be.
- Pags

Raw point totals are often misleading and can often be a function of ice time. Thats why it's important to consider a player's points per 60 minutes. Yes, sometimes if a player is logging low minutes, their P/60 can get super inflated with high shooting percentages, but again, as always, it's important to look at while pictures and never take a stat in isolation.

However, in Vermette's case, he logged top 6 minutes in Arizona and put up a very pedestrian 4th line, almost replacement level P/60 this last year, and his P/60 has been steadily declining for the past four or so years. Factor in his poor possetion metrics and it's easy to see that Vernette is neither a player that gives adequate tangible production, nor is the majority of the offense in his team's favor when he's on the ice.

So to answer your question, do I think Fehr or Sundqvist will put up 37 points next year? No, and a lot of that is because Fehr and Sundqvist will be on a great team next year where their role will be small and mostly defensive. Now, do I think Fehr or Sundqvist could outprofuce Vermette in the role of a defensive 4C? I am at the very least very confident Fehr can and I'm of a mind that Sundqvist, who'll come cheaper, wouldn't be too far behind Vermette.

As for faceoffs? I think for the most part advanced stats guy try to play down the importance of faceoffs a little bit too much like they do with hits. They're not worthless parts of the game, but for the most part I agree that the way we keep track of them is pretty worthless. Hits and faceoffs aren't inherently worthless but there tons of worthless hits and faceoffs, so keeping raw numbers isn't very helpful and there's no current metric that can help substantiate them. So hiring a guy that logs a ton of hits or has a high F/O% is not a good reason to hire that person over another if he doesn't do much else. The Pens need to look no further than Joe Vitalle or Mac Lapierre for a reason to not sign a guy just because they draw at over 50%, and in Vermette's case his raw draw numbers are really the only thing keeping him relevant.

Raw point totals are often misleading and can often be a function of ice time. Thats why it's important to consider a player's points per 60 minutes. Yes, sometimes if a player is logging low minutes, their P/60 can get super inflated with high shooting percentages, but again, as always, it's important to look at while pictures and never take a stat in isolation.

However, in Vermette's case, he logged top 6 minutes in Arizona and put up a very pedestrian 4th line, almost replacement level P/60 this last year, and his P/60 has been steadily declining for the past four or so years. Factor in his poor possetion metrics and it's easy to see that Vernette is neither a player that gives adequate tangible production, nor is the majority of the offense in his team's favor when he's on the ice.

So to answer your question, do I think Fehr or Sundqvist will put up 37 points next year? No, and a lot of that is because Fehr and Sundqvist will be on a great team next year where their role will be small and mostly defensive. Now, do I think Fehr or Sundqvist could outprofuce Vermette in the role of a defensive 4C? I am at the very least very confident Fehr can and I'm of a mind that Sundqvist, who'll come cheaper, wouldn't be too far behind Vermette.

As for faceoffs? I think for the most part advanced stats guy try to play down the importance of faceoffs a little bit too much like they do with hits. They're not worthless parts of the game, but for the most part I agree that the way we keep track of them is pretty worthless. Hits and faceoffs aren't inherently worthless but there tons of worthless hits and faceoffs, so keeping raw numbers isn't very helpful and there's no current metric that can help substantiate them. So hiring a guy that logs a ton of hits or has a high F/O% is not a good reason to hire that person over another if he doesn't do much else. The Pens need to look no further than Joe Vitalle or Mac Lapierre for a reason to not sign a guy just because they draw at over 50%, and in Vermette's case his raw draw numbers are really the only thing keeping him relevant.
- Victoro311

puckbase.com

faceoffs are irrelevant.

Here is Vermette's projected point totals being generous and assuming no regression by using his production rate from last season and giving him Cullen's ice time.


5v5: 848.12 mins @ 1.07 p/60 = 15.12 pts
5v4: 34.98 mins @ 3.51 p/60 = 2.04 pts
4v5: 203.93 mins @ 0.00 p/60
4v4: 23.28 mins @ 0.00 p/60
5v3: 0.83 mins @ 12.61 p/60 = 0.17 pts
3v5: 3.58 mins @ 0.00 p/60
3v3: 8.05 mins @ 0.00 p/60

A whopping total of 17.33 pts

Vermette was actually used mostly as the 2nd line C. He wasn't really that good, but he definitely held his own. Scored some pretty clutch goals too. Given the depth of the team, he'd be fine as the 4th line C. There will be a time where 2 of our top 3 centers are injured, and vermette could hold down the fort as a 2nd or 3rd center for a short time if needed. Im also assuming he would sign a cheap 1 year contract at no more than 1 mil.
- drummer829

Putting up 1.04 P/60 at even strength as a 2C is inexcusable I don't care how bad the team around him was. That's not holding your own in that role.

puckbase.com

faceoffs are irrelevant.
- Feds91Stammer

Something tells me faceoffs are not completely irrelevant. But I do get the basic logic that they really are not that important.

Doesn't it come down to a very low % of faceoffs being won cleanly. So if a faceoff guy at 45% and a guy at 55% isn't reallly as big of difference as you would think at first.

And other stats are probably catching the advantages. IE. a Faceoff won leading to a goal would often give the center an assist. Or a faceoff won leading to possession would show up in corsi. So if a guy has the same corsi% but lower faceoff win rate then he's doing something else that is driving corsi and it would basically be double counting the possession gain by looking at the faceoff %.

You're (not you but analytics), trying to apply scientific strategies to a volatile entity that abides by no scientific principals. How does a players form from years previous show what he will do next season? As I said before Sid finished strong to produce a season of stats that would follow previous years. But how he got there is important but is 'generally' ignored by the statistics. They generalise & are basically an average. I would love for someone to provide the data on teams using analytics & aquiring players based on them & see how many succeed & how many fail. which point of view will people take - look at the comments on here alone for Vermette. What is he, a solid 3/4 C or a poop player due to his 'numbers' telling us that? (Numbers that probably don't tell us what he can do playing in a 3/4 C role)

I've seen it on here that the analytics can produce contradictory results for the same person, it's just which results you want to use to justify your prospective. That in itself should sound the alarm.
- Aussiepenguin

You do realize all scientific laws are based on "volatile entities". Scientific principles measure the exact same things as we are measuring here, but they are measuring things with sample sizes larger than I can even pronounce. So they average out into neat scientific laws.

Analytics are just a way of predicting likely results. But yes any given entity is a bit random, but still predictable in a broad sense.

Something tells me faceoffs are not completely irrelevant. But I do get the basic logic that they really are not that important.

Doesn't it come down to a very low % of faceoffs being won cleanly. So if a faceoff guy at 45% and a guy at 55% isn't reallly as big of difference as you would think at first.

And other stats are probably catching the advantages. IE. a Faceoff won leading to a goal would often give the center an assist. Or a faceoff won leading to possession would show up in corsi. So if a guy has the same corsi% but lower faceoff win rate then he's doing something else that is driving corsi and it would basically be double counting the possession gain by looking at the faceoff %.
- sditulli

Did you actually look at the site?

Net Shots Post-Faceoff (NSPF) estimates a face-off taker's contribution to puck possession by counting shot flow (shots-for minus shots-against) during the 10 seconds following a player's face-offs in 5-versus-5 situations, and comparing it to a league-average face-off taker. NSPF is calculated for each of the offensive, defensive, and netural zones; the overall NSPF is their sum.

Crosby had the best NSPF at 76.14. That is less than one shot attempt per game.

NSPF/FO is the NSPF per face-off, allowing for easier comparison among players with different numbers of face-offs taken.

He also led the league per face-off at 0.0554 shot attempts per face-off

They are irrelevant over the course of a season. Now maybe in the playoffs when we are talking about a small sample of 7 games they come into play but even then it is going to be luck based in that small sample.

Did you actually look at the site?

Net Shots Post-Faceoff (NSPF) estimates a face-off taker's contribution to puck possession by counting shot flow (shots-for minus shots-against) during the 10 seconds following a player's face-offs in 5-versus-5 situations, and comparing it to a league-average face-off taker. NSPF is calculated for each of the offensive, defensive, and netural zones; the overall NSPF is their sum.

Crosby had the best NSPF at 76.14. That is less than one shot attempt per game.

NSPF/FO is the NSPF per face-off, allowing for easier comparison among players with different numbers of face-offs taken.

He also led the league per face-off at 0.0554 shot attempts per face-off

They are irrelevant over the course of a season. Now maybe in the playoffs when we are talking about a small sample of 7 games they come into play but even then it is going to be luck based in that small sample.
- Feds91Stammer

I pretty much admit they don't matter much. Small samples don't matter its a fallacy of the eye test of seeing a random event where a goal was scored off a faceoff.

Though I wouldn't say faceoffs are irrelevant. 1 shot attempt per game should translate in 4-5 goals per year. So 1-2 extra wins per year. That could make the difference between playoffs or not. Or home ice or not. Which can be worth to a team a few million dollars.

I pretty much admit they don't matter much. Small samples don't matter its a fallacy of the eye test of seeing a random event where a goal was scored off a faceoff.

Though I wouldn't say faceoffs are irrelevant. 1 shot attempt per game should translate in 4-5 goals per year. So 1-2 extra wins per year. That could make the difference between playoffs or not. Or home ice or not. Which can be worth to a team a few million dollars.
- sditulli

For even-strength faceoffs, all we need to go is integrate the area under the blue curve and multiply by EV shooting percentage, which is 5.97%. This gives us a goal differential of +2.45 goals per 100 extra faceoff wins, or 245 faceoffs per two points in the standings. (Incidentally, a neutral zone faceoff is worth +0.9 goals per 100 extra faceoff wins, or two points in the standings per 657 extra faceoff wins.)

http://www.arcticicehocke...ff-even-strength-vs-power

The only 3 teams with a differential above 245 this season were Arizona, Carolina, and Minnesota.

A player has 4th line possession metrics with a 56 FO%. That means DESPITE being able to begin play with the puck 56% of the time (an obvious advantage) they are still getting wrecked afterward.

Arizona and their shiny new Math GM determined that Vermette wasn't even good enough to hold a roster spot over a developing young depth player on a team of developing young players. Why should the Pens disagree with that?

Because he wins 56% of faceoffs (like Cullen) while posting that HERO chart? Because he had 38 points (16 on the power play!) while getting more ice time and MUCH more favorable zone starts versus Cullen's 32 points ( 0 on the power play!)?

If we aren't signing a player near Cullen's quality for the 4th line, we should continue to allow young players to develop in that 4th line depth role.

You do realize all scientific laws are based on "volatile entities". Scientific principles measure the exact same things as we are measuring here, but they are measuring things with sample sizes larger than I can even pronounce. So they average out into neat scientific laws.

Analytics are just a way of predicting likely results. But yes any given entity is a bit random, but still predictable in a broad sense.
- sditulli

Ah no I don't believe they are.

Just as a simple basic example science produces results that nominates the environment they were produced under - does analytics do that? Simple yes no answer.

Putting up 1.04 P/60 at even strength as a 2C is inexcusable I don't care how bad the team around him was. That's not holding your own in that role.
- Victoro311

Of course he didn't hold is role. He shouldn't have been playing 2C. That's like saying the same thing if you put Sundvqvist as the Pens' 2nd C. You can't possibly expect him to perform as well as he does at a 4th line level.

I'm not even trying to make a case for signing him anyway. He holds no value to the Pens.

Dominic Moore's name has now been thrown around and I'd jump at that in a second. Pens shouldn't sit around waiting on Cullen. Again though, I don't necessarily think the Pens need to add a 4th C.