New to Advanced NHL Stats

[Juxtaposer]

I'm not sure I agree completely with this definition of PDO. I know it's the official definition and all but still.

The average PDO (weighted for #shots) for the whole league will be exactly 1000, that's obvious. But a good team could be consistently over 1000 and a bad team could be consistently under. By definition, it doesn't have to regress to 1000 for one specific team. Or it's not entirely due to luck.

A very skilled team can finish a season with a PDO of 1020 or 1030. It's not luck anymore, IMO. Take an NHL team and send it to the AHL and I would bet my house that its PDO won't 'regress towards 1000'.

What could be considered 'lucky' is to have a very high PDO (like 1050 or 1100) over a stretch because of hot goaltending or unusually high shooting%. But that's it.

Here's the thing with PDO: It's a great concept and the idea behind it makes a ton of sense: Hockey is a game played on a not-entirely-smooth ice surface with ten guys skating around at high speed. Where exactly a bouncing disk of rubber goes is sometimes decided by luck. A bad bounce, etc.

The problem with PDO is that it generally regresses towards a mean of 1000 (on-ice save percentage of ~920 (.920%) + on-ice shooting percentage of ~80 (8.0%) at even strength, but not really. Yes, that's a contradiction, but let me explain my thinking.

There have been a lot of recent studies that suggest two interesting things: Defensemen have a tiny bit of control over their on-ice save percentage and none over their on-ice shooting percentage. Likewise, forwards have a tiny bit of control over their on-ice shooting percentage but none over their on-ice save percentage. Likewise a very poor defensive defenseman would allow a greater amount of quality of shots on his goaltender and therefore earn a below-average on-ice shooting percentage.

So I believe that a defenseman who is very good defensively at blocking/diverting/directing elsewhere/keeping elsewhere shots can have a little effect on the on-ice save percentage of their goaltender. Likewise, I think that a very creative, talented offensive forward can influence to a small degree the quality (location/speed/type/etc) of shot they direct at the net. 4th line face-punchers, on the other hand, rarely take shots that are not from 30 feet out with no screen.

However, I think the range of these things is very small (see #1 at bottom of post). Example: Here are Sidney Crosby's on-ice shooting percentages from as long as the stat has been measured. See how large the range is? But here's another subtle thing to notice; as the sample size gets larger, the on-ice shooting percentage gets lower. The season he played just 22 games, he had a huge 15.17% shooting. The two seasons he played close to the full 82 games (in this sample, which is just the seasons in which BTN tracked stats) he had a 11.02 and 11.54 on-ice shooting. I think if the rest of his injury-riddled seasons had been completed, you would have seen his on-ice shooting regress to a more reasonable 11%. And Crosby is the very top of the food chain. I think he sets the upper limit. Now, just browsing through BTN, by eye I would guess the lower limit for the 5 minute per game 4th line facepunchers is about 6%, with more players at this end than at the Crosby end (there are endlessly more 4th line facepunchers than Crosby's in the league). I think that if you ignore the facepunching 4th line grinders and the elite talents like Crosby* (the former because who cares about the offensive talent of 4th line facepunchers and the latter because players like Crosby are outliers in the NHL), you'd get a true talent range of between about 7.5% and 9.0% for forwards (and I feel I may actually be being generous here. See #2 for a good read about the sustainability of on-ice shooting percentages). Now considering that there is a good amount of data suggesting that forwards have little to no effect on their own on-ice save percentage, that puts true talent PDOs for individual non-facepuncher non-Crosby NHL forwards at a range of 995 and 1015, assuming league average goaltender. That's a very small range. And that's why PDO is so safe, because even if you're accounting for differences in true talent on-ice shooting percentage, the rest is so very uncontrollable for that forward and is thereby susceptible to fluctuations in luck.

Now compare the range of likely true talent to the range of on-ice shooting percentages you see in individual seasons. That's where the luck plays in.

If I were so inclined, I would look at the range of true talent on-ice save percentages for defenseman, but for simplicity's sake, I'll just say it's ±.10% at the extremely of defenseman's capabilities. I think that's generous, I think a truer estimate would be closer to ±.05%. Of course, the thing about on-ice save percentage is that it is largely based on the true talent of that skater's team's goaltender. But on average, the league's goaltender's put up a .920, so on average the true talent PDO of a defenseman would be 995-1005. Again, this is a much smaller range than what we often see on a season-by-season range (see #3 for a good read on the sustainability of defensemen's on-ice save percentages). Plus, if you look at the on-ice save percentages of defensemen in any given year, the names look positively random (see #4). So I think it's a safe assumption that forwards can affect shot quality for more than defensemen can affect shot quality against.

But regardless, here's what you should take away from this:

1) In a basic investigation, assuming a true talent PDO of 1000 is safe, no one should freak out about it. Anyone who cries about a 1010 PDO being hugely inflated clearly doesn't know what they're talking about and probably shouldn't be trusted in this sort of analysis.

2) a) When evaluating a forward on an individual basis, look for a true talent PDO comprised of the career even strength save percentage of the team's goaltender plus the forward's career average on-ice shooting percentage, and then look for season-to-season deviations to determine how lucky/unlucky that player's season/stretch of games/whatever sample size you're looking at has been.

b) When evaluating a defenseman on an individual basis, look for a true talent PDO comprised of the league average shooting percentage (8%, which includes both the elite talents like Crosby and 4th line facepunchers, as defensemen face all players, although I'd love to see eventually a stat which averages the career on-ice shooting percentages of players a defenseman faces weighted by ice-time against, but that would be a huge project) plus the career average of the defenseman's goaltender ± the established effect that defenseman has had on his goaltenders over his career, and compare that to the season/sample size you're looking for the luck factor over.

3) The funny thing about PDO is that very often a high on-ice shooting percentage is coupled with a low on-ice save percentage, resulting in a PDO of, wait for it, 1000. Why is that? Is it because high-octane offensive forwards with true talent on-ice shooting percentages are likely to be paired with great offensive but poor defensive defensemen who cause a true talent deflation of their goaltender's save percentage behind them? Or is it a system-based deal, where run-and-gun systems cause higher quality scoring chances at both ends and conservative defense-first systems cause lower quality shots at both ends of the ice? I would bet more on the former, but it'll take someone more versed in systems and analytics than I to determine that.

4) LUCK IS AN ACTUAL PHENOMENON. There is a degree of shot quality on an individual basis but it is not as large as what people who scream about shot quality would have you think. A PDO of 970 or 1030 (see #5) (like the poster I'm quoting alludes to) WOULD be in fact very lucky, even if that team had very good goaltending and a high proportion of high true-talent shooters.

So while I see the flaws in PDO and the ways it could be improved, I still think it's a very important concept that should not be disregarded.


*Crosby is one example, other forwards that appear to be outside the general range are Ovechkin, Stamkos, Malkin, Semin, Getzlaf.

#1: http://www.broadstreethockey.com/2011/12/9/2623793/do-playmakers-drive-teammates-shooting-percentage

#2: http://www.boysonthebus.com/2013/07/18/can-players-sustain-on-ice-shooting-percentages/

#3: http://hockeyanalysis.com/2012/02/09/defenders-effect-on-save/

#4: https://twitter.com/camcharron/status/357617826808336386

#5: http://nhlnumbers.com/2013/4/22/pdo-numbers-by-nhl-team-april-22

 

[No Ragretzky]

Just browing behindthenet and I noticed this past season Jamie Benn's on-ice SV% was 881 whereas the rest of his team had a minimum 905 SV%. Was he THAT unlucky? Or was he that much of a liability? Or a bit of both?

 

[sharks9]

Just browing behindthenet and I noticed this past season Jamie Benn's on-ice SV% was 881 whereas the rest of his team had a minimum 905 SV%. Was he THAT unlucky? Or was he that much of a liability? Or a bit of both?

Last year he was pretty low on the team too, but not as bad as this year. It's a combination of being bad defensively and bad luck that would give such a terrible on-ice SV%

 

[Tak7]

There's definitely a role for advanced stats in today's NHL. However, I also feel as though lots of people put way too much emphasis on them when it comes to analysing players and performance.

What bothers me about advanced stats, is fans who understand them and interpret them acting like they are better, more superior fans than anyone else.

I've lost track of the number of times I've offered an opinion on a player here, or on twitter, and have had someone throw back some convoluted advanced stat concoction in my face in a real matter-of-fact way

 

[Bear of Bad News]

What bothers me about advanced stats, is fans who understand them and interpret them acting like they are better, more superior fans than anyone else.

In fairness, those types of fans also exist on the other side of the argument ("the only way to understand players is to watch them, and anyone who tries to use numbers is wasting their time").

 

[Kane One]

I'm also new to this, but I'm interested in how behindthenet.ca calculates these stats. Are there other sites I could get this from aside from them? I'm honestly not sure if I want to trust any stat I can only get from one place.

 

[hatterson]

I'm also new to this, but I'm interested in how behindthenet.ca calculates these stats. Are there other sites I could get this from aside from them? I'm honestly not sure if I want to trust any stat I can only get from one place.

stats.hockeyanalysis.com has many (if not all) of the same stats.

 

[Kane One]

Why is Corsi shown in per 60 minutes?

 

[Ishdul]

Why is Corsi shown in per 60 minutes?

That's just how behindthenet does it. In a lot of sports rate stats are expressed by the length of the game since there's not always a standard. stats.hockeyanalysis.com has it by 20 minutes.

 

[Kane One]

That's just how behindthenet does it. In a lot of sports rate stats are expressed by the length of the game since there's not always a standard. stats.hockeyanalysis.com has it by 20 minutes.

To me, 20 minutes makes more sense because players play closer to 20 minutes a game than they do to 60.

 

[The Winter Soldier]

I am new to advanced NHL stats and a bit overwhelmed.

Can of you recommend the best websites that give a comprehensive overview and the top oh three or five stats?
I saw the first thread but it was overwhelming.

Advanced stats should never be a replacement for watching the game, for instance can someone prove puck possession is a positive as to knowing how to play without the puck is? Is there a stat out there that measures a player's intelligence on how to intercept passes, playing positionally, all attributes to playing without the puck.

Analytics has a place in the game, a small part, but they should be taken with a grain of salt.

There are some bloggers that have taken them to a whole new extreme, one's that love to use catchy circles and graphs.

Makes me ill when I see them, hockey is not about reading boxscores or how many shots you shoot at the net and even missing it to accumulate a high corsi stat.

People should remember this. Crosby once finished 10th or thereabouts on Pittsburg in Rel Corsi, nothing replaces watching a player play.

 

[Bear of Bad News]

Advanced stats should never be a replacement for watching the game, for instance can someone prove puck possession is a positive as to knowing how to play without the puck is? Is there a stat out there that measures a player's intelligence on how to intercept passes, playing positionally, all attributes to playing without the puck.

Analytics has a place in the game, a small part, but they should be taken with a grain of salt.

There are some bloggers that have taken them to a whole new extreme, one's that love to use catchy circles and graphs.

Makes me ill when I see them, hockey is not about reading boxscores or how many shots you shoot at the net and even missing it to accumulate a high corsi stat.

People should remember this. Crosby once finished 10th or thereabouts on Pittsburg in Rel Corsi, nothing replaces watching a player play.

That's all well and good. However, there are also biases involved with "watching the game", and analytics (done properly) can insulate against those effects.

If you haven't read Daniel Kahneman's "Thinking, Fast and Slow", I highly recommend it. It shows how your brain can (and does) trick you into making sub-optimal decisions.

One best learns about the game with both (watching and analyzing). Anyone who offers only one or the other either has a vested interest, or is missing half of the equation. For what it's worth, most people who work with hockey analytics do so because they love watching the game - they aren't substituting one for the other.

Your post above suggests that you don't fully understand how analytics are properly used (perhaps you've only read the "pop" stuff that's out there). I'd suggest reading up around this subforum - particularly at the links and resources. You might enjoy yourself.

 

[GuineaPig]

People should remember this. Crosby once finished 10th or thereabouts on Pittsburg in Rel Corsi, nothing replaces watching a player play.

Show me somewhere, anywhere, where someone has concluded Crosby isn't all that based on this.

You couldn't, because no one reasonable just takes one stat and ignores all others.

 

[Stephen]

Advanced stats are Pejorative Slured. What they should really do is just put some transponders on the players like they do in soccer to actually track puck possession and the location of their possession and not try to find mathematical ways to approximate what can be quantified. What is the point in trying to quantifying something by measuring something else, then trying to infer the meaning out of a mountain of garbage?

 

[The Winter Soldier]

That's all well and good. However, there are also biases involved with "watching the game", and analytics (done properly) can insulate against those effects.

If you haven't read Daniel Kahneman's "Thinking, Fast and Slow", I highly recommend it. It shows how your brain can (and does) trick you into making sub-optimal decisions.

One best learns about the game with both (watching and analyzing). Anyone who offers only one or the other either has a vested interest, or is missing half of the equation. For what it's worth, most people who work with hockey analytics do so because they love watching the game - they aren't substituting one for the other.

Your post above suggests that you don't fully understand how analytics are properly used (perhaps you've only read the "pop" stuff that's out there). I'd suggest reading up around this subforum - particularly at the links and resources. You might enjoy yourself.

My objection with advanced stats is I often see bloggers use them to try to justify bad performances or use them to mislead for argument sake. If you look hard enough, I am sure you can use a selected stat to make a player look better than what they are.

Defensive zone starts I have read a lot was the reason for one player's bad year for instance, a point reiterated so much, it seemed like this was the only reason for this player's off year. Such nonsense I thought to myself, that some would buy into the narrative.

Even less advanced stats such as face-off % can be misleading. Usually the best face-off men in the league face each other, a face off % such as 52.6% for a player like Bozak is misleading for instance when you compare him to Mcclement's 51.6% or Grabovski's 50.6% since he faces the other team's best face off men nightly, whereas Mcclement and Grabovoski face the 2nd or 3rd best on the opposing teams.

Saying this, I would agree there is a place for advanced stats in the game, just as plus minus stats are. Yet I hear people scoff at them.

Context is the key for me, watching games is still the best way to gauge the effectiveness of players in a game, nothing replaces scouts, no metric or analytic will. One of my main beefs is the over emphasis placed on possession stats. When playing without the puck is just as important. But if you read enough the analytics will almost always point to possession stats for a player being superior, this is faulty thinking IMO. A lot more to hockey than possessing the puck. Vision, Ice Q, grit, heart, playing without the puck, net presence, knowing when to shoot, when to pass. ect...

I would weight it this way:

Viewing
Observation
Stats(advanced and old fashioned stats)
Conclusion

My belief is hockey will never be a stats game like baseball, too many intangibles to have it down to a science.

 

[Ishdul]

Advanced stats are Pejorative Slured. What they should really do is just put some transponders on the players like they do in soccer to actually track puck possession and the location of their possession and not try to find mathematical ways to approximate what can be quantified. What is the point in trying to quantifying something by measuring something else, then trying to infer the meaning out of a mountain of garbage?

The point is to measure meaningful possession, notably in terms of shots attempted. Possession stats would not be improved by counting the amount of time a player holds the puck behind his net waiting for his teammates to change.

 

[The Winter Soldier]

The point is to measure meaningful possession, notably in terms of shots attempted. Possession stats would not be improved by counting the amount of time a player holds the puck behind his net waiting for his teammates to change.

Zone time is more important to possession time. Playing without the puck is just as important if not more.

 

[The Winter Soldier]

Show me somewhere, anywhere, where someone has concluded Crosby isn't all that based on this.

You couldn't, because no one reasonable just takes one stat and ignores all others.

I don't understand what you are saying, are you saying Alex Ponikarovsky is a better player than Crosby was in the 09-10 season because he had a higher Relative and On ice Corsi?

 

[do0glas]

My objection with advanced stats is I often see bloggers use them to try to justify bad performances or use them to mislead for argument sake. If you look hard enough, I am sure you can use a selected stat to make a player look better than what they are.

Defensive zone starts I have read a lot was the reason for one player's bad year for instance, a point reiterated so much, it seemed like this was the only reason for this player's off year. Such nonsense I thought to myself, that some would buy into the narrative.

Even less advanced stats such as face-off % can be misleading. Usually the best face-off men in the league face each other, a face off % such as 52.6% for a player like Bozak is misleading for instance when you compare him to Mcclement's 51.6% or Grabovski's 50.6% since he faces the other team's best face off men nightly, whereas Mcclement and Grabovoski face the 2nd or 3rd best on the opposing teams.

Saying this, I would agree there is a place for advanced stats in the game, just as plus minus stats are. Yet I hear people scoff at them.

Context is the key for me, watching games is still the best way to gauge the effectiveness of players in a game, nothing replaces scouts, no metric or analytic will. One of my main beefs is the over emphasis placed on possession stats. When playing without the puck is just as important. But if you read enough the analytics will almost always point to possession stats for a player being superior, this is faulty thinking IMO. A lot more to hockey than possessing the puck. Vision, Ice Q, grit, heart, playing without the puck, net presence, knowing when to shoot, when to pass. ect...

I would weight it this way:

Viewing
Observation
Stats(advanced and old fashioned stats)
Conclusion

My belief is hockey will never be a stats game like baseball, too many intangibles to have it down to a science.

as far as faceoffs go, it has been argued that they arent as important as it seems.

im on the fence about it personally. I think anything over 40% starts in the defensive will have an affect on your offensive game depending on your total TOI. (4th liner vs. top 6).

There have also been studies showing that winning a faceoff can dramatically improve your shots on goal/chances of scoring...so winning more than the other team will probably help...also a good shoutout to why your best faceoff guys are typically either an offensive specialist or a defensive specialist. obviously, you have your two way guys that can do it all as well...but whatever.

the possession/shot studies show over a considerable amount of sample size that out-possessing your opponents will generally lead to more wins than losses.

when you boil that down to one game, you run into sample size variance. (yuo can out-shoot and lose or get out shot and win) but over the course of a season if you consistently are shooting more, you will likely have a winning record. from the studies i have read, this has not only been shown to be true, but also predictive to an extent.

you are attributing an avoidance of cherry picking stats on what you see from HF boards users. read the studies, decide for yourself. but just because you believe that watching is the be all end all, doesnt mean that stats are wrong.

 

[Ishdul]

My objection with advanced stats is I often see bloggers use them to try to justify bad performances or use them to mislead for argument sake. If you look hard enough, I am sure you can use a selected stat to make a player look better than what they are.

Defensive zone starts I have read a lot was the reason for one player's bad year for instance, a point reiterated so much, it seemed like this was the only reason for this player's off year. Such nonsense I thought to myself, that some would buy into the narrative.

It is an indisputable fact that starting in the defensive zone more than the offensive zone is going to hurt your stats (both traditional and advanced). Different players are hurt differently, but that much is law, and it is futile to argue against it. For this reason it is important to note the zone starts for all players. A really good defensive player might be one who takes a lot of defensive zones and doesn't get scored on at a high rate, and might be properly used as a high defensive zone starts guy, but such a player might have bad traditional stats overall. It can also show misuse of certain players who are better offensively than defensively, who just don't have the same effectiveness but are still forced into such roles by either bad coaching (who misjudges the strengths of the player) or a bad team (who doesn't have any other options).

So yes, it can directly contribute to a player's "bad year", and it certainly hurt Grabovski's stats in comparison with say his 2010-11.

Even less advanced stats such as face-off % can be misleading. Usually the best face-off men in the league face each other, a face off % such as 52.6% for a player like Bozak is misleading for instance when you compare him to Mcclement's 51.6% or Grabovski's 50.6% since he faces the other team's best face off men nightly, whereas Mcclement and Grabovoski face the 2nd or 3rd best on the opposing teams.

This seems like a non sequitur in relation to the rest of your post. Faceoff% is not at all an advanced stat and most of the advanced stats suggest that it is largely unimportant except at the extreme ends of the spectrum. Your point about Bozak also doesn't hold true.

Saying this, I would agree there is a place for advanced stats in the game, just as plus minus stats are. Yet I hear people scoff at them.

The more raw you get with +/- stats, the worse they are, which is important because the commonly listed +/- is incredibly raw. When you get into stuff like on-ice/off-ice you're getting into advanced stat territory.

Context is the key for me, watching games is still the best way to gauge the effectiveness of players in a game, nothing replaces scouts, no metric or analytic will. One of my main beefs is the over emphasis placed on possession stats. When playing without the puck is just as important.

I don't think many people dispute this. Even in baseball this is true, prospects for example are judged on far more than just their stats and "counting up the WARs" analysis is derided by every reputable person.

 

[Bear of Bad News]

I don't understand what you are saying, are you saying Alex Ponikarovsky is a better player than Crosby was in the 09-10 season because he had a higher Relative and On ice Corsi?

Why in the world would you say that that's what he's saying? Nothing he said implied that; in fact, what he said implied the exact opposite of what you're claiming that he implied.

 

[Bear of Bad News]

Advanced stats are Pejorative Slured. What they should really do is just put some transponders on the players like they do in soccer to actually track puck possession and the location of their possession and not try to find mathematical ways to approximate what can be quantified. What is the point in trying to quantifying something by measuring something else, then trying to infer the meaning out of a mountain of garbage?

Way to keep an open mind, Stephen.

Meaningful analysis can be done (and has been done) - if you want to leave your head in the sand, then by all means - leave your head in the sand. And go ahead and call the work that we do "Pejorative Slured" again - that should help.

And before you call this entire subforum "Pejorative Slured" again, I'll ask you to read the forum-specific rules:

http://hfboards.mandatory.com/showthread.php?t=1237235

Pay particular attention to the first one, and thanks for stopping by.

 

[Bear of Bad News]

My objection with advanced stats is I often see bloggers use them to try to justify bad performances or use them to mislead for argument sake. If you look hard enough, I am sure you can use a selected stat to make a player look better than what they are.

What you're missing is this - there are people out there (like myself) who are doing analysis to *learn* about the game, not to promote an agenda or "make a player look better".

I couldn't care less who "looks good" when I do analyses - my goal is to learn more about the sport and what leads to success.

Yes, there are people who use analytics to promote agendas. There are also folks who use "I watched the games, and you can't tell me otherwise!" to promote agendas. Biased fans isn't a problem unique to analytics.

I'll say it again - you seem to have preconceived notions about what goes on here. I'll invite you to read the folks' work who are driving the train - I recommend starting with Puck Prospectus and going from there.

 

[Hank Chinaski]

Advanced stats are Pejorative Slured. What they should really do is just put some transponders on the players like they do in soccer to actually track puck possession and the location of their possession and not try to find mathematical ways to approximate what can be quantified. What is the point in trying to quantifying something by measuring something else, then trying to infer the meaning out of a mountain of garbage?

Perhaps because "advanced stats" (assume your referring to Corsi in this case, which isn't all that advanced, it's just +/- of shots for vs. shots against) have already been proven to be an accurate proxy for puck possession? Why would anyone spend money on the system you describe when it's essentially redundant.

BTW, the San Jose Sharks, Phoenix Coyotes, Vancouver Canucks and I'm sure a number of other NHL teams found that "mountain of garbage" meaningful enough to track. Guess they're Pejorative Slured as well.

 

[Kershaw]

generally the people that say the advanced stats are useless have no idea what they mean, which always brings in a loop of argument between "stat nerds" and self proclaimed experts who'd rather "watch the game". since a lot of people in general don't understand it, it becomes mob mentality on said set of statistics. It's the same thing in baseball with advanced stats like UZR, FIP, SIERA, etc.