Advanced Stats

[The Mars Volchenkov]

Just wanted to start a thread on this because, even though a good amount of teams seem to have been using this for a while, it's really taken off in the last year or so and become huge in the NHL community. What is everyone's opinion on this stuff?

I'm learning more about it every day, and I don't know if I'm just stubborn or what, but I'm having trouble really accepting it and taking it too seriously, at least as far as judging players based only on advanced stats. Certain things, I do like and think are helpful, such as zone starts and zone finish, as you see how a player is sheltered and what players are moving the play forward, but a lot of the other stuff must be going over my head.

Just like other things, it seems like there are going to be flaws. Yes, a lot of shots against usually aren't a good thing, but are they giving up a lot of shots because they're leading most games and sitting back, and do they even take into account the quality of these shots?

When I'm at the games or watching on TV, I tend to notice myself who is being given the tougher matchups and whether they're doing well in the position the coach has put them in, so maybe that's why I tend to ignore these advanced stats. I'm a pretty astute observer when I watch the games, so I don't need some stats to tell me whose playing well and who isn't.

Am I in the minority or is this stuff just going to keep getting bigger and I better get on the train?

Click to read more...

 

[Freudian]

If you use it carefully and in combination with watching the games, it's helpful. I love looking at zone starts. It doesn't really tell us anything except how coaches use their players, which is very interesting.

I don't think corsi/fenwick/shots have as big explanatory power as most advanced stat fans, especially comparing players on different teams with different systems. But if you use it carefully in conjunction with watching games and with zone starts/quality of opposition you can get a decent idea about who wins the possession battle and who doesn't.

Hockey is such a chaotic game and the stats tracked aren't very useful for describing what happens. Some here on HF (none on this sub-forum) seem to base 100% of their views on advanced stats, which shows they don't have a basic grasp of how weak the explanatory power of them are and how small samples we have during a season.

They assume every team that's winning while outshot is lucky and will eventually start to lose, which is absurd. And when teams don't start to lose while being outshot like Preds, Rangers, Leafs it's explained as the teams being lucky. Corsi/Fenwick isn't that strongly correlated to wins. It's slightly stronger than points or goals for are, but not significantly stronger.

So I guess I find the advanced stats interesting to browse and sometimes use but I realize all of them have big flaws. I'd never speak in definitive terms about any player or team based on it. Sometimes you notice something interesting there (like for example that when Barrie is on the ice Avs tend to outshoot the opposition) and then you can watch for it during games. If used cautiously it can increase your enjoyment of hockey. It's not the Rosetta Stone that has cracked the code of hockey, like some seem to think it is.

On a side note. I think it was on MvsW they mentioned that Roy is the first coach in the NHL that only uses man on man in the defensive zone. Other coaches use zone defense or a zone/man on man marking system. I wonder what that results in as far as allowing shots and opposition creating scoring chances. I have noticed fewer unmarked opponents in the slot, but it's only six games so it might be random.

 

[Avs_19]

Like you guys, zone starts/finishes is the only thing I look at but that's because I haven't really gotten into the other stuff yet and I don't know enough about it. I've read that some advanced stats for hockey are flawed and you never hear broadcasters or analysts use it on air. Advanced stats are getting used more and more in baseball and if you're a fan, I think it's something you should probably know so I read up on it. However, it doesn't seem to be catching on for hockey. Unless I'm just looking in the wrong places and reading the wrong people.

What site do you guys use, behindthenet.ca?

 

[zxcvnm]

How do they actually measure zone starts/finishes? For example, are zone starts only based on when you start a shift from a face-off? What if a defenseman is holding the puck behind the net while a line change is taking place or is this situation ignored in current methodology?

Zone finishes seem flawed since I'm guessing it cannot differentiate between cases when a line plays the entire time in the defensive zone only to get it out and into the offensive zone at the end of a shift and when a line immediately takes possession and moves the puck into the offensive zone for an extended period of time. It would seem that zone possession time could be used to offer a better measuring stick. That said, I imagine corsi can be adjusted to account for this as well.

Another question I have is if anyone knows why corsi is preferred to zone possession time?

 

[Freudian]

http://www.extraskater.com/ has stats game by game and is updated almost in real time.

I like http://www.leftwinglock.com/line-combinations/ to see who plays with who. That's not really stats, but useful info.

 

[Freudian]

How do they actually measure zone starts/finishes? For example, are zone starts only based on when you start a shift from a face-off? What if a defenseman is holding the puck behind the net while a line change is taking place or is this situation ignored in current methodology?

Zone finishes seem flawed since I'm guessing it cannot differentiate between cases when a line plays the entire time in the defensive zone only to get it out and into the offensive zone at the end of a shift and when a line immediately takes possession and moves the puck into the offensive zone for an extended period of time. It would seem that zone possession time could be used to offer a better measuring stick.

They don't track who has possession of the puck and where they have it. If they did, it would obviously be more useful.

Zone starts/finish isn't perfect but since they are events that happen so often during the course of a season all these what if things tend to even out. The odd shift where you spend most of the time in the offensive zone and then end up in your own zone will be drowned out by all the times where you end up in the offensive zone at the end.

 

[zxcvnm]

They don't track who has possession of the puck and where they have it. If they did, it would obviously be more useful.

Zone starts/finish isn't perfect but since they are events that happen so often during the course of a season all these what if things tend to even out. The odd shift where you spend most of the time in the offensive zone and then end up in your own zone will be drowned out by all the times where you end up in the offensive zone at the end.

So are zone starts irrespective of whether or not the shift started with a face-off? Part of my confusion is that it seems situations in which your defenseman possess the puck in the defensive zone while a line change is taking place is dramatically different than a defensive zone draw, but would they be counted the same in zone starts?

As for your statement in the second paragraph, I'm skeptical. It seems in hockey there are plenty of shifts where you get pinched in your defensive zone for the majority of your shift, but when you finally clear it, you possess it into the offensive zone, wait for your line to change, and pass it off or dump it down low before going for a change yourself. I don't think this is a sufficiently uncommon situation. Just briefly perusing zone start/finish stats seem to indicate that many players with majority defensive zone starts have a lower percentage of defensive zone finishes and vice versa. This is certainly biased sampling since players employed in majority defensive zone situations, for example, tend to be better than average defensive players, but I'm still skeptical of the usefulness of the zone finishes stat in most cases (zone starts, on the other hand, do seem like a good indicator of a sheltering/non-sheltering role).

EDIT: I would figure that corsi adjusted for quality of competition and game situation (e.g. score) would be a better measuring stick. Is there any statistic that adjusts corsi for competition and game situation along with zone start?

 

[Freudian]

Like you guys, zone starts/finishes is the only thing I look at but that's because I haven't really gotten into the other stuff yet and I don't know enough about it. I've read that some advanced stats for hockey are flawed and you never hear broadcasters or analysts use it on air. Advanced stats are getting used more and more in baseball and if you're a fan, I think it's something you should probably know so I read up on it. However, it doesn't seem to be catching on for hockey. Unless I'm just looking in the wrong places and reading the wrong people.

What site do you guys use, behindthenet.ca?

Baseball is the perfect sport for stats. It's short controlled events with a finite amount of outcomes. You can track it perfectly and it's very easy to see how it affects the outcome of a game.

Games that are as fluid as hockey and where there are so many things affecting the outcome of a game is much less suited for it. Advanced hockey stats basically ignores goaltending. It's basically assumed they are all equally good in the long run, simply because they have no way to incorporate it into the formulas without them collapsing completely. They tend to ignore special teams for the most part. Hockey advanced stats in a nutshell is basically the more shots you take the more you win.

 

[Freudian]

So are zone starts irrespective of whether or not the shift started with a face-off? Part of my confusion is that it seems situations in which your defenseman possess the puck in the defensive zone while a line change is taking place is dramatically different than a defensive zone draw, but would they be counted the same in zone starts?

As for your statement in the second paragraph, I'm skeptical. It seems in hockey there are plenty of shifts where you get pinched in your defensive zone for the majority of your shift, but when you finally clear it, you possess it into the offensive zone, wait for your line to change, and pass it off or dump it down low before going for a change yourself. I don't think this is a sufficiently uncommon situation. Just briefly perusing zone start/finish stats seem to indicate that many players with majority defensive zone starts have a lower percentage of defensive zone finishes and vice versa. This is certainly biased sampling since players employed in majority defensive zone situations, for example, tend to be better than average defensive players, but I'm still skeptical of the usefulness of the zone finishes stat in most cases (zone starts, on the other hand, do seem like a good indicator of a sheltering/non-sheltering role).

No doubt zone finishes are much less interesting than zone starts. I tend to only take note of it when there is a big difference between zone starts and zone finishes. For most players the difference is so small it doesn't tell us much.

For example last year Bordeleau started in the Ozone 61% and finished there 53%. It tells us basically what the eye test tells us. He's being sheltered and when he's on the ice, Avs tend to lose the possession battle a bit. Not massively but noticeably.

 

[zxcvnm]

Baseball is the perfect sport for stats. It's short controlled events with a finite amount of outcomes. You can track it perfectly and it's very easy to see how it affects the outcome of a game.

Games that are as fluid as hockey and where there are so many things affecting the outcome of a game is much less suited for it. Advanced hockey stats basically ignores goaltending. It's basically assumed they are all equally good in the long run, simply because they have no way to incorporate it into the formulas without them collapsing completely. They tend to ignore special teams for the most part. Hockey advanced stats in a nutshell is basically the more shots you take the more you win.

Baseball is better because there are a lot more individual elements that can be measured. Basically, most of the game is pitcher vs batter. For the most part, teammates don't play a huge role in an individual's performance. This is in stark contrast with hockey, but it simply means that old-school player evaluation will always be necessary.

That advanced stats ignores goaltending is a good thing. Good goaltending can mask deficiencies in team play and separate stats can be used to evaluate a specific goaltender, but you don't want to evaluate team performance on the basis of an opponent's goalie standing on his head. On average you will win more games than not if you outperform the other team regardless of their goaltender (unless yours is particularly weak but again it would simply illustrate to the team the need to improve the goaltending rather than other areas).

From what I've read, corsi/fenwick correlates pretty well with scoring chances (e.g. here) and I'm guess scoring chances correlates, over the course of a season, pretty well with goals. You said in an earlier post that corsi doesn't correlate to wins pretty well. Do you have justification (like an article link, for example)?

 

[Boulder Avalanche]

Advanced stats are not useful by themselves but with watching games they become useful. Until someone creates a complex computer simulation that tests advanced stats and their effect on hockey they are useless by themselves.

 

[Freudian]

Baseball is better because there are a lot more individual elements that can be measured. Basically, most of the game is pitcher vs batter. For the most part, teammates don't play a huge role in an individual's performance. This is in stark contrast with hockey, but it simply means that old-school player evaluation will always be necessary.

That advanced stats ignores goaltending is a good thing. Good goaltending can mask deficiencies in team play and separate stats can be used to evaluate a specific goaltender, but you don't want to evaluate team performance on the basis of an opponent's goalie standing on his head. On average you will win more games than not if you outperform the other team regardless of their goaltender (unless yours is particularly weak but again it would simply illustrate to the team the need to improve the goaltending rather than other areas).

From what I've read, corsi/fenwick correlates pretty well with scoring chances (e.g. here) and I'm guess scoring chances correlates, over the course of a season, pretty well with goals. You said in an earlier post that corsi doesn't correlate to wins pretty well. Do you have justification (like an article link, for example)?

There are tonnes of articles about it. It's been a while but I think the correlation between corsi tied and future results was something like 0.45 and since you don't get critical mass until later in a season (corsi and fenwick tend to fluctuate a lot early in the season) the predictive value of these stats is very low. Add that with injuries/trades the team you put on the ice rarely is the team that created the stats to start with. Add goaltending into the mix and it's not hard to see the absurdity in claiming that a team will inevitably start losing/winning because their fenwick/corsi is at a certain level.

I don't mind these stats in themselves. I just dislike how they are used by people.

 

[tigervixxxen]

I'm glad to see this discussion because I am interested in some advanced stats analysis but it can be overwhelming. The numbers gurus seem to throw all of the data out there and analyze everything. It would be more helpful to have a few of these measures become mainstream understood and accepted rather than constantly trying to stare at a table with 26 columns and try to make sense of it. I like that the advanced measures are trying to expand on shots/saves and +/- but it is definitely not perfect either.

Corsi and Fenwick are interesting from a puck possession standpoint. The theory there is you can't shoot the puck if you don't posess it but doesn't take into account possessing without shooting. I think it's more interesting comparing players on the same team rather than from team to team because pretty much the better teams have higher measures. Quality of competition Corsi is interesting too just to see who is facing the toughest competition. We pretty much know that anyway but always good to see if it matches perception.

PDO is another one, the luck factor apparently. I guess this one accounts for a team having either an unusually high shooting or save percentage, and then also useful in comparing a player's shooting percentage against the mean. It's probably way too early to glean anything from PDO as the team's save % has everyone's PDO high.

I like the Behind the Net site because it's easier to follow but them seem slower to update. Here is one more link but it's another big table. http://stats.hockeyanalysis.com/

 

[zxcvnm]

@Freudian: I wouldn't use them for predictive value. For player evaluation, it seems that it provides some important context. I would think that advanced stats, in general, have more value for player evaluation than team evaluation.

On the team front, while corsi might not correlate too well to future results (why should it), I wonder how well it correlates to current team performance. For example, if you track corsi over the course of the season, would it correlate to the team's performance during that same time period. In other words, as you said, injuries/trades can impact team performance, but I imagine that impact would likely be reflected in the corsi as well. So, while it wouldn't have predictive value, it would have important value in evaluating a team's performance during a particular game (not that this is too critical).

EDIT: In other words, past corsi will only correlate so well with future corsi for many reasons. But, it would mean that teams that underperform corsi-wise will not continue winning unless they correct that deficiency. As it relates to the Avs, it would mean that our struggles in this area (I'm guessing as I haven't looked at those stats closely), indicate that unless we fix it, we will come back to down to Earth. It doesn't mean that will necessarily happen. Simply that, if we continue our winning ways, we improved our corsi along the way.

 

[Freudian]

I think the correlation between the strongest stat (fenwick tied, I think) and winning was about 0.55. Nothing worth bothering with, in my book.

Personally I think corsi on an individual level is much weaker than on the team level. Put a Winnik with Landeskog, who shoots every chance he gets, and Winnik's suddenly a corsi god. Trade him and his corsi plummets because he doesn't have a guy with 300 shots on his line anymore.

Anyway, I'm not an expert and it's not like I follow the advanced stats debate closely. I just check in from time to time to see if they have come up with something more interesting than corsi/fenwick. I guess PDO is fun, since we now can put a number to what bad shooting luck we have and what hot goaltenders we run into. Science for the win.

 

[tigervixxxen]

Another one I like is the Player Usage charts, which combine zone starts with Corsi. It gives an idea of the role each player is in (offensive or defensive starts, easier or more difficult competition) and then how effective they are at possessing the puck in that role. Again, probably more interesting to compare players on the same team but is also interesting to see how other teams' set up differs.

http://www.hockeyabstract.com/playerusagecharts

 

[InjuredChoker]

here is steve burtch' article how reliable ton of statistics are, how they correlate with winning in given season or year to year.

 

[Lonewolfe2015]

Advanced stats require a lot of supplementary knowledge and you need to know which ones to use. Within a team they are extremely telling, from team to team they are less telling, from player to player they are even less so. You need big differences and multiple stats to get a decent picture.

Take Cody Franson for instance, last year he was extremely sheltered with some of the lowest QoC relative to his team relative to their Corsi numbers. So he played guys he had favorable advantages against to generate more shots against, relative to his team and relative to the opponent's players. However his own ability to generate shots relative to his team versus shots against was still only middle of the pack, despite this advantage given to him.

Meanwhile his PP numbers indicate even easier competition, B-unit PK teams probably, similar to Kessel's competition on the PP. This combined with his 17% on ice shooting percentage for his team while on the PP and rather high PDO of 1080 indicates an overachieving year for him on the PP and regression to the mean would be following that during any longer season other than a lockout shortened one. His PP numbers are inflated relative to his actual abilities, so taken off his team and outside of his favorable matchups and favorable season you would expect a reasonable drop in production.

Just my two cents on ways advanced statistics can be useful under certain circumstances.

 

[henchman21]

Behind the Net and Extra Skater are the two sites that I use.

Advanced stats in general should be taken with a grain of salt. Some people preach them as gospel and think you can make decisions solely on them. Really you need to watch the game, remember or take notes on what you see, and then read the advanced stats to see if the stories jive. If they don't use the eye test over the advanced stats.

Zone starts, QoC, on ice shot and save percentage, Fenwick and its breakdowns (I prefer Fenwick over Corsi), and PDO (on ice shot plus on ice save percentage) are the numbers I really pay attention to (I only look at 5 on 5 numbers). I will look at Corsi, but I put less weight in it compared to Fenwick. The difference between Fenwick and Corsi is that Corsi counts all shot attempts, and Fenwick takes out blocked shots as blocked shots are considered a skill and can be controlled by the defensive team. Reading those and understanding what is going on in the game can show who is more effective on a team. Now all of this can be skewed greatly by the quality of the goaltender, which is my biggest complaint of advanced stats.

 

[zxcvnm]

Now all of this can be skewed greatly by the quality of the goaltender, which is my biggest complaint of advanced stats.

I don't see this as a problem with advanced stats. It would simply indicate that teams with better goaltending can get away with otherwise lower fenwick numbers. It wouldn't, by itself, negate a desire to fill a team with players with positive fenwick.

The real problem with advanced stats that I see is that isolating an individual contribution is currently not possible or current methodology doesn't do too great of a job. I don't think these statistics can tell us how a player with positive fenwick (say adjusted for QoC, zone starts, etc.) would perform if he were given different linemates, placed on a different team, or a different role. If you could isolate individual contribution effectively in some WAR-like stat, it could then be possible to build a solid team by searching the market for players who are otherwise undervalued based on this stat.

 

[henchman21]

I don't see this as a problem with advanced stats. It would simply indicate that teams with better goaltending can get away with otherwise lower fenwick numbers. It wouldn't, by itself, negate a desire to fill a team with players with positive fenwick.

It doesn't change Fenwick or Corsi at all, but it does greatly skew on and off ice save percentage, PDO (maybe the most important advanced stat), among others including the win probability models.

 

[Holyhell]

On a side note. I think it was on MvsW they mentioned that Roy is the first coach in the NHL that only uses man on man in the defensive zone. Other coaches use zone defense or a zone/man on man marking system. I wonder what that results in as far as allowing shots and opposition creating scoring chances. I have noticed fewer unmarked opponents in the slot, but it's only six games so it might be random.

I agree with everything you said in your post.

This is very interesting though (no offense to the rest of your post) that Roy is the only coach to use man to man. I like the strategy because I think that Colorado has the speed and smarts for most of our players to be able to keep up with 'their man' in the defensive zone. Ovechkin was the biggest test so far and he was pretty well neutralized... let's see what these guys can do against Zetterberg and Datsyuk.

 

[henchman21]

IMO the real test of Roy's defensive system will be when we face a dynamic offensive defensemen. Players like a Subban, Byfuglien, Keith, Karlsson, Yandle, etc. I think those player have the real potential to break down the defense. To date the most dynamic offensive defensemen the Avs have face would be Mike Green and he isn't the Mike Green of 5 years ago.

Looking at the schedule we will see this very soon with Letang on the 21st.

 

[S E P H]

I don't have a problem with advanced stats, but I prefer an observation point when talking about hockey. IMO Any observation > any advanced stats, though when a person has both at their disposal then they can seriously make a point. The problem I have is that majority of the people who do use advanced stats, do not watch the games. When this happens they're basically talking out of their ass.

 

[Lonewolfe2015]

On a side note. I think it was on MvsW they mentioned that Roy is the first coach in the NHL that only uses man on man in the defensive zone. Other coaches use zone defense or a zone/man on man marking system. I wonder what that results in as far as allowing shots and opposition creating scoring chances. I have noticed fewer unmarked opponents in the slot, but it's only six games so it might be random.

Roy doesn't use a man on man defensive system. He uses a combination of zone defense and 3rd man back man on man, whatever the hell the official term is.

Take the typical 1st line shift where they lose the puck. Opposing forwards create a 3 on 2, spread wide and force the defenders to not isolate out any of the potential other two players who do not have the puck. If ROR or Duchy are able to backcheck the defense pressure the puck carrier and the 3rd "defender" aka the trailing forward isolates the high slot player (generally the center) and tracks them throughout the play until the center gets back into the play defensively.

So whomever says he plays man on man strictly is wrong, what is interesting is that he tends to play a collapsing 3-1-1 if you will, where one winger is always around the high slot and one winger is always around the point with the center acting as a third defender and cycling the defensive pickups with the defenders. You'll notice EJ/Hejda/Duchy switching their man constantly depending who is closer.