+/- anomalies

[sr edler]

thomas vanek

2005-06 -11
2006-07 +47
2007-08 -5

 

[Psycho Papa Joe]

I have spelled it out probably 10 times in previous threads. I don't have the energy to do it yet again.

In short, there are 12 players on the ice when a goal is scored. Giving out an individual statistic for the product of 12 players efforts is completely and totally inaccurate and lazy. People are looking for an easy way to measure if a player is good or not.

If a goal is scored, there are 10 non-goalies on the ice who are all credited with a + or a -. If it was simply one guy that screwed up, there is a 10% chance any given player on the ice was involved. Heck, let's assume 4 players efforts, positive or negative, led to a goal. That still means there is a 60% chance that any given player on the ice had nothing to do with it.

Conclusion: 60% of the time +/- means absolutely NOTHING in regards to an individual player. People are being credited with a stat that has nothing to do with their efforts. No wonder there are so many anomalies - most of the time the stat is complete BS. People trying to find an easy way to measure performance are simply fooling themselves into thinking +/- is somehow useful.

Then, of course, they come out with the classic line, in their most arrogant tone "You have to use it in context". They can cling to that myth and continue to fool themselves. The stat is so skewed and so random that it means nothing. And, if you need to spend time trying to figure out the "context" for every single player then, the stat is worthless in itself. Why not just do the research, watch the guy play then figure out if he is good or not? The number tells you nothing - unless you spend the time to figure out if it tells you something or not. That is moronic.

Imagine if a thermometer worked that way. Sometimes when you see the temperature sitting at 20 degrees celsius, it is a very comfortable day outside. But, sometimes 20 celsius means it is snowing. Other times, it says 20 but it is actually 40 and people are falling over with heat stroke. A thermometer like that would pretty quickly be tossed out as a piece of garbage because every single time you need to figure out the "context" to know if that temperature actually means what you think it might mean.

+/- is completely and utterly garbage. When people act so condescending, telling me I don't understand how to use the stat, I just have to laugh at their ignorance.

Look at that, you got me to spell it out one more time.

So in essence, you don't know how to use the stat to evaluate player performance. Gotch ya.

 

[overpass]

I have spelled it out probably 10 times in previous threads. I don't have the energy to do it yet again.

In short, there are 12 players on the ice when a goal is scored. Giving out an individual statistic for the product of 12 players efforts is completely and totally inaccurate and lazy. People are looking for an easy way to measure if a player is good or not.

If a goal is scored, there are 10 non-goalies on the ice who are all credited with a + or a -. If it was simply one guy that screwed up, there is a 10% chance any given player on the ice was involved. Heck, let's assume 4 players efforts, positive or negative, led to a goal. That still means there is a 60% chance that any given player on the ice had nothing to do with it.

Conclusion: 60% of the time +/- means absolutely NOTHING in regards to an individual player. People are being credited with a stat that has nothing to do with their efforts. No wonder there are so many anomalies - most of the time the stat is complete BS. People trying to find an easy way to measure performance are simply fooling themselves into thinking +/- is somehow useful.

Then, of course, they come out with the classic line, in their most arrogant tone "You have to use it in context". They can cling to that myth and continue to fool themselves. The stat is so skewed and so random that it means nothing. And, if you need to spend time trying to figure out the "context" for every single player then, the stat is worthless in itself. Why not just do the research, watch the guy play then figure out if he is good or not? The number tells you nothing - unless you spend the time to figure out if it tells you something or not. That is moronic.

Imagine if a thermometer worked that way. Sometimes when you see the temperature sitting at 20 degrees celsius, it is a very comfortable day outside. But, sometimes 20 celsius means it is snowing. Other times, it says 20 but it is actually 40 and people are falling over with heat stroke. A thermometer like that would pretty quickly be tossed out as a piece of garbage because every single time you need to figure out the "context" to know if that temperature actually means what you think it might mean.

+/- is completely and utterly garbage. When people act so condescending, telling me I don't understand how to use the stat, I just have to laugh at their ignorance.

Look at that, you got me to spell it out one more time.

I guess the Montreal Canadiens were wasting their time tracking the stat back in the 50s and 60s. Imagine what they could have done if they weren't so ignorant...

 

[Ogopogo*]

So in essence, you don't know how to use the stat to evaluate player performance. Gotch ya.

Keep on truckin' friend. You have obviously got it figured out.


When your neighbor gets convicted of dealing heroin, won't it be great when you get a criminal record too....just for living next door! Welcome to +/-.

 

[Rob Scuderi]

Keep on truckin' friend. You have obviously got it figured out.


When your neighbor gets convicted of dealing heroin, won't it be great when you get a criminal record too....just for living next door! Welcome to +/-.

I'm sure this perspective is somehow more ignorant that taking +/- at face value.

It's not like 80-82 games provide a large enough sample size to limit what you're suggesting or anything...

 

[Prophecy35]

Thanks for all the great responses, although, as I guessed it has devolved into a debate about the merits of +/-

My view is that +/- can be very useful when taken in context, not just at face value. If a player has a considerably higher or lower +/- than his teammates, then I think that is a good starting point. It's still more complicated than that, as you then have to look at situational stats (if the data is available).

Here is an example of what I consider to be "grounds to look further". Player X has a +/- rating of 50+ over a two year span. Those directly behind him on his team (player who placed second each year) over that same two year span had a +/- of around 35. This is not just one player, but the totals of the player finishing directly behind him in each of the two seasons, so Year #1 runner-up added to Year #2 runner-up. So, Player X was more than +15 higher over a two year span than the next two closest finishes combined.

In addition, Orr, as an extreme example, has simply ridiculous finishes in comparison with his teammates. From 68-75 (years he played 60+ games) he was a combined +549 (leading the league 6/7 years and being an average +78 in each season). The totals of the runners up on his team are +378 (+54 per season). So, during this span, on average, he was "out-plussing" his Bruins teammates by +24. It can be assumed from this, how effective Orr was for his team because of the insanely large gap.

Like how comparing eras should ideally (and fairly) be done by comparing their performance vs. their peers in the league and on their teams, comparing +/- data is best served when a player significantly outperforms the rest of his team consistently. Does that mean that a player who gets bad ratings is a lesser player? No, since they could be put in tougher situational matchups. However, I think there is some merit, when looking at +/- finishes vs. the rest of the team, especially over multiple seasons to rule out anomalies.

 

[tarheelhockey]

Like how comparing eras should ideally (and fairly) be done by comparing their performance vs. their peers in the league and on their teams, comparing +/- data is best served when a player significantly outperforms the rest of his team consistently. Does that mean that a player who gets bad ratings is a lesser player? No, since they could be put in tougher situational matchups. However, I think there is some merit, when looking at +/- finishes vs. the rest of the team, especially over multiple seasons to rule out anomalies.

So you're saying that even after you finish compiling multiple seasons of +/- and doing all these breakdowns, you still need to go find their Zone Starts and Qualcomp numbers for yet another round of analysis. Don't you see how that defeats the purpose of listing +/- in the first place?

I think it would do a lot of good if +/- were simply removed from the "standard" stat line. While it has some utility in the sort of advanced research described above, where it's being combined with other stats to produce an index, as a standalone number it really is truly useless. Comparable to putting GAON/60 on the standard stat line... a seriously misguided way to represent player performance in a vacuum. To the point where even defending it as being accurate "in context" doesn't go far enough toward explaining just how bad of a measure it really is. Leave it to the people who really know what they're doing with the numbers, rather than putting it right between Points and PIM as one of the standard measurements of performance.

 

[Psycho Papa Joe]

So you're saying that even after you finish compiling multiple seasons of +/- and doing all these breakdowns, you still need to go find their Zone Starts and Qualcomp numbers for yet another round of analysis. Don't you see how that defeats the purpose of listing +/- in the first place?

I think it would do a lot of good if +/- were simply removed from the "standard" stat line. While it has some utility in the sort of advanced research described above, where it's being combined with other stats to produce an index, as a standalone number it really is truly useless. Comparable to putting GAON/60 on the standard stat line... a seriously misguided way to represent player performance in a vacuum. To the point where even defending it as being accurate "in context" doesn't go far enough toward explaining just how bad of a measure it really is. Leave it to the people who really know what they're doing with the numbers, rather than putting it right between Points and PIM as one of the standard measurements of performance.

See now, this is a good answer, rather than just calling it a 'garbage stat'. The underlying data that goes into calculating the number is very usefull and that is how it should be used by scouts, coaches and managers. Basically taking the data, comparing it to the players that were on the ice at the same time and where the faceoff occurred and putting into a database where scouts, coaches and managers can analyse it, makes it very usefull. That data, plus somebody who knows how to interpret it, plus a decent computer would give a usefull indicator of performance.

 

[Padan]

Sakic had some strange seasons in the late 90's. In 96/97 he had by far the worst +/- rating on the Avs (-10) but was +23 two years later (second to Forsberg).

 

[plusandminus]

Jari Kurri: +18, -24, +19, -24
http://www.hockey-reference.com/players/k/kurrija01.html

You may find his 1992-93 season alone interesting.
http://www.hockey-reference.com/players/k/kurrija01/gamelog/1993/
During the first 22 games of the season, he was:
22 16+30 46 +24 (Calculated manually, so may be wrong.)
which was best on team in scoring and second in +/- (after Huddy). He accomplished that without Gretzky, who was injured during the first "half" of the season.
Then Tomas Sandstrom got injured, and (correlated or not) Kurri's scoring and +/- declines considerably.
The rest of the season, Kurri was:
60 11+30 41 -5
despite (or partly because of?) Gretzky playing in 45 of those games. Sandstrom coming back to play the last 13 or so games didn't seem to help his scoring either/neither.

 

[Dark Shadows]

Jari Kurri: +18, -24, +19, -24
http://www.hockey-reference.com/players/k/kurrija01.html

You may find his 1992-93 season alone interesting.
http://www.hockey-reference.com/players/k/kurrija01/gamelog/1993/
During the first 22 games of the season, he was:
22 16+30 46 +24 (Calculated manually, so may be wrong.)
which was best on team in scoring and second in +/- (after Huddy). He accomplished that without Gretzky, who was injured during the first "half" of the season.
Then Tomas Sandstrom got injured, and (correlated or not) Kurri's scoring and +/- declines considerably.
The rest of the season, Kurri was:
60 11+30 41 -5
despite (or partly because of?) Gretzky playing in 45 of those games. Sandstrom coming back to play the last 13 or so games didn't seem to help his scoring either/neither.

How did you get his game log to come up on that site?
Edit: Nevermind I found it

 

[seventieslord]

ogopogo is right. raw +/- is pretty useless. prophecy is right. adjusting +/- and doing things such as analyzing relative to teammates, can be useful.

Still, goals are so random. You can allow 10 shots against in a night and your goalie stops them all, and another guy allows just one and the goalie allows it. Who did better defensively? I prefer CORSI over +/-, easily, because it is basically +/- for shots. It removes a lot of noise and does a better job showing which players actually tilt the ice when they're on it. It allows for fun analysis at the end of the year, like who has a +/- that doesn't fit their CORSI? (it means they were either lucky or unlucky with regards to their team's and their opposition's goaltending - and you'd be surprised how much variance can show up in one player in just 82 games!)

of course, CORSI is only available post-lockout IIRC.

 

[Ogopogo*]

. It can be assumed from this, how effective Orr was for his team because of the insanely large gap..

Or, it could simply be that Orr had the most ice time on an incredibly dominant team. More playing time on a team that scores a lot more than it allows gives you a much higher +/-.

Unless you spend the time to dig deeper +/- gives you nothing of value. If you have the spend the time to dig deeper then the +/- number is worthless.

How do you explain Orr only being a +10 in '75-'76? He was 12th on the Bruins in +/-! FTW?????

 

[Iain Fyffe]

Or, it could simply be that Orr had the most ice time on an incredibly dominant team. More playing time on a team that scores a lot more than it allows gives you a much higher +/-.

If only Orr had something to do with the team scoring a lot more goals than it allowed. Then we might have something!

How do you explain Orr only being a +10 in '75-'76? He was 12th on the Bruins in +/-! FTW?????

You could at least try to come up with a good example. In 1975/76 Orr was +10 in 10 games played.

If you haven't at least moved beyond counting stats to rates, there's no helping you.

 

[Psycho Papa Joe]

ogopogo is right. raw +/- is pretty useless. prophecy is right. adjusting +/- and doing things such as analyzing relative to teammates, can be useful.

Still, goals are so random. You can allow 10 shots against in a night and your goalie stops them all, and another guy allows just one and the goalie allows it. Who did better defensively? I prefer CORSI over +/-, easily, because it is basically +/- for shots. It removes a lot of noise and does a better job showing which players actually tilt the ice when they're on it. It allows for fun analysis at the end of the year, like who has a +/- that doesn't fit their CORSI? (it means they were either lucky or unlucky with regards to their team's and their opposition's goaltending - and you'd be surprised how much variance can show up in one player in just 82 games!)

of course, CORSI is only available post-lockout IIRC.

Sure the odd game can produce something random, which is why you shouldn't look at the data on a one game basis. But when assessing a player over a quarter season or more, such randomness does tend to even itself out.

Hell goals and assists stats themselves are subject to randomness as well. Fluke goal or bad goal, vs beauty goal. Beautifull pass vs puck off one's butt. They all produce the same stat at the end of the day, but not all goals or assists are created equal. WHich is why you look at the data over an extended period of time, rather than one day's work.

 

[Dark Shadows]

If only Orr had something to do with the team scoring a lot more goals than it allowed. Then we might have something!

Exactly

You could at least try to come up with a good example. In 1975/76 Orr was +10 in 10 games played.

If you haven't at least moved beyond counting stats to rates, there's no helping you.

 

[Nalyd Psycho]

ogopogo is right. raw +/- is pretty useless. prophecy is right. adjusting +/- and doing things such as analyzing relative to teammates, can be useful.

Still, goals are so random. You can allow 10 shots against in a night and your goalie stops them all, and another guy allows just one and the goalie allows it. Who did better defensively? I prefer CORSI over +/-, easily, because it is basically +/- for shots. It removes a lot of noise and does a better job showing which players actually tilt the ice when they're on it. It allows for fun analysis at the end of the year, like who has a +/- that doesn't fit their CORSI? (it means they were either lucky or unlucky with regards to their team's and their opposition's goaltending - and you'd be surprised how much variance can show up in one player in just 82 games!)

of course, CORSI is only available post-lockout IIRC.

I prefer +/-. The theory is that goals resulting is a representation of the quality of shots. And with modern goaltending, letting 100 bad shots is better than letting 1 good shot. When you cross reference an individuals +/- to the rest of the team you can get a view of how goaltending effected it and then when broken down further to how a player played with different teammates it is a solid if unspectacular player efficiency stat.

 

[Ogopogo*]

If only Orr had something to do with the team scoring a lot more goals than it allowed. Then we might have something!


You could at least try to come up with a good example. In 1975/76 Orr was +10 in 10 games played.

If you haven't at least moved beyond counting stats to rates, there's no helping you.

Thank you for taking my bait and proving my point.

10 games played is EXACTLY my point. 11 players had better +/- than Orr that year because they played more games on a good team. +/- has so many external factors that are not controlled by the individual that it is a completely worthless number.

 

[Bear of Bad News]

Thank you for taking my bait and proving my point.

10 games played is EXACTLY my point. 11 players had better +/- than Orr that year because they played more games on a good team. +/- has so many external factors that are not controlled by the individual that it is a completely worthless number.

That was bait?

If your point is that every counting stat has the same bias as this, then congratulations on a job well done. Sidney Crosby only has two goals and ten assists this season, despite being incredibly talented. Therefore, we should no longer count goals and assists!

 

[Johnny Engine]

Sure the odd game can produce something random, which is why you shouldn't look at the data on a one game basis. But when assessing a player over a quarter season or more, such randomness does tend to even itself out.

Rory Fitzpatrick disagrees.

So you're saying that even after you finish compiling multiple seasons of +/- and doing all these breakdowns, you still need to go find their Zone Starts and Qualcomp numbers for yet another round of analysis. Don't you see how that defeats the purpose of listing +/- in the first place?

I think it would do a lot of good if +/- were simply removed from the "standard" stat line. While it has some utility in the sort of advanced research described above, where it's being combined with other stats to produce an index, as a standalone number it really is truly useless. Comparable to putting GAON/60 on the standard stat line... a seriously misguided way to represent player performance in a vacuum. To the point where even defending it as being accurate "in context" doesn't go far enough toward explaining just how bad of a measure it really is. Leave it to the people who really know what they're doing with the numbers, rather than putting it right between Points and PIM as one of the standard measurements of performance.

Absolutely right, but I don't think that it's even that hard to get some sort of context out of +/- from a standard stat line, providing you're not an alien who has never heard of the players he's looking at. If player X has an excellent +/-, not many points, high TOI, and plays on a bad team, that does all point towards "good defensively," no? None of those factors require much digging at all to quantify.
Of course, it does mean that I have to put up with dolts on my facebook feed day in and day out and their "LULZ, (insert Leaf) is a -6! He sux!", but is that really an important thing to worry about?
...that said, player agents could theoretically leverage that kind of popular opinion into disproportionate contracts, but I don't think any of us have enough fly-on-the-wall knowledge to say that.

 

[seventieslord]

I prefer +/-. The theory is that goals resulting is a representation of the quality of shots. And with modern goaltending, letting 100 bad shots is better than letting 1 good shot. When you cross reference an individuals +/- to the rest of the team you can get a view of how goaltending effected it and then when broken down further to how a player played with different teammates it is a solid if unspectacular player efficiency stat.

actually, if you can do shot-adjusted corsi, then you probably have the best stat of all. goals are the most important thing, but there's no denying they can be random. a stat based on something with such a small sample size (i.e. 80 GF, 60 GA) is bound to be a fail as opposed to something like "800 shots for, 600 shots against"

 

[Ogopogo*]

That was bait?

If your point is that every counting stat has the same bias as this, then congratulations on a job well done. Sidney Crosby only has two goals and ten assists this season, despite being incredibly talented. Therefore, we should no longer count goals and assists!

Read the post I made orignally (as per your request). It will make sense then.

 

[tarheelhockey]

Absolutely right, but I don't think that it's even that hard to get some sort of context out of +/- from a standard stat line, providing you're not an alien who has never heard of the players he's looking at. If player X has an excellent +/-, not many points, high TOI, and plays on a bad team, that does all point towards "good defensively," no? None of those factors require much digging at all to quantify.

Here's the problem, though.

http://www.nhl.com/ice/statshome.htm
Goals, assists, points, +/-

http://www.tsn.ca/nhl/statistics/
Goals, assists, points, PPG, SHG, +/-

http://espn.go.com/nhl/statistics
Goals, points, +/-

http://scores.nbcsports.msnbc.com/nhl/index.asp
Goals, points, +/-


The major outlets including the NHL present raw +/- as THE basic measure of defense. Even if you go a step deeper and look at the NHL's main stat sheets: http://www.nhl.com/ice/playerstats.htm#?navid=nav-sts-indiv

it gives you GP-G-A-P-plusminus-PIM as the basic stat line with 7 more columns until you get to TOI... which apparently ranks with shifts/game and faceoff% in their list of priorities. This sends a message to people looking at leaderboards that raw +/- is a meaningful statistic in and of itself. The people running stats for the league should surely be expected to know better, even if the media outlets don't.

 

[Nalyd Psycho]

actually, if you can do shot-adjusted corsi, then you probably have the best stat of all. goals are the most important thing, but there's no denying they can be random. a stat based on something with such a small sample size (i.e. 80 GF, 60 GA) is bound to be a fail as opposed to something like "800 shots for, 600 shots against"

While I agree. The problem is that letting a high volume of easy shots is a defensive tactic.

 

[seventieslord]

While I agree. The problem is that letting a high volume of easy shots is a defensive tactic.

I agree, but if corsi is probably adjusted for shot quality we probably have a winner.

(maybe it already is? I haven't researched it hard enough)