Adjusted Even-Strength Plus-minus 1960-2017

[Canadiens1958]

Guy Carbonneau.

Yes.

I'll show Carbonneau's numbers for the two parts of his career - when he was a decent even-strength scorer and when he was a pure defensive player.

Player | Year | SFrac | $ESGF/G | $ESGA/G | R-ON | R-OFF | XEV+/- | EV+/- | AEV+/- | /Season
Guy Carbonneau | 83-92 | 9.6 | 0.71 | 0.58 | 1.22 | 1.21 | 63 | 102 | 39 | 4
Guy Carbonneau | 93-00 | 7.0 | 0.51 | 0.48 | 1.06 | 1.26 | 42 | 18 | -24 | -4
Guy Carbonneau | Total | 17 | 0.62 | 0.54 | 1.16 | 1.23 | 105 | 120 | 15 | 1

Not sure if 1992 was the right cutoff year, but he won his last Selke that year so I thought I should include it in the first half. His numbers look even better if you just look at 83-90 (1.30 R-ON and 1.19 R-OFF), but that might be unfair cherry-picking.

To put these numbers in context, consider who Carbonneau's linemates were, what his role was in terms of which opposing lines he would play against, and how good the other forward lines on his team were. I'll defer to you on detailed knowledge of the Canadiens, but from what I know Carbonneau's numbers are very impressive considering his role. Very few checking line players have had a positive adjusted plus-minus.

1992 is the accurate cut-off. He was injured during the 1992-93 season and assumed a more defensive role afterwards.

BTW in the Q with Chicoutimi, Guy Carbonneau was an excellent point quarterback on the PP.

 

[tombombadil]

seeing as you are possibly doing requests.... Konstantinov?

 

[VanIslander]

Another reason why it's blasphemy to think Lidstrom is better than Bourque.

Rk | Player | SFrac | $ESGF/G | $ESGA/G | R-ON | R-OFF | XEV+/- | EV+/- | AdjEV+/- | /Season | PP% | $PPP/G | SH%
1 | Ray Bourque | 20.30 | 1.17 | 0.85 | 1.37 | 0.96 | -62 | 524 | 586 | 29 | 88% | 0.45 | 58%
2 | Bobby Orr | 7.68 | 1.81 | 0.84 | 2.17 | 1.08 | 54 | 610 | 556 | 72 | 98% | 0.67 | 63%
5 | Mark Howe* | 11.70 | 1.13 | 0.75 | 1.50 | 0.98 | -26 | 354 | 380 | 32 | 59% | 0.24 | 42%
7 | Larry Robinson | 17.34 | 1.31 | 0.82 | 1.60 | 1.33 | 331 | 697 | 366 | 21 | 49% | 0.20 | 45%
10 | Al MacInnis | 17.72 | 1.10 | 0.78 | 1.40 | 1.11 | 112 | 468 | 356 | 20 | 87% | 0.49 | 39%
20 | Larry Murphy | 20.35 | 1.06 | 0.88 | 1.20 | 1.02 | 21 | 298 | 276 | 14 | 65% | 0.32 | 32%
23 | Denis Potvin | 13.27 | 1.18 | 0.79 | 1.50 | 1.23 | 166 | 419 | 253 | 19 | 87% | 0.43 | 53%
25 | Nicklas Lidstrom | 15.62 | 1.19 | 0.84 | 1.41 | 1.23 | 199 | 452 | 253 | 16 | 72% | 0.42 | 54%

 

[Pear Juice]

Another reason why it's blasphemy to think Lidstrom is better than Bourque.

Rk | Player | SFrac | $ESGF/G | $ESGA/G | R-ON | R-OFF | XEV+/- | EV+/- | AdjEV+/- | /Season | PP% | $PPP/G | SH%
1 | Ray Bourque | 20.30 | 1.17 | 0.85 | 1.37 | 0.96 | -62 | 524 | 586 | 29 | 88% | 0.45 | 58%
2 | Bobby Orr | 7.68 | 1.81 | 0.84 | 2.17 | 1.08 | 54 | 610 | 556 | 72 | 98% | 0.67 | 63%
5 | Mark Howe* | 11.70 | 1.13 | 0.75 | 1.50 | 0.98 | -26 | 354 | 380 | 32 | 59% | 0.24 | 42%
7 | Larry Robinson | 17.34 | 1.31 | 0.82 | 1.60 | 1.33 | 331 | 697 | 366 | 21 | 49% | 0.20 | 45%
10 | Al MacInnis | 17.72 | 1.10 | 0.78 | 1.40 | 1.11 | 112 | 468 | 356 | 20 | 87% | 0.49 | 39%
20 | Larry Murphy | 20.35 | 1.06 | 0.88 | 1.20 | 1.02 | 21 | 298 | 276 | 14 | 65% | 0.32 | 32%
23 | Denis Potvin | 13.27 | 1.18 | 0.79 | 1.50 | 1.23 | 166 | 419 | 253 | 19 | 87% | 0.43 | 53%
25 | Nicklas Lidstrom | 15.62 | 1.19 | 0.84 | 1.41 | 1.23 | 199 | 452 | 253 | 16 | 72% | 0.42 | 54%

My first instinct when I saw these numbers was that the dynasty team players are downvalued very much. I haven't really put any effort into analyzing the adjustment algorithm here though, I'll just conclude that there does seem to be a very large adjustment. Robinson's, Potvin's and Lidström's +/- are corrected by 40-50%. Is that really sensible?

I don't mean to offend the analysis made, it's a great effort, it's just that I instinctly reacted to such enormous adjustment factors.

 

[BraveCanadian]

My first instinct when I saw these numbers was that the dynasty team players are downvalued very much. I haven't really put any effort into analyzing the adjustment algorithm here though, I'll just conclude that there does seem to be a very large adjustment. Robinson's, Potvin's and Lidström's +/- are corrected by 40-50%. Is that really sensible?

I don't mean to offend the analysis made, it's a great effort, it's just that I instinctly reacted to such enormous adjustment factors.

The better your overall team throughout your career, the worse you'll look when adjusted.

If you're head and shoulders above the rest of your team for much of your career (Bourque is a good example), or have no comparable off ice counterpart, you'll look great in comparison to them.

overpass didn't overlook these things when he posted the results but everyone just looks at the results.

 

[overpass]

seeing as you are possibly doing requests.... Konstantinov?

Player | SFrac | $ESGF/G | $ESGA/G | R-ON | R-OFF | XEV+/- | EV+/- | AEV+/- | /Season
Vladimir Konstantinov | 5.8 | 1.06 | 0.68 | 1.56 | 1.29 | 69 | 181 | 112 | 19

My first instinct when I saw these numbers was that the dynasty team players are downvalued very much. I haven't really put any effort into analyzing the adjustment algorithm here though, I'll just conclude that there does seem to be a very large adjustment. Robinson's, Potvin's and Lidström's +/- are corrected by 40-50%. Is that really sensible?

I don't mean to offend the analysis made, it's a great effort, it's just that I instinctly reacted to such enormous adjustment factors.

I've downgraded the adjustment a touch in my current version. I have Bourque at +577, Orr at +562, Robinson at +411, MacInnis at +369, M. Howe at +367, Murphy at +278, Potvin at +275, Lidstrom at +353.

I admit the adjustment process is a weak point. It's very subjective and it's a one-size-fits-all adjustment. I've run some numbers to support it, but a large part of determining the correct factor is just having the results look right. If your first reaction is that the adjustment looks too high, I'll certainly take that into account. I think it was a little too high as well, which is why I've updated my current numbers.

 

[tombombadil]

****, i love this site. i don't know one damned person that thinks like this

 

[plusandminus]

My first instinct when I saw these numbers was that the dynasty team players are downvalued very much. I haven't really put any effort into analyzing the adjustment algorithm here though, I'll just conclude that there does seem to be a very large adjustment. Robinson's, Potvin's and Lidström's +/- are corrected by 40-50%. Is that really sensible?

I don't mean to offend the analysis made, it's a great effort, it's just that I instinctly reacted to such enormous adjustment factors.

I have said that too. I think dividing actual +/- with say square root of R-Off (or similar expondent) would give better results.
If Lidstrom's actual +/- is +452, and his R-Off is 1.23.
And Bourque's actual +/- is +524 and his R-Off is 0.96.
For Lidstrom: +452 / sqr(1.23) = +407.55 (not +253)
For Bourque: +524 / sqr(0.96) = +534.80 (not +586)
To me, it's very logical.

(Is that how overpass does it too, except using another exponent? Maybe I've misunderstood something? But the results looks different.)

Of course the method is still not perfect. R-Off can be very "biased" in itself. R-On and R-off basically compares the +/- of a player with the +/- of teammates playing on the same position as him. I would encourage people to actually think more about R-On and R-Off, what affects them, what can be read out of them, etc.
Example:
An average left wing playing on a line with two great teammates, will likely see his +/- and R-On be good. Another, better, left wing on his team may play with two poorer linemates, and his +/- and R-On may look below average despite being the better player of the two left wings. Since the first left wing has the benefit of playing with great linemates, not only does his R-On become high, but his R-Off will become low. His R-Off will make him look as if he played with poor teammates, while he actually had the benefit of playinng with two great linemates. (And the reverse will be true for the other left wing. He is a better player, but plays with poorer linemates. But he gets a low R-On, and a higher R-Off, indicating that he performed below average on a team that was perhaps above average. In reality, he was the best left wing on the team, and played with very poor players.)

Actual level of player | Actual level of linemates | R-On | R-Off | adjusted +/-
Average | Great | Above average, say 1.2 | Below average, say 0.8 | e.g. +50
Above average | Poor | Below average, say 0.8 | Above average, 1.05 | e.g. -40

Same could be true to defencemen, probably to a lesser extent but still.
That's only one of the things that may bias R-On and R-Off.
Overpass may be aware of this, but my general impression is that most people aren't even aware of what the stat says, what actually affects it, etc.

 

[BraveCanadian]

That's only one of the things that may bias R-On and R-Off.
Overpass may be aware of this, but my general impression is that most people aren't even aware of what the stat says, what actually affects it, etc.

You're right that a lot of context has to be taken into account with these numbers. And overpass is aware of at least some of those factors:

See here for updated numbers through 2010 for active players.

For the above reasons, please keep the following in mind when using these numbers to evaluate players

• Adjusted plus-minus is best used to compare players who played in a similar role. For example, compare #1 defensemen who played the toughest ice-time on the team to other #1 defensemen, not to #6 defensemen who were sheltered by their coaches from the best players. For example, take Tom Preissing’s rating with a huge grain of salt.
  
• Adjusted plus-minus is measured against a baseline of average, so it will tend to underrate players with a long decline phase or several poor years at the start of their career (Mark Messier) and give high ratings to players who retired young and didn’t play a lot past their prime(Bobby Orr, Eric Lindros).
• Adjusted plus-minus is measured against a baseline of average, so it will tend to underrate players with a long decline phase or several poor years at the start of their career (Mark Messier) and give high ratings to players who retired young and didn’t play a lot past their prime(Bobby Orr, Eric Lindros).
• Check to see who the player’s linemates were. Did he have a great player on his line? Charlie Simmer and Dave Taylor both have very high ratings, and likely owe much of it to Marcel Dionne.
• Did the player play on a team with another great player who was on another line/D-pairing? If so, his adjusted plus-minus may be too low. Mark Messier in his Edmonton years is an example here, along with Ted Green. I don’t think there are too many cases of this kind, but there are certainly a few.
• There may be a significant amount of random variation in a single-year result. For that reason, I would look at multiple years when measuring a player’s peak, and would not use this stat as definite proof that one player was better than another in a given year.

There are a lot of disclaimers there, but I still believe there is a lot of good information in adjusted plus-minus when evaluating a player’s career. Even after taking the above possible biases into account, there are still some very interesting results.

 

[plusandminus]

You're right that a lot of context has to be taken into account with these numbers. And overpass is aware of at least some of those factors:

Yes, I know. And it is being repeated occasionally.

I hope you and others understood my example with Lidstrom and Bourque. (It's hard for me to know if people do, or if they think it's too advanced. ?) I responded to a comment questioning the sensibility of the AEV+/- column. I agree the adjustments - when looking at them - doesn't look very sensible. (I may be wrong, I may always be wrong.) So I suggested how I think I would have made the calculation (a relatively simple calculation). I have no idea if that part is understandable, or if readers just think like "What does he know?" without even looking at the example.

 

[Pear Juice]

I have said that too. I think dividing actual +/- with say square root of R-Off (or similar expondent) would give better results.
If Lidstrom's actual +/- is +452, and his R-Off is 1.23.
And Bourque's actual +/- is +524 and his R-Off is 0.96.
For Lidstrom: +452 / sqr(1.23) = +407.55 (not +253)
For Bourque: +524 / sqr(0.96) = +534.80 (not +586)
To me, it's very logical....continues on

I understand your point. The problem is that there's really no reason to change the exponent of the adjustment factor, except for the reason that it gives you data more similar to the reality of the actual stats. You could change it to the square root. Or raise it to the power of nine. Or multiply it with the inverse of pi. It might make sense in a hockey kind of way, because the numbers fit better to what we have in mind, but is there really any mathematical reasoning behind it?

This really applies to all kinds of adjustments. What's basically going on is that you're trying to fit your data to a curve. You have a plan in your mind about how this curve looks. Well what if the curve doesn't even exist? What if there is no correlation between these sets of data?

It has been said by some other poster here that adjustment is difficult as you don't know what you should be heading for. That really nails down the problem, there's no way to validate your answers. The only validation possible is the one I just tried, to see if the numbers look sensible or not. But then they're just relative to my view of what's sensible or not, which in turn hails back to the original stats.

 

[matnor]

This is very interesting work and there are a lot of things to take from this but obviously this stat needs very much to be viewed in context. As overpass (and others) pointed out the quality of teammates both on and off the ice as well as match-ups are important to take into consideration. Nowadays, stats like QUALCOMP and QUALTEAM can be used for this purpose but this is not possible for historical data since it requires detailed ice time data. However, I think it is possible to produce a version of the QUALTEAM stat by using point collaboration data instead of actual ice time. It wouldn't be as good but it might still be useful. Essentially it could be produced by estimating joint icetime for two players by the share of ES points they were in on together. Unfortunately, I can't think of any way to estimate the quality of competition.

Just out of curiousity, how was expected ES +/-, XEV+/-, calculated, I didn't quite get that. I haven't read through the whole thread so my apologies if it has already been answered.

 

[seventieslord]

If Lidstrom's actual +/- is +452, and his R-Off is 1.23.
And Bourque's actual +/- is +524 and his R-Off is 0.96.
For Lidstrom: +452 / sqr(1.23) = +407.55 (not +253)
For Bourque: +524 / sqr(0.96) = +534.80 (not +586)
To me, it's very logical.

I was about to reply to this but Matnor beat me to it, and said the exact same thing I would have said, only much better. It's apples divided by oranges, for one thing. An accumulated number divided by an average.

 

[matnor]

I was about to reply to this but Matnor beat me to it, and said the exact same thing I would have said, only much better. It's apples divided by oranges, for one thing. An accumulated number divided by an average.

I think you give me credit for something Der Kaiser said

 

[plusandminus]

I was about to reply to this but Matnor beat me to it, and said the exact same thing I would have said, only much better. It's apples divided by oranges, for one thing. An accumulated number divided by an average.

For some reason, you seem to downgrade (I hope that is a correct word, my English isn't as good as I wish it was) what I write about stats. You do not seem to think very highly of me, and seem to repeatedly interpret things to my disadvantage. I have to admit it does annoy/frustrate me a bit.

In what way do you associate what matnor wrote to what I wrote?

Can you elaborate why one cannot divide +/- with R-Off?
Why does it matter if R-Off is an average?

 

[plusandminus]

However, I think it is possible to produce a version of the QUALTEAM stat by using point collaboration data instead of actual ice time. It wouldn't be as good but it might still be useful. Essentially it could be produced by estimating joint icetime for two players by the share of ES points they were in on together. Unfortunately, I can't think of any way to estimate the quality of competition.

I made a quick study on the 2002-03 season.
http://hfboards.com/showthread.php?t=969551&page=2
Post #38 in the thread.
My impression is that estimating joint ice time based on point shares doesn't seem very reliable (although perhaps better than nothing). What is your impression?

 

[plusandminus]

I understand your point. The problem is that there's really no reason to change the exponent of the adjustment factor, except for the reason that it gives you data more similar to the reality of the actual stats. You could change it to the square root. Or raise it to the power of nine. Or multiply it with the inverse of pi. It might make sense in a hockey kind of way, because the numbers fit better to what we have in mind, but is there really any mathematical reasoning behind it?

This really applies to all kinds of adjustments. What's basically going on is that you're trying to fit your data to a curve. You have a plan in your mind about how this curve looks. Well what if the curve doesn't even exist? What if there is no correlation between these sets of data?

It has been said by some other poster here that adjustment is difficult as you don't know what you should be heading for. That really nails down the problem, there's no way to validate your answers. The only validation possible is the one I just tried, to see if the numbers look sensible or not. But then they're just relative to my view of what's sensible or not, which in turn hails back to the original stats.

I agree that the exponent will determine the curve. An exponent below 1 will "smooth it down", while a exponent above one will sort of "make it more dramatic" (or how to put it). The inversed pi would, if I understand you right, make basically all players' adjusted +/- look much better than their actual one.
But my point wasn't necessarily what kind of exponent one uses, but rather about the logic with dividing +/- (players +/- on ice) by R-Off (simplified: teammates' +/- when player off the ice). I have, however, found an exponent to give the "best" results according to the "eye-test".

I am driven by curiousity, and a wish to examine things. That's why I do my studies, to look how things seem to correlate, to try to put things in larger context, etc.

Often it's a process.
For example, since the 1980s was more high scoring, we might want to adjust the point totals of today's players, to make up for that difference. The assumpion usually made, is that the points should be adjusted based on a lineary correspondence between players' point totals, and goals per game. A player scoring 80 pts in an era of 8 goals per game, will end up equal to a player scoring 60 pts in an era of 6 goals per game. (Simplified.) That is a basic adjustment. And yes, I suppose it's often based on an assumption in ones mind - at least initially. Then one tries it out, study the results, think hard about them, tries to find natural explanations for things, etc, etc. One comes up with theories on how to refine them, tries them out, study results, think, looks for natural explanations. And so on.

I have written earlier that I think it's sort of "philosophical". For example, what is best? Is it to play 100 minutes in PK and allow 10 goals (not scoring any). Or to play nothing at all? What about 10 minutes and 0 goals? What is best, considering all other things equal? Not easy to determine, I think.

 

[seventieslord]

For some reason, you seem to downgrade (I hope that is a correct word, my English isn't as good as I wish it was) what I write about stats. You do not seem to think very highly of me, and seem to repeatedly interpret things to my disadvantage. I have to admit it does annoy/frustrate me a bit.

In what way do you associate what matnor wrote to what I wrote?

Can you elaborate why one cannot divide +/- with R-Off?
Why does it matter if R-Off is an average?

I meand der kaiser, not matnor.

As someone who understands numbers you should understand why an accumulated total should not just be divided by a ratio for no good statistical reason other than you think it looks good. A ratio over a ratio at least makes sense logically.

It would be possible to show you examples using alternate realities where one of the two factors changed for the better or worse, yet the players result looked the opposite of what you'd logically think, but... I won't have the time for that for at least 3 weeks. And I am not the best guy to do it anyways. I am sure the "best guys" will eventually though.

 

[Pear Juice]

I agree that the exponent will determine the curve. An exponent below 1 will "smooth it down", while a exponent above one will sort of "make it more dramatic" (or how to put it). The inversed pi would, if I understand you right, make basically all players' adjusted +/- look much better than their actual one.
But my point wasn't necessarily what kind of exponent one uses, but rather about the logic with dividing +/- (players +/- on ice) by R-Off (simplified: teammates' +/- when player off the ice). I have, however, found an exponent to give the "best" results according to the "eye-test".

I am driven by curiousity, and a wish to examine things. That's why I do my studies, to look how things seem to correlate, to try to put things in larger context, etc.

Often it's a process.
For example, since the 1980s was more high scoring, we might want to adjust the point totals of today's players, to make up for that difference. The assumpion usually made, is that the points should be adjusted based on a lineary correspondence between players' point totals, and goals per game. A player scoring 80 pts in an era of 8 goals per game, will end up equal to a player scoring 60 pts in an era of 6 goals per game. (Simplified.) That is a basic adjustment. And yes, I suppose it's often based on an assumption in ones mind - at least initially. Then one tries it out, study the results, think hard about them, tries to find natural explanations for things, etc, etc. One comes up with theories on how to refine them, tries them out, study results, think, looks for natural explanations. And so on.

I have written earlier that I think it's sort of "philosophical". For example, what is best? Is it to play 100 minutes in PK and allow 10 goals (not scoring any). Or to play nothing at all? What about 10 minutes and 0 goals? What is best, considering all other things equal? Not easy to determine, I think.

Sorry, it probably didn't come across right. My point was that it doesn't matter what exponent you use unless there's a mathematical reason that you use it. Using the sqrt will smoothen out the adjustment, but is there any point in smoothing it out, besides making it more alike the original data? Maybe the adjustment needs smoothening, maybe it doesn't. But who can tell?

What we're doing when adjusting sports statistics to account for how the game was played in different time periods is we're trying to create a hypothetic space where all players can be compared on the same premises. Given the hypothetical nature of the analysis, this is extremely hard to validate. Therefore you need good backing for any terms that you decide to introduce into your algorithm.

Why use the sqrt on the R-off value when you might aswell do the third sqrt? Or multiply it by three quarters? I have absolutely no idea as to what correction function you should use, but in order for your analysis to be of good quality, you need to present one.

Don't take it to harshly, I think it's interesting work. There's lots of information hidden in the statistics and I appreciate the effort people put into analyzing them. I for one ain't got the time nor energy to do it!

Just don't wobble the bits and pieces around too much, then you might have a hard time putting them back in the original frame.

 

[matnor]

I made a quick study on the 2002-03 season.
http://hfboards.com/showthread.php?t=969551&page=2
Post #38 in the thread.
My impression is that estimating joint ice time based on point shares doesn't seem very reliable (although perhaps better than nothing). What is your impression?

I agree it's problematic but aggregating over entire careers the errors might become relatively smaller. Unfortunately, it wouldn't be possible to do what you did in that thread since it would require knowledge of all players on the ice for each goal which is not available for historical data. In the end it would be possible to test whether such a stat correlates well with QUALTEAM which is available for the last couple of seasons so it remains to be seen if the methodology works.

 

[plusandminus]

As someone who understands numbers you should understand why an accumulated total should not just be divided by a ratio for no good statistical reason other than you think it looks good. A ratio over a ratio at least makes sense logically.

Actually, I don't understand your logic here.

A prime example would be adjusted points. The most simple model is to take Pts (accumulated during a season), and adjust it based on GPG for the season, and adjust it to number of games played. An accumulated value is divided by an average, and then multiplied by a factor. Nothing strange or wrong with that.
Pts / (GPG / standardGPG) * (teamGP / 82)
There are other ways to get the same result. For example:
Pts / (GoalsPerTeam / standardGoalsperTeam)
This is basically the same logic as my proposed +/- adjustment:
PlusMinus / rOff
or
PlusMinus / (PlusMinusWhenOff / 1)

I don't know if you have actually studied (or understood) what you question, or just replied based on some preconception or something someone else have said. If you haven't actually spent time thinking about it, and experimenting with it, I suggest you do when you get the time.

You are also welcome to think more about the danger with dividing a ratio with another ratio (in this case R-On by R-Off). Because in this case, that might cause the bias I think you think my method would cause.

R-Off tells us how good the team were without the player on the ice. When adjusting +/-, it is not necessary to bring R-On into the formula. One can do it, depending what one wants to calculate, but it's not necessary.
R-On is basically a player's +/- expressed in a different way.
Using both +/-, ROn and ROff in the formula, would sort of make +/- be used twice, if that's how you meant.

I could write a lot more about this too, but hope it's not necessary.

I have written before that I'm basically against trying to adjust +/-. When I did in-depth studies in 2002-03, I ended up thinking +/- says most when looking within single teams. Edit: "Against" may be the wrong word, I think it is often interesting. I rather meant one often cannot draw very good conclusions based on it.

It would be possible to show you examples using alternate realities where one of the two factors changed for the better or worse, yet the players result looked the opposite of what you'd logically think, but... I won't have the time for that for at least 3 weeks. And I am not the best guy to do it anyways. I am sure the "best guys" will eventually though.

I think I toughed that subject myself about 11 posts ago in this thread. Yes, it can show "opposite results", just like your method can too (if I understand you right). I even wrote about that, giving examples, and showing a table illustrating the effect.

But the "reverse" results is not a result of what you seem to think it is. It has to do with the how good the teammates are. A "below average sharing lots of ice time with above average teammates", can according to the adjustments look like an "above average player playing with below average teammates". See the post/table.
No matter what method one uses, both methods are lacking there.

 

[seventieslord]

R-On is basically a player's +/- expressed in a different way.

Right, +/- is just R-On times games played, times the scoring environment (goals per game) that the player played in. (i.e. players with 5GF,4GA and 10GF,8GA are both 1.25, but one's +1 and the other's +2, to put it simply)

so, by using +/- in the calculation that I'm criticizing, you just introduced the same bias that is involved in comparing raw numbers from eras with different scoring levels. Why reintroduce that? overpass' system takes it out, and rightly so.

You may find the results somehow more palatable, but the way you got there just doesn't make sense.

I think I toughed that subject myself about 11 posts ago in this thread. Yes, it can show "opposite results", just like your method can too (if I understand you right). I even wrote about that, giving examples, and showing a table illustrating the effect.

But the "reverse" results is not a result of what you seem to think it is. It has to do with the how good the teammates are. A "below average sharing lots of ice time with above average teammates", can according to the adjustments look like an "above average player playing with below average teammates". See the post/table.
No matter what method one uses, both methods are lacking there.

nope, you don't know what I am talking about at this point.

I am poor at explaining this stuff. I am never very comfortable explaining a statistical error to someone equally as stats-inclined as myself, or maybe a bit less. If they're far behind me, sure, no problem. But the Ian Fyffes and overpasses of the world always do a far, far better job of articulating what I'm trying to, and I'll just bow out and leave it to them now, thanks.

 

[Czech Your Math]

Just out of curiousity, how was expected ES +/-, XEV+/-, calculated, I didn't quite get that. I haven't read through the whole thread so my apologies if it has already been answered.

I don't think he explained it, but I would guess something like this:

(ROFF - 1) * 200 * [player's (ESGF + ESGA)/ (200+200)]

calculated for each season, with the individual seasons' results summed

 

[plusandminus]

Right, +/- is just R-On times games played, times the scoring environment (goals per game) that the player played in. (i.e. players with 5GF,4GA and 10GF,8GA are both 1.25, but one's +1 and the other's +2, to put it simply)

Let's see what that results in...
Made up example: Player plays 80 games. ESGF=80. ESGA=40.
80-40 = +40 (+0.5 per game)
80/40 = 2.0 (that's R-On)
2.0 * 80 = +160
+40 = +160 ?? (Doesn't sound logical, but maybe I've missed something.)
And the we multiply that by the scoring equivalent.

so, by using +/- in the calculation that I'm criticizing, you just introduced the same bias that is involved in comparing raw numbers from eras with different scoring levels. Why reintroduce that? overpass' system takes it out, and rightly so.

I thought the +/- used in the division was a +/- adjusted for era, so that shouldn't make any difference. If R-Off is OK too, then I think my points stand.

We'll see what overpass and others say, if they comment.

You may find the results somehow more palatable, but the way you got there just doesn't make sense.

To me it does.

nope, you don't know what I am talking about at this point.

I am poor at explaining this stuff. I am never very comfortable explaining a statistical error to someone equally as stats-inclined as myself, or maybe a bit less. If they're far behind me, sure, no problem. But the Ian Fyffes and overpasses of the world always do a far, far better job of articulating what I'm trying to, and I'll just bow out and leave it to them now, thanks.

OK. You started this dialogue by claiming I was wrong. Since it was the xth time during the last weeks, and I think you've been wrong (at least most of) the earlier times, I took the time to argue with you.

 

[Czech Your Math]

Sorry, it probably didn't come across right. My point was that it doesn't matter what exponent you use unless there's a mathematical reason that you use it. Using the sqrt will smoothen out the adjustment, but is there any point in smoothing it out, besides making it more alike the original data? Maybe the adjustment needs smoothening, maybe it doesn't. But who can tell?

What we're doing when adjusting sports statistics to account for how the game was played in different time periods is we're trying to create a hypothetic space where all players can be compared on the same premises. Given the hypothetical nature of the analysis, this is extremely hard to validate. Therefore you need good backing for any terms that you decide to introduce into your algorithm.

Why use the sqrt on the R-off value when you might aswell do the third sqrt? Or multiply it by three quarters? I have absolutely no idea as to what correction function you should use, but in order for your analysis to be of good quality, you need to present one.

Don't take it to harshly, I think it's interesting work. There's lots of information hidden in the statistics and I appreciate the effort people put into analyzing them. I for one ain't got the time nor energy to do it!

Just don't wobble the bits and pieces around too much, then you might have a hard time putting them back in the original frame.

I agree with you completely. This is relevant to the adjusted plus-minus as well, where R-OFF is regressed (i.e. smoothed) to the mean. As you said, if there is not a rationale behind it (besides making the results look a bit better to some), then it becomes arbitrary. I think there is a piece missing to this puzzle, but darned if any of us has found it yet.