Adjusted +/-

[Rudruff]

Trying to understand a little more about what the "adjustment" is and how it is derived.

I understand that it is trying to overcome issues with +/-.




Relatively new to the advanced stats world of basketball, but I have some grounding in statistical analysis.

[DSMok1]

Raw +/- is just how the team did on the court compared to the other team. Really dependent on the other 9 players on the floor.

Adj +/- takes every matchup through the time period analyzed and looks at it like this:

P1 + P2 + P3 ... - P6 ... - P10 + HCAconst = (Tm1Pts-Tm2Pts)/Poss

And then solves for all 500 P values as one big system of equations.

Now, there are some tweaks and modifications to the system used to try to account for when P1 and P2 almost always play together (collinearity problems) or when P1 & P11 are ONLY subbed for one another (another source of collinearity). Also, players with only a few minutes will have wild results in this method.

[Rudruff]

Ok, so I can see a lot of variability in +/- on a game by game basis.

Adjusted +/- creates a series of smaller (and possibly more variable) values that you can add together and cancel out the variability via the large number of situations.

So the adjustment is constructed to account for all the systemic errors you can find among the variables.

[Nivek]

The problem with APM is that it takes a TON of data to get standard errors down to a reasonable level. It takes several seasons of data to get there, and at that point I'm wondering whether the numbers are saying anything worthwhile about a player's current performance. I still look at APM (and raw +/- as well) sometimes, but it's not the Holy Grail.

[floppymoose]

hunting around the web i found this:
http://www.82games.com/comm30.htm

[Doctor MJ]

The problem with APM is that it takes a TON of data to get standard errors down to a reasonable level. It takes several seasons of data to get there, and at that point I'm wondering whether the numbers are saying anything worthwhile about a player's current performance. I still look at APM (and raw +/- as well) sometimes, but it's not the Holy Grail.

The gap between the best and worst players in the league is always significantly bigger than the standard errors (for players with major minutes). People are so quick to say "Oh, can't use it!", but the reality is you just use it cautiously. You don't use it to say anything between players who are close to each other.

The concern over players changing from year to year and that making multi-year measurements invalid is understandable, but when you look at the charts, you see big name guys who've been consistently great chopping the list. If you know a guy has been inconsistent, then don't use APM over the inconsistent time period to judge him, for other players, there doesn't seem to be a problem.

Agree though, certainly no Holy Grail.

[Nivek]

I am NOT saying people shouldn't use APM. I use it all the time, but ONLY as part of analysis, NEVER as the lone (or even primary) tool.

Two biggest issues I have with APM:

1. It's more or less a black box measure. Unless you're equipped to handle lots of data, you can't replicate what APMers do, which means you have to sorta take on faith that they're handling the regressions and "adjustments" properly.

2. Standard errors are too large.

Okay, I have a third. Transferability from one team to another is questionable. For example, this season, Javale McGee has a strong positive APM -- most of which is attributable to a strong offensive on/off. Looking solely at APM, one might think McGee is a good offensive player. Shoot, if you look at his box score stats, one might be more convinced (low usage, but he shoots a good percentage). In fact, he has a primitive offensive game (put him in short-shorts and film in black & white and he'd look like someone from the 60s). The positive offensive on/off is in no small part because his backups are/were even worse -- Hilton Armstrong, Yi Jianlian, Kevin Seraphin.

I think all basketball stats (except maybe FT%) are affected by team context. APM may be more contextual than most, though.

[Doctor MJ]

Okay, I have a third. Transferability from one team to another is questionable. For example, this season, Javale McGee has a strong positive APM -- most of which is attributable to a strong offensive on/off. Looking solely at APM, one might think McGee is a good offensive player. Shoot, if you look at his box score stats, one might be more convinced (low usage, but he shoots a good percentage). In fact, he has a primitive offensive game (put him in short-shorts and film in black & white and he'd look like someone from the 60s). The positive offensive on/off is in no small part because his backups are/were even worse -- Hilton Armstrong, Yi Jianlian, Kevin Seraphin.

I think all basketball stats (except maybe FT%) are affected by team context. APM may be more contextual than most, though.

I feel like by the end you've almost made my response for me.

People knock APM because of its team dependence, but what if value is just that dependent on team context? If a player goes from one team to another, and isn't forced to change behavior, he'll probably shoot about the same, which will cause his volume scoring numbers to be relatively close to before - does that mean his net impact on the team will be about the same as it was before? I don't think that's a given by any means.

Obviously, regardless of what true value is, the fact that APM varies from role to role so dramatically means you have to be very cautious about using it to assess a player's potential just based on the stat - but I can't imagine acquiring any player just by looking at his stats.

[Nivek]

Maybe I did. And, I agree that there's value in getting own-team contextual value information. But...If there isn't transferability, then the stat has limited utility in player acquisition -- at least when talking about role players. I agree with your last thought -- no one should be acquiring players based solely on his stats.

Wait, are we sure this a stats board?

[Doctor MJ]

Maybe I did. And, I agree that there's value in getting own-team contextual value information. But...If there isn't transferability, then the stat has limited utility in player acquisition -- at least when talking about role players. I agree with your last thought -- no one should be acquiring players based solely on his stats.

Wait, are we sure this a stats board?

It's quite interesting actually contrasting RealGM's GB, with this board (and the "cool kids" from the PC board"), and with the APBRmetric board. The APBRmetric board has a tendency to be arguably as fundamentalist in its pro-stats stance as the GB is in its pro-eye ball stance. I go there and participate, there's good stuff, and there some really incredible people there (I mean for Christ's sake, Dean Oliver participates there), but it's quite common to encounter people with opinions that would have any GM rolling their eyes at them - and those people would just say that the GM's were wrong.

As an example: I was recently over there and brought up the last RealGM top 100 in a relevant just to see what people would think, and one of the well-established guys there said

but it also has steve nash listed as high as 39th, yet maurice cheeks isn't even on the list?

And I didn't get anyone else indicating that they thought Nash deserved to be significantly higher than Cheeks. Felt like an alien world. Obviously, a lot of people there just think differently from me.

On average I think this crew has the most healthy mindset balancing the two components, aware of the inevitable issues that come with too much hubris in either direction.

[Nivek]

I'm on the APBRmetrics board as well, though not as frequently as I used to post. There are some good guys over there, but I agree with your comments completely.

[drza]

I wonder if APM is knocked for being contextual more than it should be, with respect to the box score stats. For example, no one would think twice about giving someone's career box score stats in a discussion, despite the fact that players obviously change over time. Yet, those same people would hesitate to use a multi-year +/- calculation over a shorter period of time because players can change from one year to the next. Seems a bit...I don't know, not right to me.

I do recognize some of the weaknesses (i.e. black box, error, and huge amounts of info) that you point out. But I don't doubt that box score stats have just as many, if not more in the way of weaknesses. I (and, I believe you) are advocating using a combo of analysis tools, which is to me clearly the way to go, but I do find it a bit frustrating that box scores are still more often relied upon as defaults and presumably reliable valuators while the +/- stats continually have to be re-confirmed as useful at all. But I guess them's is the breaks with the "new kid on the block", so to speak.

[drza]

Maybe I did. And, I agree that there's value in getting own-team contextual value information. But...If there isn't transferability, then the stat has limited utility in player acquisition -- at least when talking about role players. I agree with your last thought -- no one should be acquiring players based solely on his stats.

Wait, are we sure this a stats board?

It's quite interesting actually contrasting RealGM's GB, with this board (and the "cool kids" from the PC board"), and with the APBRmetric board. The APBRmetric board has a tendency to be arguably as fundamentalist in its pro-stats stance as the GB is in its pro-eye ball stance. I go there and participate, there's good stuff, and there some really incredible people there (I mean for Christ's sake, Dean Oliver participates there), but it's quite common to encounter people with opinions that would have any GM rolling their eyes at them - and those people would just say that the GM's were wrong.

As an example: I was recently over there and brought up the last RealGM top 100 in a relevant just to see what people would think, and one of the well-established guys there said

but it also has steve nash listed as high as 39th, yet maurice cheeks isn't even on the list?

And I didn't get anyone else indicating that they thought Nash deserved to be significantly higher than Cheeks. Felt like an alien world. Obviously, a lot of people there just think differently from me.

On average I think this crew has the most healthy mindset balancing the two components, aware of the inevitable issues that come with too much hubris in either direction.

I read that Nash/Cheeks exchange you had on APBR, and thought you showed admirable restraint in not defending your player to the death. The thing is, so much of Nash's current value is tied into his huge impact that only really shows up in APM, which we don't have for Cheeks' day. Based on the box scores, as the guy seemed to be going off of, I was at least able to understand his stance. Though, as you mention, it seemed almost bizarre after hanging around here so much.

[Ozark]

Basketballvalue has been updated. Here are your top 7 players in the NBA.

Drumroll please...

1. Lebron James
2. Kevin Durant
3. Steve Nash
4. Chris Paul
5. Dirk Nowitzki
6. Dwayne Wade

and last but not least...

7. Nick Collison.

(cue the Price is Right tuba music)

[Doctor MJ]

Basketballvalue has been updated. Here are your top 7 players in the NBA.

Drumroll please...

1. Lebron James
2. Kevin Durant
3. Steve Nash
4. Chris Paul
5. Dirk Nowitzki
6. Dwayne Wade

and last but not least...

7. Nick Collison.

(cue the Price is Right tuba music)

Aside from the sample size issue I'm long on record saying that I don't think it makes sense to compare big minute players to small minute players in this stat simply because their coach is actively choosing when to play and when not to play small minute players based on what he sees on the floor at that point in time.

[Doctor MJ]

I read that Nash/Cheeks exchange you had on APBR, and thought you showed admirable restraint in not defending your player to the death. The thing is, so much of Nash's current value is tied into his huge impact that only really shows up in APM, which we don't have for Cheeks' day. Based on the box scores, as the guy seemed to be going off of, I was at least able to understand his stance. Though, as you mention, it seemed almost bizarre after hanging around here so much.

Thanks drza, though I cringe that Nash is "my player". Not saying your wrong, just saying I try not to be *that* guy.

It's absolutely the case that Nash's superiority in APM was completely off his radar - else I don't know how he could possibly feel confident that Cheeks was his equal. - and this is actually the type of fight I've been having less concretely with people on their for 6 years. When I talk to the so called "stat guys" at RealGM, they're well versed in all stats and use them to a variety of degrees. When I talk to guys over there, there are people who only focus on box score stats and only focus on +/- type stats, and both groups tend to be dead convinced that they are supremely rational in a world full of fools.

That sounds harsher than I mean it to be. The average quality of poster over there is certainly higher than on RealGM, and there are some brilliant people there. But there is a core of people here I think very highly of.

[AussieBuck]

Berri takes a shot at APM

And Keyon Dooling is one of the best players in the league according to Adjusted Plus-Minus (seriously, APM ranks him 15th in the league this year). But scorers are not always the most productive player (a story we have told many times before). And APM is not a very reliable measure of player performance (a story we have also told more than once before).

http://dberri.wordpress.com/2011/04/07/ ... milwaukee/

[floppymoose]

unadjusted +- has him around 30th. I'm not yet convinced that adjusted +- solves more problems than it causes.

[Ozark]

In all fairness, Nick Collison may not be top 25% in minutes, but he isn't a small sample size either.

Also I don't think his APM is as crazy as it seems. Very little box score stats, but the Thunder do always seem to pull away predictably when he is on the court. I'm not here for trolling purposes, I'd just like to hear an APM guy explain to me how it can pull a no box score guy like Collison out of a hat, and distinguish him and select him from among the other players on a very effective second team unit.

Teach me how to Dougie, c'mon now.

PS - I joined the Church of Berritology TM for two weeks. I was the dude who left Jonestown before the Kool-aid got passed around. Thank God I have a natural radar for creepy authoritarian vibes.

[DSMok1]

In all fairness, Nick Collison may not be top 25% in minutes, but he isn't a small sample size either.

Also I don't think his APM is as crazy as it seems. Very little box score stats, but the Thunder do always seem to pull away predictably when he is on the court. I'm not here for trolling purposes, I'd just like to hear an APM guy explain to me how it can pull a no box score guy like Collison out of a hat, and distinguish him and select him from among the other players on a very effective second team unit.

I've done With-or-Without-You at game level for OKC, and it shows the same amazing impact for Nick Collision. See http://godismyjudgeok.com/DStats/2011/n ... -and-nazr/