This is Google's cache of viewtopic.php?p=24011&sid=45282926e028c57d8c45de86586f4482. It is a snapshot of the page as it appeared on Apr 11, 2011 13:43:45 GMT. The current page could have changed in the meantime. Learn more

Text-only version
These search terms are highlighted: adjusted plus minus  
APBRmetrics :: View topic - Points Added: An Alternative to Adjusted Plus/Minus
APBRmetrics Forum Index APBRmetrics
The statistical revolution will not be televised.
 
 FAQFAQ   SearchSearch   MemberlistMemberlist   UsergroupsUsergroups   RegisterRegister 
 ProfileProfile   Log in to check your private messagesLog in to check your private messages   Log inLog in 

Points Added: An Alternative to Adjusted Plus/Minus
Goto page 1, 2  Next
 
Post new topic   Reply to topic    APBRmetrics Forum Index -> General discussion
View previous topic :: View next topic  
Author Message
Ryan J. Parker



Joined: 23 Mar 2007
Posts: 711
Location: Raleigh, NC

PostPosted: Sun Dec 14, 2008 5:54 pm    Post subject: Points Added: An Alternative to Adjusted Plus/Minus Reply with quote

I don't want to spend too much time on trying to rate players to a single number (since new research in this area will probably have the least impact), but I really want to understand a method for measuring a player's overall contribution while on the court. Adjusted plus/minus is the best way to do that. Problem is, I have a hard time understanding it. Therefore, I created something similar that I feel is more basic that I can more easily understand.

I call it points added, and you can find my full explanation here.

The points added offensive ratings:

http://spreadsheets.google.com/pub?key=pLJimPjd7oqu1jTSBj0LoPA&hl=en

The points added defensive ratings:

http://spreadsheets.google.com/pub?key=pLJimPjd7oquIUwMwHOHOPQ&hl=en

I don't exactly expect this to be preferred over adjusted plus/minus, but in my small world I am able to understand it better. Hopefully someone else finds this useful. Smile
_________________
I am a basketball geek.
Back to top
View user's profile Send private message Visit poster's website
deepak



Joined: 26 Apr 2006
Posts: 665

PostPosted: Sun Dec 14, 2008 9:40 pm    Post subject: Reply with quote

Perhaps I missed it ... you created these ratings based on which season(s)?
Back to top
View user's profile Send private message
Ryan J. Parker



Joined: 23 Mar 2007
Posts: 711
Location: Raleigh, NC

PostPosted: Sun Dec 14, 2008 10:13 pm    Post subject: Reply with quote

Yeah sorry about that. It's from last year, 07-08.
_________________
I am a basketball geek.
Back to top
View user's profile Send private message Visit poster's website
Mountain



Joined: 13 Mar 2007
Posts: 1527

PostPosted: Mon Dec 15, 2008 4:18 am    Post subject: Reply with quote

How does combination of offensive and defensive ratings correlate to single season 07-08 adjusted at basketballvalue or to the splits using the multi-season version at 82games?

Does your method have smaller errors than single season adjusted or the multi-season weighted version?

As a % of value is the method better at estimating the best on offense and worst on defense or am I getting the wrong impression from the ratio of the values to the standard deviations?

If you subtracting 12 or 13 (half the difference in the top to bottom spread, setting a zero value midpoint) from all the values to put them on a positive-negative would that look like adjusted or be properly scaled? Not saying you have or want to go back to that but my mind drifts back to that as it might make it more immediately apparent to me where a player falls on the distribution and whether their contribution is above or below average.

Given position trend of who is tops is points added somehow "perimeter friendly" on offense (probably about 90 perimeter guys in top 100) and bigman friendly on defense (at least 80 bigs in top 100) or is that to be taken as the objective measure of these impacts ?


Last edited by Mountain on Mon Dec 15, 2008 6:16 pm; edited 1 time in total
Back to top
View user's profile Send private message
cherokee_ACB



Joined: 22 Mar 2006
Posts: 157

PostPosted: Mon Dec 15, 2008 2:07 pm    Post subject: Re: Points Added: An Alternative to Adjusted Plus/Minus Reply with quote

Ryan J. Parker wrote:

I don't exactly expect this to be preferred over adjusted plus/minus, ...


It seems to me you are running essentially the same regression. The only difference between both ratings should be a constant, unless I'm misunderstanding what "sampling from the Normal distribution when fitting the data" means

By the way, I don't understand how that sampling constrains the ratings to be strictly positive.

Another question, how are you dealing with low minute players? What value do you assign them?
Back to top
View user's profile Send private message
Ryan J. Parker



Joined: 23 Mar 2007
Posts: 711
Location: Raleigh, NC

PostPosted: Mon Dec 15, 2008 2:48 pm    Post subject: Reply with quote

The biggest difference in the Points Added fit and the one that Adjusted +/- uses is that Adjusted +/- uses "shifts" of data in terms of points per 100 possessions, where as I use each individual possession in terms of the points scored on that possession. This is one of the reasons Adjusted +/- doesn't feel "clean" to me. I have read some articles on adjusted +/- that a weighting system is used, but thanks in large part to Eli's work I haven't been able to figure out how that works in a sensible manner. Hence my motivation for creating something I could wrap my head around, so I use each individual possession.

As for constraining the coefficients to be positive: With WinBUGS you can specify limits on the sampling distributions. In this case I limit the samples to be positive with the I(0,) piece of the model specification.

The low possession players are not included in any of the samples. I've read that Adjusted +/- uses these as "reference" players, yet another concept I don't understand. I'm not sure what the impact is of throwing these guys out instead of making them appear as one player.
_________________
I am a basketball geek.
Back to top
View user's profile Send private message Visit poster's website
Kevin Pelton
Site Admin


Joined: 30 Dec 2004
Posts: 979
Location: Seattle

PostPosted: Mon Dec 15, 2008 4:05 pm    Post subject: Reply with quote

Ryan J. Parker wrote:
The low possession players are not included in any of the samples. I've read that Adjusted +/- uses these as "reference" players, yet another concept I don't understand. I'm not sure what the impact is of throwing these guys out instead of making them appear as one player.

When you say not included, do you mean they're not part of the regression or that you're not including any possessions with low-possession players?

With adjusted plus-minus, I'm not sure we've spent enough time discussing the various ways different people have treated reference players, since the odd outlier can have an impact (I think).
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Ryan J. Parker



Joined: 23 Mar 2007
Posts: 711
Location: Raleigh, NC

PostPosted: Mon Dec 15, 2008 4:18 pm    Post subject: Reply with quote

What I've done is simply removed any possessions involving players that did not make the cutoff point. So only possessions that involved the 340 players that made the cutoff were included in the fit. The rest of the possessions were thrown in the trash. Cool

For the interested, that still left a total of 165,403 possessions.
_________________
I am a basketball geek.
Back to top
View user's profile Send private message Visit poster's website
Mountain



Joined: 13 Mar 2007
Posts: 1527

PostPosted: Mon Dec 15, 2008 5:20 pm    Post subject: Reply with quote

If you are willing to make this chop down would be willing to look at cutting the data again down to just when playoff teams play against playoff teams? That would interest me as a possibly improved indicator of playoff performance.
Back to top
View user's profile Send private message
Mike G



Joined: 14 Jan 2005
Posts: 3624
Location: Hendersonville, NC

PostPosted: Mon Dec 15, 2008 5:32 pm    Post subject: Reply with quote

I put DougStats.com's position designation next to player names, sorted by position, and got these averages:
Code:
pos    #    off    def    net
 C    61    7.8    6.8    0.9
PF    60    8.9    9.2   -0.2
SF    80    9.2    8.9    0.3
SG    68   11.1   11.6   -0.5
PG    81   10.9   13.0   -2.2

These are raw averages, not minutes-weighted.
Whatever you've done, you haven't ironed out the wrinkle suggesting that PG's are bad for you. Fine for offense, but a collective sieve on D.

Weird, huh?
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Ryan J. Parker



Joined: 23 Mar 2007
Posts: 711
Location: Raleigh, NC

PostPosted: Mon Dec 15, 2008 5:47 pm    Post subject: Reply with quote

I'm not sure what's "weird" about that, really. I'd expect some sort of distributional difference for each position.

Do you happen to have the Doug's Stats names in the same format as mine? If so, that would be helpful. If not, I'll parse them up later this week.

What I'll do is put player's together and present the distribution by position.
_________________
I am a basketball geek.
Back to top
View user's profile Send private message Visit poster's website
Mike G



Joined: 14 Jan 2005
Posts: 3624
Location: Hendersonville, NC

PostPosted: Mon Dec 15, 2008 6:11 pm    Post subject: Reply with quote

Heck, I already converted to First Name in one column, Last in another. E-Mail me if you wish I'd send it to you.

What's alarming about the PG deficit is that it's the one position that cannot go unfilled for more than a few seconds. Maybe, when an all-defense lineup goes in for a stop at the end of a quarter?

In football, there are just a few plays where there's no quarterback on the field: field goals, PAT's, punts, kickoffs. They might produce a stat that shows teams do better, in points per play, without a QB.

Meanwhile, how can all PG's net such a substantial negative? Even w the '08 Raptors, who seldom used Ford and Calderon together: both are majorly negative.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Kevin Pelton
Site Admin


Joined: 30 Dec 2004
Posts: 979
Location: Seattle

PostPosted: Mon Dec 15, 2008 7:21 pm    Post subject: Reply with quote

Mike G wrote:
In football, there are just a few plays where there's no quarterback on the field: field goals, PAT's, punts, kickoffs. They might produce a stat that shows teams do better, in points per play, without a QB.

Someone hasn't been watching the Miami Dolphins this year ...

Quote:
Meanwhile, how can all PG's net such a substantial negative? Even w the '08 Raptors, who seldom used Ford and Calderon together: both are majorly negative.

Didn't someone explain that this had to do with the Raptors playing ridiculously well in Darrick Martin's limited minutes? If so, that shouldn't necessarily translate into what Ryan is doing here, I would imagine.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Ryan J. Parker



Joined: 23 Mar 2007
Posts: 711
Location: Raleigh, NC

PostPosted: Mon Dec 15, 2008 9:26 pm    Post subject: Reply with quote

Well this is certainly something worth keeping an eye on, but there should not be a positional bias in the way the data is fit. If it turns out that there are perhaps proportionally more poor PGs than good PGs in relation to players at other positions then that does say something, but not that there is a positional bias.

That said, the size of the data means there is a lot of uncertainty. TJ Ford's net rating, for example, allows us to approximate that there is a 35% chance he has a net positive impact. Thus I would certainly want to look at other things when comparing him against other players to determine who I would want running my team.

What I do find interesting is that from this data a guy like Devin Harris has a high probability of being a net positive player (over 95%), and the probability that Devin Harris has a a higher impact than Jason Kidd is almost as high (over 93%). Hindsight being what it is, you have to wonder what the Mavs were thinking. That being said, this certainly doesn't measure the impact a coach has on these ratings, so maybe there is room for debate.
_________________
I am a basketball geek.
Back to top
View user's profile Send private message Visit poster's website
cherokee_ACB



Joined: 22 Mar 2006
Posts: 157

PostPosted: Tue Dec 16, 2008 4:49 am    Post subject: Reply with quote

Ryan J. Parker wrote:
The biggest difference in the Points Added fit and the one that Adjusted +/- uses is that Adjusted +/- uses "shifts" of data in terms of points per 100 possessions, where as I use each individual possession in terms of the points scored on that possession. This is one of the reasons Adjusted +/- doesn't feel "clean" to me. I have read some articles on adjusted +/- that a weighting system is used, but thanks in large part to Eli's work I haven't been able to figure out how that works in a sensible manner. Hence my motivation for creating something I could wrap my head around, so I use each individual possession.


Well, nothing prevents to run the Adj+/- regression using individual possessions. That is, nothing but processing power and data availability (at least in my case). But I see your point.

Quote:
As for constraining the coefficients to be positive: With WinBUGS you can specify limits on the sampling distributions.


Ok, I see. Anyway, this should have little impact as you get no near zero ratings. Note that the spread is the same as with EliW's ratings.

Another question: how would you combine the offensive and defensive ratings? Just subtract one from the other and get to this top players list?
Code:

Jamario Moon     14.3
Ronnie Price     12.7
Mike Dunleavy    12.0
Peja Stojakovic  11.5
Dirk Nowitzki    11.4
Andre Iguodala   11.3
Kobe Bryant      11.0
Baron Davis      10.7
Steve Nash       10.2
Dwight Howard     9.9
Jamal Crawford    9.7
Dwyane Wade       9.4
Thaddeus Young    9.2
Joel Anthony      9.1
Aaron Gray        9.1

Back to top
View user's profile Send private message
Display posts from previous:   
Post new topic   Reply to topic    APBRmetrics Forum Index -> General discussion All times are GMT - 5 Hours
Goto page 1, 2  Next
Page 1 of 2

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © 2001, 2005 phpBB Group