 |
APBRmetrics
The statistical revolution will not be televised.
|
| View previous topic :: View next topic |
| Author |
Message |
basketballvalue
Joined: 07 Mar 2006
Posts: 195
|
 Posted: Thu Feb 07, 2008 12:21 am Post subject: BasketballValue.com update: Adjusted +/- now available |
 |
|
All,
I'm pleased to let you know that basketballvalue.com has been updated to show adjusted +/- data in addition to the simple +/- data that has always been available. A brief article about the addition is available on 82games. The calculations follow the approach Dan has outlined, modeling players who are in the top 75% in minutes in the league. Each matchup observation is weighted by the number of possessions. As always, the data will be updated daily throughout the season.
I hope you find it useful and I'll be interested to hear your thoughts. As usual, there are a few surprises, both positive (Rafer Alston) and negative(Deron Williams, Andre Miller, Richard Hamilton). Generally, though, I do think the results are plausible.
Thanks,
Aaron
_________________
www.basketballvalue.com
Follow on Twitter |
|
| Back to top |
|
 |
Mike G
Joined: 14 Jan 2005
Posts: 3293
Location: Delphi, Indiana
|
| Posted: Thu Feb 07, 2008 7:14 am Post subject: |
|
|
Going in, I think, "This is the Holy Grail, at last". And there are still some questionable results.
Why is Dwight Howard's standard error about twice as large as everyone else's? If it's lineup instability, why isn't everyone for the Magic's as large? (Foyle's is also huge).
Al Jefferson is the Wolves' worst player, followed by Craig Smith. I'd thought they were the best.
The Celts only have 4 positive players -- Pierce, KG, TAllen, GDavis -- of 10 rated.
Jamison is carrying the Wiz's, with modest assistance from Caron.
Led by Millsap, all the Jazz frontline players -- even Okur -- are positive. All their backcourt players are losers; especially Deron and Harpring.
Nash carries the Suns; after him, only Bell and Hill are positive.
Gotta run: Worst players in the league? (among the top 224 in minutes) :
Al Jefferson, Brevin Knight, Cassell, Charlie V, Curry, Bobby Jackson, Yi, Hughes, Azubuike, Andre.
Also near the bottom: Deron, Gay, Yao, Calderon, Rondo, BGordon, Roy, Rip, Zach.
_________________
`
39% of all statistics are wrong |
|
| Back to top |
|
|
basketballvalue
Joined: 07 Mar 2006
Posts: 195
|
| Posted: Thu Feb 07, 2008 10:02 am Post subject: |
|
|
Mike,
Thanks for the thoughts, I appreciate it. A couple of quick replies:
| Mike G wrote: | Going in, I think, "This is the Holy Grail, at last". And there are still some questionable results.
|
Certainly true. I tried to be clear in the 82games article ("It is important to note that the adjusted +/- rating is not a “holy grail” statistic that perfectly captures each player’s overall value. The ratings reported here are limited by a number of factors. "). I think others such as Dan and Dave have expressed similar sentiments. As I mentioned in the conclusion, "adjusted +/- ratings can help NBA teams, their fans, and the media identify hidden gems, but the results must be one piece of a broader assessment of the player."
| Mike G wrote: |
Why is Dwight Howard's standard error about twice as large as everyone else's? If it's lineup instability, why isn't everyone for the Magic's as large? (Foyle's is also huge).
|
Good question, Steve and I noted this as well as we were drafting the article. I'm looking into it but I am not certain of the exact cause right now. I'll continue to investigate, but we ultiamtely decided that it shouldn't hold up publishing the article.
You also raised some good questions about individual teams. In some cases (e.g. Al Jefferson), I think the adjusted +/- results are similar to Overall rating. I'd welcome everyone's thoughts on these questions. You can download the full list off basketballvalue.com if you or anyone else wants to investigate. I haven't had a chance to evaluate the averages by position and see if they systematically rewards or hurt a position, for instance.
Thanks,
Aaron
_________________
www.basketballvalue.com
Follow on Twitter |
|
| Back to top |
|
|
Ben
Joined: 13 Jan 2005
Posts: 264
Location: Iowa City
|
| Posted: Thu Feb 07, 2008 11:43 am Post subject: |
|
|
Looking at the list, Steve Kerr must be using +/-!
Shaq 9.4
Marion -1.5
Banks -9.7 |
|
| Back to top |
|
|
basketballvalue
Joined: 07 Mar 2006
Posts: 195
|
| Posted: Thu Feb 07, 2008 12:20 pm Post subject: |
|
|
Indeed. It is remarkable that despite all the talk of Shaq's decline, his adjusted +/- is high and even his simple +/- is positive as well (+6.2). He's still having a positive impact (22nd by adjusted).
On the other hand, Marion has a strong simple +/-(+10.5), so it seems he's effectively getting penalized for playing with good players and often being replaced with good players. Still, it's a significant adjustment that I'm sure gets into the realm of questionable in some people's minds.
Thanks,
Aaron
_________________
www.basketballvalue.com
Follow on Twitter |
|
| Back to top |
|
|
DLew
Joined: 13 Nov 2006
Posts: 219
|
| Posted: Thu Feb 07, 2008 2:17 pm Post subject: |
|
|
| Nice work Aaron, just get ready for a lot of people who don't understand how incredibly noisy adjusted plus-minus is, especially for only half a season. |
|
| Back to top |
|
|
Mountain
Joined: 13 Mar 2007
Posts: 1527
|
| Posted: Thu Feb 07, 2008 2:34 pm Post subject: |
|
|
Thanks very much for adding current season adjusted +/- data to the broad public basketball community’s toolbox.
At this point I am set to use it as one tool, an enhancement over raw +/-, but still with caution, the amount of caution I now better see it deserves than I did 1-3 years ago.
i've stated before that going to "overall +/-" incorporating statistical plus / minus as the dominant part of the value formula and adjusted +/- as the tail makes good sense to me. The method can be approximated using past disclosures. But do you have any plans to go to that level at your site? |
|
| Back to top |
|
|
Eli W
Joined: 01 Feb 2005
Posts: 399
|
| Posted: Thu Feb 07, 2008 6:20 pm Post subject: |
|
|
Great stuff, Aaron.
Here's my quick adjusted +/- wishlist:
- Getting new and better coefficients for statistical +/-. The more public adjusted +/- data we have, the better weights we can calculate for boxscore stats. It would also be interesting to regress adjusted +/- not just onto the typical boxscore stats but also some more advanced player stats that have become available in the last few years (on 82games as well as elsewhere).
- For the site, I'd be interested in seeing each player's adjusted +/- broken down into offensive and defensive adjusted +/-.
- Calculating which players have played with the best/worst teammates and played against the best/worst opposing lineups. Presumably this info is used in calculating adjusted +/- (raw +/- plus teammate strength plus opposing lineup strength equals adjusted +/-).
_________________
Eli W. (formerly John Quincy)
CountTheBasket.com |
|
| Back to top |
|
|
cherokee_ACB
Joined: 22 Mar 2006
Posts: 157
|
| Posted: Fri Feb 08, 2008 3:11 am Post subject: |
|
|
| basketballvalue wrote: |
| Mike G wrote: |
Why is Dwight Howard's standard error about twice as large as everyone else's? If it's lineup instability, why isn't everyone for the Magic's as large? (Foyle's is also huge).
|
Good question, Steve and I noted this as well as we were drafting the article. I'm looking into it but I am not certain of the exact cause right now. I'll continue to investigate, but we ultimately decided that it shouldn't hold up publishing the article. |
Mi guess is that this is because Foyle and Howard have been just 8 minutes together on the court, and Orlando has only played 25 minutes without both of them. With such a small sample, it's pretty hard for the correlation to separate them from their teammates. It's the player clusters effect in action (see Cassell - Knight for another example of an one position cluster with high errors) |
|
| Back to top |
|
|
Analyze This
Joined: 17 May 2005
Posts: 364
|
| Posted: Sat Feb 09, 2008 10:13 am Post subject: |
|
|
| DLew wrote: | | Nice work Aaron, just get ready for a lot of people who don't understand how incredibly noisy adjusted plus-minus is, especially for only half a season. |
If it's "incredibly noisy", is the system fitted to value players after 1 season or 1/2 season? Take the New Orleans Hornets for example. Chris Paul (-2,24) and Peja Stojakovic(11,84). Noisy indeed. That does not pass the "laugh test" for me based on what I have seen from both players this season. It's not that I don't want to adjust what I expected from those 2 players, but it's so far out of line based on what I saw that the chance that's it's correct is very small.
And if the system is not fitted to value players with a very high degree of accuracy after only half or 1 season of data (not enough data) what is the use/value of all those adjusted +/- ratings that are on the site for this season at this moment?
How can I trust an adj +/- value for player x if it seems like an adj +/- rating for player y (for example Stojakovic in comparison with Paul) seems to be really off? Another rating can be also really off but that can be less clear if it does not stand out so much. If the adj +/-ratings seem ok (= what people expect. for example high ones for Bryant and James) than you can say aha the system works, if the ratings are a bit off you can say that those players are a bit over or underrated and the system shows this but if the +/-'s are so far out off line that it's not believable than you say ah but we still have not enough data, it's noisy. That's very easy. Those very strange ratings could also be a sign that there is something seriously wrong with the whole measuring system for all players.
And if(?) you need more than 1/2 or 1(?) season of data (or 2 or 3 or more seasons?) to wipe those Paul-Stojakovic errors(?) out (and are they out with double the amount of data?), is it not so that the system requires too much data and therefore is not fitted for the nba where you need faster results and can not wait years and years before you have results that have a very high level of accuracy?(Besides with other rules or another interpretation of rules can you just take the data from before and mix it with the data from after to get enough data or do you start all over?) And what if players move from team to team each season (that happens a lot in the nba), what is the influence of that on the amount of data you need before you have results with a high level of accuracy?
Show me that this system does a better job in measuring the value of a player than other systems and it is not just something that has been sold with a lot of expensive words to some nba teams but does not bring us a better way to measure the value of players. The only thing I see up till now(after a bit more than 1/2 season) is that some results are so out of line that I must wonder how accurate the system is for all players.
_________________
Where There's a WilT There's a Way |
|
| Back to top |
|
|
Mountain
Joined: 13 Mar 2007
Posts: 1527
|
| Posted: Mon Feb 11, 2008 5:17 pm Post subject: |
|
|
Using data from dougstats and Eli's previous spreadsheet formulas for statistical +/- (broken out to offense and defense) and pace adjustment I have posted a statistical +/- file in the yahoo apbr file group for this season to date.
A Statistical +/- value weight + some weight for the adjusted +/- figures provided at basketball value (probably at a fixed weight instead of variable as in the original study for simplicity) could produce an approximation of "overall +/-". I will do that for at least a sample now and perhaps more later. If anyone wants to do for the full league be my guest. (The adjusted +/- values are reported in last name order on pages at basketballvalue.com and could be matched up with this file with a bit of work.)
Using 20% weight for adjusted +/- (the non-minute weighted average of "alpha" in the original study was 19.3%) and adjusting the statistical +/- value from I believe their per 40 minute basis to per 48 and weighting it 80% I find these estimated "overall +/-" values for the top 12 on adjusted +/- as reported in the 82 games article:
10.0 James
8.3 Howard
7.8 Garnett
7.2 Nash
7.2 Bryant
6.1 Pierce
6.1 Bosh
5.9 Jamison
5.2 Nowitzki
4.9 Kidd
4.8 Boozer
2.4 West
Of course others from lower on that adjusted +/- list might find their way into a rank on this "overall +/-" list if they were included so this is not a top list merely the rank order of that dozen.
A brief look at a few values is not enough to evaluate the method but they are presented to give a sense what values the method produces. Overall +/- seems pretty good to me, better than raw or pure adjusted and morely better than pure individual stat linear weight formulas.
But if nobody wants to discuss... so be it
(file deleted)
Last edited by Mountain on Mon Mar 17, 2008 2:30 pm; edited 2 times in total |
|
| Back to top |
|
|
Mountain
Joined: 13 Mar 2007
Posts: 1527
|
| Posted: Tue Feb 12, 2008 5:03 pm Post subject: |
|
|
A few more named above
by this version of overall +/-
(to compare with impression created by pure adjusted +/-):
Shaq in Miami +3.7
Al Jefferson -2.5
Deron Williams -3.7
Yao 0.4
Calderon 4.0
Roy -2.0
Zach -0.9 |
|
| Back to top |
|
|
Serhat Ugur (hoopseng)
Joined: 13 Oct 2006
Posts: 181
Location: Basketball Research
|
| Posted: Tue Jul 21, 2009 10:31 am Post subject: |
|
|
Hey Aaron,
How about adding Player vs. Player comparison? Would be great to get the impacts in one table under available data sets.
| Code: |
PLAYER DATA SETS PLAYER
Hedo Turkoglu x.y 2009 Playoffs - Matt Barnes
Hedo Turkoglu c.d 2009-09 Regular Season f.g Matt Barnes
|
_________________
http://www.nbastuffer.com |
|
| Back to top |
|
|
THWilson
Joined: 19 Jul 2005
Posts: 164
Location: phoenix
|
| Posted: Mon Feb 08, 2010 11:20 am Post subject: |
|
|
Hi Aaron,
There is a problem with your calculation of two-year adjusted +/-.
http://basketballvalue.com/index.php Marcus Camby is listed at +573. |
|
| Back to top |
|
|
basketballvalue
Joined: 07 Mar 2006
Posts: 195
|
| Posted: Tue Feb 09, 2010 8:25 pm Post subject: |
|
|
Thanks for pointing this out, it's fixed now. Sorry it took a little time to identify the issue.
Thanks,
aaron
_________________
www.basketballvalue.com
Follow on Twitter |
|
| Back to top |
|
|
|
|
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
|
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
