Author Message
schtevie
Joined: 18 Apr 2005
Posts: 408
PostPosted: Wed Jul 02, 2008 1:03 pm Post subject: Celtics '08 pre(&post)diction etc. Reply with quote
Let me try to kick off the summer fun by initiating what could be a productive (and almost certainly amusing) discussion on the correctness of the assumptions behind the various the NBA prediction methods employed here and elsewhere.
The motivation is the curious case of the 2008 Boston Celtics. Quite simply, all methods not based on Adjusted +/- failed and failed spectacularly. To such a degree that there is no way to casually explain away the estimation errors. That this is so is, on its face, surprising. The players involved in the beefing up of the Cs were as known quantities as known quantities could be in this league. That their collective effort could (be so misconstrued should be a shot across the bow.
Eyeballing the average error, the consensus of those posting here was that the Celtics would win about 15 fewer games than they actually did (16 in a Pythagorean world) and this figure is actually biased low, given that KG and Ray Allen played fewer minutes and games than (I think) was generally anticipated (certainly, at least, in the case of Bob Cs simulation). And a further bias is that I don't think that any estimate (mine included) explicitly factored in the generally expected age-related diminution of the Big Three's productivity.
So, the question is what is the source of these estimation errors? I think it would be very useful if everyone who played the estimation game reran their Celtics estimate, using actual minutes played, then posted these along with their explanation(s) of the differential. My guess is that the common answer resides in the great statistical void of defense, but maybe there are other truths to be revealed.
So, in the spirit of such self-examination and reflection, let me begin with my own reappraisal. I had based my estimate on incomplete data. I didn't have the 2006-2007 Adj. +/- data at hand, and I didn't have the complete roster either. If I did, I would have presented the following estimates of the Celtics performance based on the reported year's Adjusted +/-:
2003 & 2004 avg.: 24.0
2005: 20.7
2006: 12.2
2007: 16.2
2008 actual: 11.3
In this light, my (actual) estimate based on (year ending) 2006 data looks pretty darn good. Were I to have used 2007 data, my guess would have been less impressive. But then again, fitting a trend would also have led to a very solid estimate.
A brief digression is relevant however on the 2007 datum.
On the one hand, perhaps the unexpected jump up (given the age-profile of the players) reflects the imprecision of the regression results (a known weakness of the approach). On the other hand, the 2007 data is derived from Steve Ilardi's 82games article. His approach then (as opposed to now) was to estimate Adjusted +/- on a 40 minutes per game basis, as opposed to per 100 possessions. To put these numbers to the common basis, I multiplied them, player by player, by approximately 1.37 (1.2 to get them from 40 to 48 minutes, then dividing by approximately .91 to reflect the actual pace of games). Perhaps I am making a scaling error. Perhaps there is something about the per minute regressions. (David Lewin had hinted last fall that his 2007 Adj. +/- numbers indicated that, indeed, time was not on the Big 3's side, and David is far too nice of a guy to intentionally mislead.) Or perhaps the 2007 estimates are correct but imprecise.
This aside, some other commentary is relevant. Given that the story of the Celtics is typically viewed through the prism of the Big 3, let's look at their collective Adj. +/- over these same years. Given 2008 minutes, this would be their estimated contribution to team success:
2003 & 2004 avg.: 34.5
2005: 31.1
2006: 20.0
2007: 29.6
2008 actual: 17.8
Comparing these numbers with the counterparts above, we see that indeed the 3 was Big, but that their collective skills, though still formidable, appear to be waning as expected (again, 2007 being an open question).
Peeling one more layer off the onion, we see that this deterioration is driven by the diminution of Ray Allen, who posted a -0.87 Adj. +/- last year, whereas Paul Pierce and KG have remained remarkably resilient. Will the wheels stay on the bus for another year, however? I don't know if I would bet on that, Ray's in particular, but a related thought in a moment.
A couple other observations about the Adjusted +/- estimate.
As I noted before, my guess is that the superiority of Adjusted +/- in this instance is largely attributable to the measurement of defense in general, and that of Kevin Garnett in particular. Using Eli W's recently posted Adj. +/- numbers, on average, three quarters of the Cs margin of superiority came from defense, and three quarters of that was attributable to KG. A higher fraction still in terms of games played. Phenomenal. By contrast, his contribution to offense was nil. (A similar relative contribution, by the way, as in 2003 and 2004, as estimated by Prof. Rosenbaum.) You miss this part of the story, and there is no hope of being on target.
Finally, looking forward rather than back, I hope and trust, as a nominal Celtics fan, that the contribution of James Posey is well understood. There has been too much hagiography written about his contribution to the team's success, and it was disturbing to see Danny Ainge being quoted words to the effect that signing him was very important to the Celtics future. (Then again, there is now expressed formal interest in Corey Magette, and I just can't imagine him taking the required championship discount, so perhaps all the talk is all just noise.)
James Posey may be a good spot-up three point shooter, a good locker room influence, and a good man defender at times, but according to +/- and contrary to advertisement, he is not the guy who does all the little things well. The last time he made a positive contribution to a team was 2003 & 2004 (an average +/- of 3.0) and alarmingly, last years contribution was by far his worst, dipping down to -7.25 which Eli W tells us was all negative effect on offense with no discernible effect on defense.
If Danny invests his MLE in an aging, below-average player, barring great advances in the Celtic youth, I fear that there will be no two-peat, what with the expected deterioration of the Big 3.
Pssst. Na-je-ra.
(I am sure that DeanO won't notice he is gone.)
I look forward to a lively discussion.
Back to top
View user's profile Send private message
thref23
Joined: 13 Aug 2007
Posts: 90
PostPosted: Wed Jul 02, 2008 3:27 pm Post subject: Re: Celtics '08 pre(&post)diction etc. Reply with quote
schtevie wrote:
James Posey may be a good spot-up three point shooter, a good locker room influence, and a good man defender at times, but according to +/- and contrary to advertisement, he is not the guy who does all the little things well.
As a C's fan....I would agree he may be overrated right now amongst our fans. By all indications he is, at least a little. I think he was underrated by our fans beginning of our season, and is now overrated perhaps.
But, its tough to measure the locker room impact, and he was played out of position @ PF many times last season...statistically we were better when he was @ SF I believe. And he was clutch when it mattered most.
Also, he is inconsistent. On some nights he might be nothing special,but he is almost never a liability, and on other nights he is extremely valuable on both ends. If depth grows further around him, the ability to be the man on any given night becomes more valuable.
I have mixed feelings about giving him the full MLE or close to it, especially considring his age, but I trust Danny. Only thing I know, we can't offer more than two years guaranteed because not only is that arguably stupid to begin with, it cuts into the caproom we have two offseasons from now.
I would be curious to get your opinions on Leon Powe, and whether you feel his actual value measures up to what his trade value may currently be. Of course he's an outstanding guy to have in the locker-room, and that can't mbe measured by any statistics, but by certain indications he wasn't that great this past season, he's not that young,and if some GM out there was willing to eat up the fact that he was a young big with a PER of 20+ and value him accordingly, then I think Danny has to consider options.
Back to top
View user's profile Send private message Send e-mail
jmethven
Joined: 16 May 2005
Posts: 51
PostPosted: Wed Jul 02, 2008 3:46 pm Post subject: Reply with quote
I'd be really curious (once the off-season transactions have cooled down a bit) to try and predict win-loss records for each team using adjusted plus-minus. I've already taken a stab at this on my blog, reacting to the recent signing of Baron Davis by the Clippers. My (very) rough estimation says that the Clippers, getting full, productive seasons from Davis and Brand, will have an efficiency margin of +3.45, putting them around 50 wins. I think this estimate has face validity at the least and I'm really curious to see how this method will do going forward.
For those who don't want to click, my projection looks like this:
Baron Davis 35.5 mpg, +7.02
Cuttino Mobley 37.0 mpg, -1.42
Al Thornton 27.3 mpg, -4.43
Elton Brand 38.3 mpg, +5.91
Chris Kaman 29.1 mpg, -2.00
Tim Thomas 21.7 mpg, -0.33
Brevin Knight 16.3 mpg, +1.18
Quinton Ross 16.1 mpg, -2.33
Eric Gordon 18.7 mpg, -2.85
Back to top
View user's profile Send private message
Ben
Joined: 13 Jan 2005
Posts: 264
Location: Iowa City
PostPosted: Wed Jul 02, 2008 4:36 pm Post subject: Reply with quote
The Lakers with Bynum and Gasol are an interesting candidate for this method.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3548
Location: Hendersonville, NC
PostPosted: Thu Jul 03, 2008 11:00 am Post subject: Re: Celtics '08 pre(&post)diction etc. Reply with quote
schtevie wrote:
... the guy who does all the little things well. ..
I assume among the little things that +/- picks up is 'enhancing teammate shooting'. How else to describe these improvements?
Code:
Celtics 2007 2008
Pierce,Paul .558 .579
Allen,Ray .556 .572
Garnett,Kevin .536 .572
Rondo,Rajon .462 .505
Perkins,Kendrick .508 .608
Posey,James .572 .574
House,Eddie .533 .532
Allen,Tony .581 .510
Davis, Glen --- .521
Powe,Leon .530 .601
Scalabrine,Brian .533 .421
Rebound rates, assists, steals, and blocks don't seem to show improvements. But turnovers do (per 36):
Code:
Celtics 2007 2008
Pierce,Paul 3.3 2.8
Allen,Ray 2.6 1.8
Garnett,Kevin 2.6 2.2
Rondo,Rajon 2.6 2.4
Perkins,Kendrick 2.8 2.5
Posey,James 1.0 1.3
House,Eddie .9 1.9
Allen,Tony 3.4 2.9
Davis, Glen 2.5
Powe,Leon 2.0 2.0
Scalabrine,Brian 1.5 1.9
These lists are in order of minutes played. All the starters improved notably in each list. Up to +.100 in effective shooting%, and from 69-90% the TO.
If +/- reliably predicts these offensive stats, and even more regarding defense, then that's wonderful. Presumably we can also anticipate ups and downs due to players' ages.
Here are my projected and actual (2008) minutes, eWins, and eW/1000 minutes (using 2007 rates when available, guessing for rookies). I didn't project Sam and PJ, so they have blanks.
Code:
Minutes eWins eW/1000
Celtics proj 2008 proj. 2008 2007 2008
Pierce,Paul 2774 2873 11.0 10.6 3.9 3.7
Allen,Ray 2628 2621 8.9 7.1 3.4 2.7
Garnett,Kevin 2849 2329 13.3 11.7 4.7 5.0
Rondo,Rajon 1976 2308 3.0 5.6 1.5 2.4
Perkins,Kendrick 1752 1912 1.9 4.3 1.1 2.3
Posey,James 1608 1816 2.1 3.3 1.3 1.8
House,Eddie 858 1481 1.4 3.1 1.6 2.1
Allen,Tony 1360 1377 3.2 2.1 2.4 1.5
Davis, Glen 1120 941 1.3 1.5 (1.2) 1.6
Powe,Leon 704 809 .9 3.2 1.2 4.0
Scalabrine,Brian 455 512 .0 .2 .0 .3
cassell,sam --- 297 --- .5 --- 1.8
brown,pj --- 208 --- .3 --- 1.4
Pollard,Scot 660 173 .3 .2 .4 1.2
Pruitt, Gabe 780 96 .4 .1 (.5) 1.3
Projected/actual eWins total 47.3/53.5 . By the formula xW = 2*eW - 41, expected wins were/are = 54/66 .
The final columns show actual player improvement. The Allens were big losers. Perkins, Rondo, Posey, House, and Powe cleared expectations by +10 eW; netting +20 actual wins and overwhelming the few dropoffs.
Several of us expected many replacement-quality minutes, and in fact there were virtually none.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
supersub15
Joined: 21 Sep 2006
Posts: 273
PostPosted: Thu Jul 03, 2008 11:20 am Post subject: Reply with quote
jmethven wrote:
I'd be really curious (once the off-season transactions have cooled down a bit) to try and predict win-loss records for each team using adjusted plus-minus. I've already taken a stab at this on my blog, reacting to the recent signing of Baron Davis by the Clippers. My (very) rough estimation says that the Clippers, getting full, productive seasons from Davis and Brand, will have an efficiency margin of +3.45, putting them around 50 wins. I think this estimate has face validity at the least and I'm really curious to see how this method will do going forward.
For those who don't want to click, my projection looks like this:
Baron Davis 35.5 mpg, +7.02
Cuttino Mobley 37.0 mpg, -1.42
Al Thornton 27.3 mpg, -4.43
Elton Brand 38.3 mpg, +5.91
Chris Kaman 29.1 mpg, -2.00
Tim Thomas 21.7 mpg, -0.33
Brevin Knight 16.3 mpg, +1.18
Quinton Ross 16.1 mpg, -2.33
Eric Gordon 18.7 mpg, -2.85
How did you reach the +3.45 number from the posted P/M? I couldn't figure it out. Thanks.
Back to top
View user's profile Send private message
jmethven
Joined: 16 May 2005
Posts: 51
PostPosted: Thu Jul 03, 2008 11:55 am Post subject: Reply with quote
supersub15 wrote:
How did you reach the +3.45 number from the posted P/M? I couldn't figure it out. Thanks.
I just weighted by minutes played - for example, Baron Davis gets a weighted value of (35.5/48)*7.02, yielding a value of 5.19. Each player's weighted value is then added up to get the team score.
I'm not sure if this is the best way to do it, although it seems logical. One obvious limitation is that according to this method, a starting lineup of Steve Nash, Kobe Bryant, LeBron James, Kevin Garnett and Dwight Howard would have an efficiency margin of +48.77 (based on 2008 regular season values), which is outrageous.
Back to top
View user's profile Send private message
dquinn1575
Joined: 03 Jun 2008
Posts: 12
PostPosted: Thu Jul 03, 2008 11:58 am Post subject: Reply with quote
Take each minutes played by rate per minute (for Davis 35.5 * 7.02 = 249.21) Sum the total to get 165.149 - Take that divided by 48 minutes to get 3.44Baron Davis 35.5 mpg, +7.02
Cuttino Mobley 37.0 mpg, -1.42
Al Thornton 27.3 mpg, -4.43
Elton Brand 38.3 mpg, +5.91
Chris Kaman 29.1 mpg, -2.00
Tim Thomas 21.7 mpg, -0.33
Brevin Knight 16.3 mpg, +1.18
Quinton Ross 16.1 mpg, -2.33
Eric Gordon 18.7 mpg, -2.85
How did you reach the +3.45 number from the posted P/M? I couldn't figure it out. Thanks.
I'm thinking this method would do well to compare teams, but you are omitting minutes outside of the top 9 players - for example for the 07-08 Clippers that would equal about 62 mpg ~ 15% of all minutes played. I would think you should have a baseline of a certain amount of minutes played with low contribution to reflect the remainder of the bench.
If you get 85% at 3.44 and 15% at -2, you get 2.62 for example.
Back to top
View user's profile Send private message
supersub15
Joined: 21 Sep 2006
Posts: 273
PostPosted: Thu Jul 03, 2008 12:31 pm Post subject: Reply with quote
Thanks for the answers. Now, I'm checking the Adjusted P/M for the Clippers here, and it shows different results for Mobley, Thornton, etc. Is your source different or am I missing something concerning the minutes?
Back to top
View user's profile Send private message
jmethven
Joined: 16 May 2005
Posts: 51
PostPosted: Thu Jul 03, 2008 12:48 pm Post subject: Reply with quote
dquinn1575 wrote:
I'm thinking this method would do well to compare teams, but you are omitting minutes outside of the top 9 players - for example for the 07-08 Clippers that would equal about 62 mpg ~ 15% of all minutes played. I would think you should have a baseline of a certain amount of minutes played with low contribution to reflect the remainder of the bench.
If you get 85% at 3.44 and 15% at -2, you get 2.62 for example.
Yeah, that's a smart way to do it without having to try to invent ratings for players who didn't play many minutes. I should have mentioned that my estimate would be skewed upwards by not including the end of the bench. On the other hand, you could look at the 9-man rotation as the Clippers' playoff rotation although of course every other team benefits from whittling down their rotation as well.
supersub15 wrote:
Thanks for the answers. Now, I'm checking the Adjusted P/M for the Clippers here, and it shows different results for Mobley, Thornton, etc. Is your source different or am I missing something concerning the minutes?
Sorry, I went into more detail on my blog post and not as much here. What I did was to go on the 82games archives and dig up the plus minuses for 2007 and 2006. Then I weighted the rating towards the most recent season - 3 parts 2008, 2 parts 2007, and 1 part 2006. Looking at the ratings again, it might have made more sense to just include the 2007 ratings since they were already weighted by Ilardi to include 2006 results to some extent.
Back to top
View user's profile Send private message
supersub15
Joined: 21 Sep 2006
Posts: 273
PostPosted: Fri Jul 04, 2008 6:20 am Post subject: Reply with quote
According to the Adjusted P/M, a team composed of Chris Paul (-0.06), Jose Calderon (-5.47), Ray Allen (-0.87), Rip Hamilton (-2.57) Richard Jefferson (-6.16), Carmello Anthony (-4.79), Udonis Haslem (-8.61), Emeka Okafor (-4.76) and Yao Ming (-0.82) would be a losing team. Is this a limitation of this tool?
Back to top
View user's profile Send private message
Ilardi
Joined: 15 May 2008
Posts: 262
Location: Lawrence, KS
PostPosted: Fri Jul 04, 2008 9:26 am Post subject: Re: Celtics '08 pre(&post)diction etc. Reply with quote
schtevie wrote:
Let me try to kick off the summer fun by initiating what could be a productive (and almost certainly amusing) discussion on the correctness of the assumptions behind the various the NBA prediction methods employed here and elsewhere.
The motivation is the curious case of the 2008 Boston Celtics. Quite simply, all methods not based on Adjusted +/- failed and failed spectacularly. To such a degree that there is no way to casually explain away the estimation errors. That this is so is, on its face, surprising. The players involved in the beefing up of the Cs were as known quantities as known quantities could be in this league. That their collective effort could (be so misconstrued should be a shot across the bow.
For what it's worth, I had a correspondence with Dan Rosenbaum and Dave Lewin last November on this very topic (see excerpt from my email below), but had trouble believing my own numbers on how good the Celtics would be. My projection for the 07-08 Celtics, based on 06-07 adjusted +/- data, was a per-game point differential of +12.7, good for roughly 71 wins. This compared with a forecast using David Berri's win score ratings of 54 total wins (using the same number of projected minutes per player).
I came pretty darned close to the team's actual point differential (+10.3), and would have hit it right on the nose had I not slightly overestimated projected minutes played for KG and Pierce.
November 21, 2007
Dave and Dan,
I just ran a quick, rough projection of Boston's projected 2007-2008 wins under WP versus adj +/- (using my published 06-07 numbers with a bit of age-related decline for 'Big 3') and came up with 71 expected wins (pythagorean) under Adjusted Plus Minus and 54 with WP. Frankly, I'm not sure which estimate looks more plausible . . . if forced to guess, I'd probably go with Celtics winning about 60 this year, which would put Berri's number a little closer. Obviously, if the Celtics actually win 65-70 (as they're on pace to do) then they become a nice 'poster child' case for Adjusted Plus-Minus, but I suppose I won't be holding my breath. It'll be interesting to try identifying other teams that yield highly divergent predictions this year between the two models . . .
Back to top
View user's profile Send private message
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Fri Jul 04, 2008 11:07 am Post subject: Reply with quote
I assumed Garnett, Allen and Posey were important to improving the geometry of the offensive threat and thus to helping teammate shooting efficiency but the case is mixed in the player pairs. Playing beside Garnett helped the young bigs but really others. Allen helped Pierce a little. Posey on court was associated with decline in teammate FG%s in most cases and the bigs often by large amounts.
Garnett and Thibodeau enhanced the defensive game. Ainge and Rivers knew who knows defense and went out and got them.
They also showed attention to balance.
According to Eli's article, Rondo and Allen made their adjusted +/- contribution on offense. Pierce was the 2-way helpful core piece. Garnett and Perkins made their contribution on the defensive side and playing inside had the most opportunity to do so. Pretty straightforward, solid design. Garnett gave enough post player offense that is wasn't an impediment.
The Spurs were built to be strong on defense inside and have offensive firepower on perimeter too but they managed to have 3 2way helpful players.
The Pistons mixed up the perimeter offense / interior defense design some- Billups and McDyess provided primarily positive offensive adjusted +/- while Hamilton and Prince primarily contributed on defensive adjusted +/- and Wallace was the 2way helpful core.
For the Lakers adjusted +/- suggests Fisher was a dual liability but mainly on defense (age/weight). Kobe was a very strong offensive player with only a small positive defensive impact and not really-right now- 2 way core. Gasol balanced him and was the opposite with surprisingly a modest positive adjusted defensive +/- impact and just a small one on offense.
Radmanovic and Vujacic were helpful on offense but Vujacic was a much bigger and harder to justify defensive liability. Odom was a positive contributor on defense but nearly gave it all back on offense. Turiaf mildly helpful on defense. Walton mildly negative on offense. More of a jumble than a simple / good balance like these other leading contenders. But Bynum was a modest 2 way positive player and maybe will become more.
Can you win a title without a clearcut 2-way strong core piece?
You probably can but when was the last time? 80s Pistons or did Joe Dumars play that role? If you need one, adjusted +/- offensive/defensive splits say that the Hornets are "relying" on Peja. The Rockets' closest was Battier. Utah doesn't have a strong 2way player - unless you count marginal Ronnie Price. Denver had Iverson and Najera as dual positive contributors but strong on one side and just above +1 on the other side like Bryant/Gasol. Phoenix doesn't. Dallas has Nowitski / Kidd / Howard. The Cavs don't have a 2way player. The Magic clearly do with Howard. Toronto has Bosh and Moon. Washington has Jamison. Philly has Thaddeus Young.
Last edited by Mountain on Sat Jul 05, 2008 9:05 am; edited 8 times in total
Back to top
View user's profile Send private message
anarcholis
Joined: 12 Jun 2007
Posts: 19
PostPosted: Fri Jul 04, 2008 10:49 pm Post subject: Reply with quote
Quote:
According to the Adjusted P/M, a team composed of Chris Paul (-0.06), Jose Calderon (-5.47), Ray Allen (-0.87), Rip Hamilton (-2.57) Richard Jefferson (-6.16), Carmello Anthony (-4.79), Udonis Haslem (-8.61), Emeka Okafor (-4.76) and Yao Ming (-0.82) would be a losing team. Is this a limitation of this tool?
I'm not sure this team would defend well enough to be very good. Other than Okafor and Haslem, I'm not sure any of these players are considered above average defenders. I think that the great advantage of adjusted plus minus is that it takes defense into account.
Back to top
View user's profile Send private message
bchaikin
Joined: 27 Jan 2005
Posts: 681
Location: cleveland, ohio
PostPosted: Sat Jul 05, 2008 1:10 am Post subject: Reply with quote
Other than Okafor and Haslem, I'm not sure any of these players are considered above average defenders.
over the past 5 seasons the houston rockets have averaged the 4th best/lowest points allowed per team defensive possession in the league, they have the lowest FG% allowed (42.5%), the 2nd lowest eFG% allowed (46.2%), and the lowest 2pt FG% allowed (44.2%)...
yao ming has played the most minutes of any rockets players during this time, has almost double the defensive rebounds of any other rockets player, and over twice as many blocked shots as any other rockets player. if you look at ming's counterpart production at http://www.82games.com from 03-04 to 07-08, you'll see rockets opposing Cs have done poorly when ming was on the floor...
i'd say all that puts yao ming in some elite defensive territory...
page 2
Author Message
Mike G
Joined: 14 Jan 2005
Posts: 3629
Location: Hendersonville, NC
PostPosted: Sat Jul 05, 2008 8:20 am Post subject: Reply with quote
Well, we were briefly back on the Celtics.
According to Sagarin, the Celts weren't hugely better than LA (1.2 ppg), Utah (2.8), NO (3.1), SA (3.0). They did, however, play in the East. In the West, they'd have gotten around 61 wins. So I don't know how much effort we should go into explaining the overachievement to 66 W (or 67 pythagorean) when in a balanced league, they'd be around 63-64.
Which players are doing the 'little things' that don't show up in the boxscore, and which are doing the little things and/or showing serious boxscore improvements?
At least 5 Celtics showed massive individual improvements, and only a couple dropped off noticeably. Both rookies did better than expected, and both were hardly needed. A couple of veterans (PJ and Sam) joined late. How many of these phenomena are predictable thru +/- analysis?
How much of the Celtics' success can be attributed to the players' known quantities, and how much was individual improvement by several of them?
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Sat Jul 05, 2008 11:43 am Post subject: Reply with quote
Much of the Celtics' success can be attributed to their very productive big 5 lineup (almost +20 pts per 100 possessions each). They used it the 3rd most in the league and it had the best net rating of any lineup used in the league over 200 minutes. It represented less than 25% of team minutes but nearly 45% of the team's +/- edge. The average edge produced per minute by all other lineups was only about 40% of the big 5. Theoretically there would be room for them to use it far more, but even 25-50% more could give them another point or two differential if performance held that net rating.
Extrapolating from the lineup performance stats if the Hornets or Pistons used their big 5s about 40 to 60% more (or stated a less intimidating way, just used them 5-7 minutes more per game) they could add about a point to their differential.
This suggestion isn't simple / automatic of course (match-ups matter but I doubt that all of the opportunities to apply a team's best lineup positively have been exhausted; their usage is relatively low at no more than 10-14 minutes a game and with many team far less) but I think increased use of the best performing of the most used lineups significantly is a pretty good strategy to try. If it works less or stops working you recognize that and adjust.
Not every competitive team would see big gains by cranking the volume on the most used lineup because the performance and performance gap isn't as large as in these cases. Many would have to sort among moderate used lineups or prospect among small use lineups to find something new and powerful and accumulate gain from a set of lineup usage changes. But in these 3 cases the teams appear to have the potential for noticeable gains simply by using their most used / trusted lineup (all of it) at a significantly greater but feasible rate. (I don't know if the data shows on average lineups beyond the smallest samples getting more efficient with more use but I'd guess unit returns or improvement more likely than decline.)
Last edited by Mountain on Sun Jul 06, 2008 11:24 am; edited 1 time in total
Back to top
View user's profile Send private message
schtevie
Joined: 18 Apr 2005
Posts: 414
PostPosted: Sat Jul 05, 2008 2:40 pm Post subject: Reply with quote
A few thoughts on the conversation...
About the Celtics:
One way to minimize the apparent failure of non-Adj. +/- prediction methods is to argue that in fact the Celtics were not as good as they were. I don't think that dog will hunt, both in the sense that knocking the Cs down to 63-64 wins doesn't really redeem the alternatives and also because the argument is false.
Being off by 13.5 vs. 15.5 wins is not that big of a correction, relative to the fact that the roster changes involved such well-known players. And recall my observation, where I wait to stand corrected, that I don't think anyone explicitly took into account the expected (and realized) age-related decline of the Big 3 in their projections. As I understand things, the true error is a lot closer to 20 games than 10.
And then there is the issue that the argument, at least its relationship to the Adj. +/- prediction, is false. Adj. +/- is not biased by East vs. West schedule strength differentials (or at least I don't think so).
What needs to be explained is why all other predictive methods were in a cluster, way below the realized target.
Another issue raised but where there appears to be some confusion is the relative contributions of the Big 3 vs. all others. In the original string, Hollinger bad-mouthed the Celtics bench, prior to the arrival of Posey, as being the worst in the league. Imagine! Maybe so by his methods, but not by Adj. +/- (and recall that Posey turned out to be the weakest link therein.)
Let me replay the numbers. The following data take actual 2008 minutes and weight the Bench (Roster - Big 3 - Posey and Perkins) by various years' Adj. +/- (as previously described). For years where current players were not in the league, I simply take their first years' numbers for backcasting. Similarly, for players who played insufficient minutes (e.g. Pollard) I take the last available Adj. +/- on record. What this suggests is that the bench never was weak. To the contrary....
2003 & 2004 avg.: Big 3 = 21.86, Others = 2.09 of which Bench = 2.70
2005: Big 3 = 21.13, Others = -0.40 of which Bench = 4.04
2006: Big 3 = 12.78, Others = -0.62 of which Bench = 1.98
2007: Big 3 = 19.02, Others = -2.46 of which Bench = 0.04
2008 Actual: Big 3 = 11.61, Others = -1.44 of which Bench = 3.56
Note the trend. A fall and rise in the contribution of Others and Bench, reflecting the aging of the veterans and the improvement of youth.
But also note the overarching fact, a bench which gains you rather than loses you points (in each and every yearly estimate) is very impressive achievement. Period. PER, but not it alone, is missing something.
Regarding another issue raised in the string:
supersub15 lays down a challenge to an Adj.+/- approach by listing an apparently horrible line-up, but one which strains credulity. The general reply is that it is not appropriate to cherry-pick.
It is well known (?) that Adj. +/- estimates tend to be imprecise. So, picking one year's (possible) outliers is not a fair criticism of the approach employed in the instance of the 2008 Cs, where there was an average of many player-year's data as a basis. Similarly, with the proposed line-up it is useful and possible to take a longer-term view.
Ray Allen has already been discussed, but the fact of the matter is that he appears to be in decline (I would not be surprised - having no projections of expected age profiles to refer to - if he were to bounce back a bit this year, but I wouldn't bet on it). Chris Paul is an interesting case, as his stock has risen in his three year career as his Adj. +/- has declined. This year's offense/defense breakdown is weird, suggesting that his prodigious skills in the former category are offset by atrocious performance in the latter. Time will tell. And time is more suggestive with Yao Ming. Since 2003, his career shows an Adj.+/- average just over 5, with a strong peak of 13.63 in 2006. Was he an elite center? Sure. Only Brad Miller (6.65) and Shaquille O'Neal (5.76) top his six year average. Will he ever be again? I wouldn't bet either way, but I would be surprised if his "true" value weren't still positive.
Relatedly, I certainly wouldn't bet on Yao Ming being an elite defensive center. Eli W's breakout shows him with a defensive Adj. +/- of 2.19 compared to a center average of 2.5. This is just one year's datum, imprecisely estimated and all, but this is the kind of estimate (accumulated across many, many players) that probably explains most of Bob C's simulation error.
As for the other players on the proposed roster, I haven't done the career summaries, but I would be surprised if an informed reading of the data suggested a notional elite team in 2009.
Back to top
View user's profile Send private message
Neil Paine
Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA
PostPosted: Sat Jul 05, 2008 3:30 pm Post subject: Reply with quote
Well, speaking for myself, much of the error came on the defensive side of the ball... I eventually settled on a projection of 108.3 ORtg/104.9 DRtg; in reality, the numbers were 110.2/98.9. The emergence of Rondo offensively and underestimating the Big 3 in general (after generic age adjustments) explains the ORtg difference, but defensively only something non-boxscore like Adj +/- could have probably foreseen the impact of adding KG to a team with a 106.9 rating the year before. Sadly, individual defensive ratings just don't tell much of the story, and they don't transfer well from one team to another.
Back to top
View user's profile Send private message Visit poster's website
Mike G
Joined: 14 Jan 2005
Posts: 3629
Location: Hendersonville, NC
PostPosted: Sun Jul 06, 2008 9:22 am Post subject: Reply with quote
schtevie wrote:
...
What needs to be explained is why all other predictive methods were in a cluster, way below the realized target....
.
I guess a sample of 1 is fairly small, and you'd have to run predictions on other teams which had large roster changes. The one outlier prediction will sometimes be closer than the rest, and other times it will do worse.
My own 'productivity-based' system predicted 59 wins in a best-case (for the Big3) scenario, in a .500 conference, absent any great help from the bench. The Big3 didn't meet that high standard, but the rest of the team certainly overachieved, productively.
I don't recall anyone predicting the East would go 43% vs the West, yet again. Of course a given team will win more vs inferior competition.
If the imbalance had gone the other way, and the Celts win but 61, then 59 looks pretty good. But then I'd have been 'right' for the wrong reasons.
So until we see a second, third, etc, examples of +/- outperforming other prediction methods, I'll accept that for now, it's worked once; luck falls where it will. At the same time, I hope it's the right track.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
eyriq
Joined: 04 Jun 2008
Posts: 54
Location: Orlando
PostPosted: Sun Jul 06, 2008 9:35 pm Post subject: Reply with quote
jmethven wrote:
I just weighted by minutes played - for example, Baron Davis gets a weighted value of (35.5/4Cool*7.02, yielding a value of 5.19. Each player's weighted value is then added up to get the team score.
I'm not sure if this is the best way to do it, although it seems logical. One obvious limitation is that according to this method, a starting lineup of Steve Nash, Kobe Bryant, LeBron James, Kevin Garnett and Dwight Howard would have an efficiency margin of +48.77 (based on 2008 regular season values), which is outrageous.
Hey, this methodology really intrigued me (simple and straightforward) and I am wondering if a similar application can be applied to Basketball-Reference's WS metric, as I'd like to find some kind of universal application that can be transferred over to another team without the dependence on the current teams performance. For instance with Maggette, he has a pretty crappy WS for a player of his talents, but his team was horrendous and I think that is clouding the picture. Also, his Adjusted Plus-Minus seems to come out like a rose, as does his PER, ect. So can you or anyone shed some light on how to apply WS to a transferable metric (something almost like WP, but with an accessible database) for evaluating players impact on a new team?
I thought that maybe going by the percentage of a teams total WS that a player accounted for would be something to work off of, in which case Maggette looks great. But how do you apply that to another team? That really only gives you a ranking scale and I am looking for something that can tell me what a players "WS Value" is, like Howard is worth so many WS a seaon, regardless of team, sort of thing. Anyway, I am new to this stuff completely and have been banging this question around in my head, so any help would be great.
Back to top
View user's profile Send private message
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Sun Jul 06, 2008 9:51 pm Post subject: Reply with quote
Adjusted plus/minus beats many other metrics by fully including defense. Winshares does too, though defensive winshares are based on defensive rating and may not capture individual defensive contribution as well as the play by play based adjusted plus/minus method.
I wonder if offensive winshares transfer more consistently than defensive winshares. Would be interesting to see a study of 50 or 100+ players who changed locations.
Maybe the way to predict winshares in new location would be by expecting offensive winshares to largely transfer (but with some adjustment for improved or declining offensive environment) and splitting defensive rating into player stop rating and largely transfer that and making the team defense component mainly based on the new team with perhaps some share of the old (relative to the new) to reflect the value of the individual arriving.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3629
Location: Hendersonville, NC
PostPosted: Mon Jul 07, 2008 9:20 am Post subject: Reply with quote
The Clipps won 23 games, and presumably their players total 23 WS. You might first estimate a conversion factor, yielding "equivalent WS for a 41-41 team": -- EDITED formula --
eWS = WS*((W+41)/2)/W = WS*((23+41)/2)/23 = WS*1.391
This suggests Maggette's WS are 39% undervalued, if he should turn up on a random team. Once you have assembled a hypothetical 'team', you may estimate the team's wins by allotting minutes, multiplying by eWS/min., and summing their players' new eWS; then unconvert by the same formula:
W = eWS*2 - 41
Even better, perhaps, first convert WS to 'pythagorean WS' by the factor pW/W. Some teams are better than their W-L record, others weaker.
_________________
`
36% of all statistics are wrong
Last edited by Mike G on Tue Jul 22, 2008 9:00 am; edited 1 time in total
Back to top
View user's profile Send private message Send e-mail
Harold Almonte
Joined: 04 Aug 2006
Posts: 616
PostPosted: Mon Jul 07, 2008 11:14 am Post subject: Reply with quote
anarcholis wrote:
Quote:
Quote:
According to the Adjusted P/M, a team composed of Chris Paul (-0.06), Jose Calderon (-5.47), Ray Allen (-0.87), Rip Hamilton (-2.57) Richard Jefferson (-6.16), Carmello Anthony (-4.79), Udonis Haslem (-8.61), Emeka Okafor (-4.76) and Yao Ming (-0.82) would be a losing team. Is this a limitation of this tool?
I'm not sure this team would defend well enough to be very good. Other than Okafor and Haslem, I'm not sure any of these players are considered above average defenders. I think that the great advantage of adjusted plus minus is that it takes defense into account.
I think this team will defend enough for not to be a losing team. One of the problems of plus/minus (I think) is that the players's OFF numbers are built from their own replacement players (context), and not adjusted to a league average replacement position player (or lineup), to equalize everybody (something like MikeG is doing above). Then those numbers are for their contexts, and are as predictive as they remain with their own replacements, and with their own lineups in a less proportion.
Last edited by Harold Almonte on Mon Jul 07, 2008 11:52 am; edited 2 times in total
Back to top
View user's profile Send private message
eyriq
Joined: 04 Jun 2008
Posts: 54
Location: Orlando
PostPosted: Mon Jul 07, 2008 11:37 am Post subject: Reply with quote
Mike G and Mountain, thanks for the input, it was exactly what I was looking for!
Edit: Lol, ok Mike, how about a tutorial for that formula? Or specifically, what function does "*" relate too? I'm searching my old college stat book but figured I'd ask (maths never been a strong suite for me). My first instinct was that it was a multiplier but that did not seem right after putting it in practice. Confused
Back to top
View user's profile Send private message
schtevie
Joined: 18 Apr 2005
Posts: 414
PostPosted: Mon Jul 07, 2008 12:11 pm Post subject: Reply with quote
Mike, yes, a sample of one is small, but in the instance, the results were highly suggestive.
For those who might wish to fine tune their prediction models before next year, let me propose an experiment, the results of which might prove very useful.
To see whether Adj.+/- can be used to improve the model's predictive powers, take a season's worth of prediction errors, say 2006-2007, as opposed to the year just finished, and see the degree to which the these correlate with the sum of the 2007-2008 roster Adj. +/- statistics.
I would propose using Eli W's data as the defensive and offensive components can be looked at independently. My guess is that the sum of player offensive Adj. +/- will be much less correlated with the error than the defensive component.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3629
Location: Hendersonville, NC
PostPosted: Mon Jul 07, 2008 1:42 pm Post subject: Reply with quote
eyriq wrote:
...what function does "*" relate to? ..
On your keyboard, look to the number pad on the right. Note the keys marked / * - +
* of course is multiplication.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
eyriq
Joined: 04 Jun 2008
Posts: 54
Location: Orlando
PostPosted: Tue Jul 08, 2008 1:43 pm Post subject: Reply with quote
Mike G wrote:
eyriq wrote:
...what function does "*" relate to? ..
On your keyboard, look to the number pad on the right. Note the keys marked / * - +
* of course is multiplication.
Err, of course! Embarassed
I still can't figure out the formula though, but never mind that I guess.
What I would like to know is how to get projected wins from scoring margin. I like the idea of compiling +- rankings and adjusting for expected minutes played and then computing a eW/L record as was laid out earlier in this thread.
My main goal is to be able to make sense of all the different metrics out there in the internet, like PER, WS, Adjusted +-, and whatever else I am missing, and make sense of it for myself instead of having to rely on others to write an article about its different applications.
Edit: And obviously the main application that I'm focused on is carrying over any applicable metrics to different teams to predict wins, exactly like you guys are doing with your personal models, but I want (don't have a choice really) to do it with publicly accessible models.
Back to top
View user's profile Send private message
schtevie
Joined: 18 Apr 2005
Posts: 414
PostPosted: Sat Jul 12, 2008 8:16 am Post subject: Reply with quote
Well, today began with a scanning of the headlines and, from my perspective, the dismal news (for the Celtics) that Eduardo Najera is signing with the Nets for a paltry $12 million(ish) over four years.
The evidence strongly suggests that Adjusted +/- has little currency in the Celtics organization. Brain-typing must be (re?)ascendent. Najera was the clear bargain of the unrestricted free agents in the "6' 8", good, tough defender, good locker room guy, who can shoot the three". Compared to Posey, Najera simply has dominated. Over the last commonly played five years in the league (eliminating 2005 when Najera was injured) Najera had a 3.6 points per 100 possessions is his average margin of superiority. That is a lot. A lot, a lot. (And this is an low estimate in that 2005 was a dismal year for Posey, surpassed only by his most recent).
A one-off oversight you say? Then why isn't Tony Allen already signed? He too is a great +/- guy; perhaps unsurpassed in his player type. He has averaged a mere 5.2 since 2004, never having a negative year, and he is entering his demographic prime. Unless there is secret medical knowledge behind the decision, I cannot understand drafting Tony Allen types in lieu of the real deal. Possibly building for the future at the direct expense of winning another championship next year? This is simply a very odd time preference.
My thought before last year's season was that the Celtics could expect about 1.75 championships out of the Big 3 due to them being on the wrong side of 30. Barring unexpected good health and a Ray Allen renaissance, I fear the Celtics are one and done.
We shall see what they do.
Back to top
View user's profile Send private message
Ryan J. Parker
Joined: 23 Mar 2007
Posts: 711
Location: Raleigh, NC
PostPosted: Sat Jul 12, 2008 10:33 am Post subject: Reply with quote
Adjust +/- doesn't paint the whole picture. Unless we are mikez and know everything that goes into the Celtics' decisions making process then we'll always be able to find some metric somewhere that says their boneheads.
Clearly they're not boneheads. Laughing