2015-16 Team win projections

[tarrazu]

Pretty sure my projections mixed up 'HOU' and 'PHX'.

[Mike G]

I would guess it's reversed, with the West having the better rookies (Towns, WCS), player movement being relatively flat (Monroe and Carroll stayed in the East, Batum and Hibbert mostly cancel), so the remaining Eastern improvement must be coming from player talent improvement.

It looks as flat as it could possibly be. Players totaling 35,000 minutes last year went from east to west; and an equal number from west to east.

The talent was also equal, with 60 eWins totaled by each group. In both cases, they totaled .82 eW/484 last year; and they both are hitting .74 eW/484 this year.
But the equivalence stops there. The west-to-east group has totaled 30% more minutes thus far, than the east-to-west players.

Of this latter group, 25 got over 10 mpg last year, and 15 of these are getting notably less in the West.

Code: Select all

. east to west   1415  mpg   1516  mpg
Gerald Henderson  Cha   29    Por   15
David West        Ind   29    SAS   15
Luc Mbah a Moute  Phl   29    LAC    5
Paul Pierce       Was   26    LAC   22
Lance Stephenson  Cha   26    LAC   20
Brandon Bass      Bos   24    LAL   15
Mirza Teletovic   Brk   22    Phx   11
Caron Butler      Det   21    Sac   11
Rasual Butler     Was   20    SAS   11
Quincy Acy        NYK   19    Sac   10
Chris Johnson     Mil   18    Uta    6
Cole Aldrich      NYK   16    LAC    5
Damjan Rudez      Ind   15    Min    5
Mike Miller       Cle   13    Den    4
Cory Jefferson    Brk   11    Phx    5

Some West teams have stockpiled players who can't all get the same minutes they're accustomed to getting.
In raw MPG, the E->W group avg'd 20 last year and 15 so far this year.
West-east movers were/are 20/20.

As an overall subpopulation, traded players tend to be below avg and also generally in decline. But they are still contributors.

[kmedved]

Pretty sure my projections mixed up 'HOU' and 'PHX'.

I have you at 29 wins for Phoenix and 57 for Houston. You sure?

[Statman]

Pretty sure my projections mixed up 'HOU' and 'PHX'.

I have you at 29 wins for Phoenix and 57 for Houston. You sure?

I believe he was joking, assuming those are the two teams his projections are off most either direction.

[tarrazu]

Pretty sure my projections mixed up 'HOU' and 'PHX'.

I have you at 29 wins for Phoenix and 57 for Houston. You sure?

I believe he was joking, assuming those are the two teams his projections are off most either direction.

Yes, sorry **insert sarcasm**. I believe I had the highest on HOU and lowest on PHX. Never liked the PHX projection but I was fine on HOU, you know when they don't attempt to start Harden and Lawson together.

[Statman]

Yes, sorry **insert sarcasm**. I believe I had the highest on HOU and lowest on PHX. Never liked the PHX projection but I was fine on HOU, you know when they don't attempt to start Harden and Lawson together.

Actually, if it makes you feel better, I had Houston a full 4 games better than you did, which is in part destroying my projections.

[Mike G]

Over in the 'improved' thread, we see Lawson and Harden at #1 and #2 biggest dropoffs from last year. Brewer is in the top (bottom) 20, and Ariza is close

[Dr Positivity]

Some Rockets efficiency (in ORTG) last season to this season

Harden 118 -> 104
Lawson 113 -> 91
Ariza 107 -> 95
Brewer 101 -> 80
Terry 112 -> 99
Beverley 107 -> 78

And the drop-off on D has been worse

[Mike G]

Team wins projected by current W + Pyth% * remaining games. No regression.
Absolute avg error from prediction. No exponent.

Code: Select all

GSW   74      Chi   48      Mil   32
SAS   67      Cha   46      Sac   31
Cle   60      Dal   46      Den   30
Mia   58      LAC   44      Was   28
Bos   57      Ind   42      Mem   27
Phx   55      NYK   41      LAL   24
Tor   54      Orl   40      Hou   23
Okl   52      Det   40      NOP   18
Uta   50      Min   37      Brk   18
Atl   49      Por   33      Phl   10


AJ   9.1      rsm   9.7     itca  10.3
DF   9.3      snd   9.8      DrP  10.4
Cal  9.5      tzu   9.9      MG   10.6
KF   9.5      fpli  9.9      BD   10.9
DSM  9.6      Crow  9.9      nr   10.9
bbs  9.6      Dan  10.1      EZ   11.3
km   9.6      yoop 10.3     taco  11.5

2015 Pyth is a distant last, at 12.5

[kmedved]

Yeah, the Rockets are basically everyone's nightmare point. Here's an extremely wide image which color codes the square errors (linked to avoid screwing with the view here): http://i.imgur.com/FbFb8L5.png

As you can see, the Rockets are also basically why the 15 Pyth + Regression model is doing so well right now. Pretty dramatically outperforming anyone else there.

[A Gravity Well]

What were your methodologies for picking games? Sum up player ratings times % of minutes for both teams, include HCA and profit?

[kmedved]

Basically, yes. I used minutes projections from a fantasy basketball site, took RPM, age-adjusted it, regressed everyone to the mean, then added an additional regression to the mean factor for players changing teams. This gave me team efficiency ratings. I then added in preseason results to adjust the team ratings further (which had a somewhat dramatic positive impact I think), and then used those to generate winning percentages. These were adjusted for travel distance (as the crow flies mileage), and back to backs as well.

I didn't do anything fancy like usage-based RPM adjustments, or RPM/BPM blends. The travel/rest stuff had minimal impact (under a win per team). Regressing players who changed teams to the mean has helped a lot so far, but I suspect may hurt me down the line.

[Mike G]

... regressed everyone to the mean, then added an additional regression to the mean factor for players changing teams...

Verry interesting. On the one hand, we could surmise that players are being moved for mostly good reasons -- not a good fit here, maybe a better fit there -- but it does seem as if there's almost always a period of adjustment, which may or may not resolve itself.

And "to the mean" would seem to suggest that really bad teams acquiring (and losing) mediocre players are expected to do better than these new players' rates would indicate?

[kmedved]

And "to the mean" would seem to suggest that really bad teams acquiring (and losing) mediocre players are expected to do better than these new players' rates would indicate?

Yes - the same coefficient is applied to all players changing teams. This drags down good players and improves bad players, so someone like Bargnani doesn't end up quite as much a detriment as you might otherwise expect. This doesn't just apply to bad teams however. David Lee's rating got pushed up by the same percent as Bargnani's.

[BasketDork]

Dead on with ATL, IND so far..closest with CHI (-1), POR (+2), LAL (+3)...far away with HOU (+33), NO (+26), PHX (-23) and BOS (-21). Within 5 wins on OKC, CLE, DAL, PHI as well. More than ten wins off on GSW, LAC, SA, MIA, MIL, MEM, WAS, TOR, CHA, DET,ORL, and BKN. It's a long season, we'll see where they end up. Should have gone higher on GSW, SA, MIA, TOR, CHA, DET, ORL, and especially NYK, lower on LAC, MIL, MEM, WAS, and BKN.