2015-16 Team win projections

[Mike G]

You might think so. But if you knew Dellavedova was going to play a lot more minutes than expected, and that Mo Williams and Shumpert were going to play a lot less, you might not expect a net gain for the Cavs.
We now know that Delly is playing a lot better than last year, and those other guys are a lot worse. Knowing both their minutes and their improvement, you get the net effect on the team.

There don't seem to be a lot of major injuries this year. For the most part, player minutes are up or down due to changes in player productivity. Every player for the Rockets (save Capela) is just worse than they were last season. You can't give fewer minutes to everyone, so they're doomed to decline. Unpredictably so.

[Mike G]

Code: Select all

.   average errors              sq rt of avg sq err
AJ   5.24    MG   6.53        AJ   6.42    yoop 7.93
KF   5.27    snd  6.57        Cal  6.43    MG   7.99
km   5.35    BD   6.66        km   6.47    BD   8.00
bbs  5.39    itca 6.70        DF   6.69    snd  8.05
DF   5.41    yoop 6.80        bbs  6.83    Dan  8.51
Cal  5.45    Dan  6.90        KF   7.01    itca 8.52
tzu  5.49    DrP  6.99        DSM  7.12    EZ   8.60
DSM  5.69    nr   7.12        rsm  7.12    nr   8.66
Crow 6.01    EZ   7.38        tzu  7.24    DrP  9.14
rsm  6.04    taco 7.85        Crow 7.44    taco 9.40
fpli 6.05    15py 8.96        fpli 7.47    15py 9.89

Almost all are looking worse these days. Reason seems to be that there's less chance that teams will regress to their expectations: GSW, SAS, Cha, Tor really are this good and keep winning. Hou, NOP, Phx, Phl really are that bad.

[Mike G]

With 7 to 9 games left:

Code: Select all

average errors
AJ   5.55      DSM  5.77      itca 6.89
tzu  5.55      rsm  6.31      Dan  6.95
Cal  5.63      Crow 6.43      yoop 6.97
DF   5.65      fpli 6.54      DrP  6.98
KF   5.65      MG   6.64      nr   7.39
km   5.67      BD   6.64      EZ   7.65
bbs  5.76      snd  6.80      taco 8.06
                     
RMSE                     
Cal  6.74      rsm  7.51      snd  8.39
AJ   6.76      tzu  7.55      Dan  8.62
km   6.94      Crow 7.83      itca 8.91
DF   7.05      fpli 7.90      EZ   8.95
bbs  7.23      yoop 8.21      nr   9.17
DSM  7.38      BD   8.27      DrP  9.21
KF   7.44      MG   8.34      taco 9.72

[kmedved]

What stands out to me is tarrazu. He leads in average error, but is much more middle of the road in RMSE (most people rank pretty close in both). Also, his picks are only 15-14-1 vs. the LV Westgate picks, which is kind of amazing for someone leading in average error.

[tarrazu]

What stands out to me is tarrazu. He leads in average error, but is much more middle of the road in RMSE (most people rank pretty close in both). Also, his picks are only 15-14-1 vs. the LV Westgate picks, which is kind of amazing for someone leading in average error.

Not sure what to make of this, but here are some things that stood out to me as I have followed along with the win projections thread this season. Maybe this sheds some light?

- I did terrible on my highest/lowest win projections. (GSW/SAS and LAL/PHI)
- A lot of my biggest misses were shared by APBR submissions as a whole (HOU, NOP, CHA as well as the teams I just mentioned on the ends to some extent)
- Some of my "closest" predictions were contrarian vs APBR as a whole and/or lucky. I believe these are the 3 teams that have helped my submission most relative to rest of APBR: MIA, PHX, CHI. Obviously helped by a disastrous season from PHX.

[Mike G]

Here are our 'records' vs our group avg.

Code: Select all

24   6   DSM     16  14   rsm
22   8   tzu     15  15   Dan
21   9   DF      15  15   BD
21   9   KF      14  16   Crow
20  10   bbs     14  16   itca
20  10   AJ      13  17   snd
20  10   km      13  17   yoop
18  12   MG      12  18   nr
18  12   Cal     10  20   15py
17  13   DrP      9  21   taco
16  14   fpli     9  21   EZ

I rank higher in this one, and some of my errors are enormous: picked GSW for 53 and LAL for 34. Both were extremes and extremely off. Also had Bos at 38 and Sac at 43.

[Nate]

...
- Some of "closest" predictions were contrarian vs APBR as a whole and/or lucky. ...
...

This year is either a huge outlier or the start of a profound shift in the NBA. So it's not that crazy to have careful predictions prove to be inaccurate.

[Mike G]

Predictors with fewer extreme predictions tend to do better, at least this year.
Here we are ranked by avg error (least to most), listing those teams for which we had the highest or lowest predictions. Versus the group avg error on these teams, we could be said to be winning or losing on our extreme guesses.

Code: Select all

apbr  W-L    Winning and Losing
tzu   3-1  W: Mem 43, Orl 39, Phx 29   L: SAS 51
KF    0-0
km    0-1   L: Atl 43
DF    0-0
AJ    1-0  W: GSW 66
bbs   0-2   L: Atl 43, Bos 52
Cal   0-0
DSM   0-0
rsm   2-0  W: LAL 20, SAS 61
Crow  0-0
fpli  1-2  W: Ind 46   L: NOP 49, Uta 36
BD    2-2  W: Den 37, Phl 15   L: Bos 36, Mil 48
MG    1-4  W: Det 42  L: Cle 51, GS 53, LA 34, Sac 43
snd   0-1   L: Cle 61
DrP   6-7  W: GS 66, Hou 48, LA 20, Por 43, Tor 54
  L: Atl 54, Mia 34, NY 14, Orl 26, Phx 42, Uta 52

itca  0-3   L: Cha 25, Chi 52, Tor 42
Dan   5-8  W: Cha 47, Chi 37, Ind 46, NO 40, Por 43
  L: Brk 17, Dal 32, Den 23, GS 53, Hou 61, Mia 55, OK 69, Was 30

yoop  0-4   L: Cle 51, Mem 55, Min 34, Tor 42
nr    0-3   L: LAC 59, Min 20, OKC 50
EZ    1-3  W: LAC 51  L: Brk 31, Det 27, Sac 29
taco  1-4  W: NYK 36  L: Chi 52, Dal 32, Ind 30, Por 27

total 23-45

Not showing whether these were high or low bids -- should be obvious?
Some of these may flip by season's end. For example, the Pacers or Cavs could heat up or cool off -- their high guesses are a bit further off than the avg error right now -- and L could become W.

Extremism doesn't pay!

[Mike G]

With just 2-3 games left:

Code: Select all

RMSE                    
AJ   6.94      rsm  7.67      MG   8.52
Cal  7.00      tzu  7.71      Dan  8.84
km   7.09      Crow 8.04      itca 9.14
DF   7.25      fpli 8.14      EZ   9.19
bbs  7.29      snd  8.43      DrP  9.32
DSM  7.47      yoop 8.46      nr   9.33
KF   7.54      BD   8.50     taco 10.02
.                            15py 10.18

average error
AJ   5.67      DF   5.93      itca 7.09
tzu  5.76      rsm  6.44      yoop 7.21
KF   5.79      Crow 6.67      DrP  7.26
km   5.80      fpli 6.79      Dan  7.26
bbs  5.86      snd  6.80      nr   7.64
DSM  5.90      BD   6.88      EZ   7.84
Cal  5.92      MG   6.96      taco 8.40
.                             15py 9.26
 
avg sqrt, squared                     
tzu  4.59      Cal  5.25      MG   6.02
KF   4.61      itca 5.59      DrP  6.12
AJ   4.77      BD   5.66      Dan  6.31
DSM  4.81      rsm  5.71      yoop 6.40
km   4.89      snd  5.81      nr   6.63
bbs  4.95      Crow 5.82      EZ   6.98
DF   5.02      fpli 5.92      taco 7.26
.                             15py 8.63

This last set compresses large errors and rewards hitting the nail on the head.

[Mike G]

FINAL STANDINGS

Code: Select all

avg error
AJ   5.73      rsm  6.47      nr    7.67
km   5.73      Crow 6.67      EZ    7.97
bbs  5.83      fpli 6.73      taco  8.60
KF   5.87      snd  6.87      15py  9.27
tzu  5.87      BD   7.00         
Cal  5.93      MG   7.07         
DF   6.00      itca 7.17         
DSM  6.00      yoop 7.20         
.              DrP  7.27         
.              Dan  7.33         

RMSE
AJ   7.03      Crow 8.10      itca  9.27
Cal  7.12      fpli 8.21      EZ    9.28
km   7.13      snd  8.54      DrP   9.44
bbs  7.34      yoop 8.58      nr    9.47
DF   7.38      BD   8.58      taco 10.19
DSM  7.58      MG   8.63      15py 10.25
KF   7.66      Dan  8.89         
rsm  7.73                  
tzu  7.84                  

[permaximum]

Player-tracking PM won again. Congrats.

Final RMSE of the winning entry is 7+ too, again.

[ampersand5]

Congratulations Andrew! What a terrific accomplishment, both for you, and the community at large.

This adds significant credence to the strength of PTPM, and consequently, I believe PTPM deserves significantly increased attention and focus.

With every additional year of Sportsvu data, PTPM is only going to be more accurate. Should we be looking at PTPM to measure individual player's impact?

[EvanZ]

Can we say for certain that PT-PM is the best projection system?

Is everyone using the same minutes projections? If not, how can you tell me this is more about PT-PM versus a more accurate prediction of minutes for each player?

I would suggest each participant do a retrodiction using minutes actually played.

[Crow]

Evan's suggestion would provide valuable insight. Was the implicit average minutes distribution for the 2016 top 5 most accurate prediction entries closer to actual than any of those five?

[tarrazu]

Can we say for certain that PT-PM is the best projection system?

Is everyone using the same minutes projections? If not, how can you tell me this is more about PT-PM versus a more accurate prediction of minutes for each player?

I would suggest each participant do a retrodiction using minutes actually played.

Should also consider how one implements their ratings into a win total. That process alone could result in different projections.