2016-17 Team Win Projection Contest / Discussion

[Crow]

Being within a half point or even a point of the best seems alright. Almost everybody is doing well imo. Seems like an in the pocket year so far. Even Vegas is better than their sucky relative performance last year if I recall correctly.

[Mike G]

http://www.basketball-reference.com/fri ... f_prob.cgi
We've got a contest, suddenly.

Code: Select all

avg error    gain      avg error    gain
RyRi   3.67   .00      kmed   4.20   .16
amp5   3.71   .27      BaDo   4.24   .09
avg    3.86   .04      sndi   4.26   .07
trzu   3.93   .00      Crow   4.33   .23
lnqi   3.94   .13      taco   4.37   .06
ATCt   3.98   .03      ncsD   4.41   .12
sbs.   3.98   .06      nrfo   4.47   .15
ncsB   4.03  -.03      cali   4.55   .07
Mike   4.11   .29      jg34   4.55   .05
vegas  4.13   .03      yoop   4.71   .21
shad   4.17   .03      GK5.   4.77   .19

Showing the gain since 2 days ago, vs the leader.
RyRi's error increased by .13, and a few of us had shrinkage.

[BasketDork]

The contest seemed a little more 'involved' last year. Am I wrong ? Glaring absences I can think of are no DSmok1, no Statman, no Evan Z, no Dr. Positivity..

[Mike G]

They may have put predictions elsewhere, but I haven't gone looking, and no one has brought them here.
Also last year, two other participants were putting up tables with lots more detail than the ones I'm posting. They're in the contest this year but not saying much.

[Mike G]

Well that didn't last long.

Code: Select all

.  avg error  RMSE         avg error  RMSE
RyRi   3.58   4.59      kmed   4.13   5.21
amp5   3.73   4.64      sndi   4.18   5.06
avg    3.73   4.68      ncsD   4.22   4.95
lnqi   3.78   4.88      taco   4.24   5.17
trzu   3.81   4.90      nrfo   4.25   5.63
ncsB   3.83   4.83      Crow   4.33   5.56
ATCt   3.84   5.09      cali   4.38   5.34
sbs.   3.84   4.83      jg34   4.46   5.53
shad   3.99   5.22      yoop   4.76   5.76
Mike   4.00   5.19      GK5.   4.81   5.73
vegas  4.04   5.08     16PyR   5.72   6.73
BaDo   4.07   5.25            

#2 thru 7 are closer than #2 is to #1

Relative to shadow's latest weighted projections:

Code: Select all

avg. errors   RMSE       avg errors   RMSE
RyRi   3.62   4.80      ncsD   4.21   4.98
sbs.   3.71   4.94      sndi   4.21   5.03
avg    3.76   4.84      shad   4.35   5.49
amp5   3.82   4.81      vegas  4.38   5.41
ATCt   3.89   5.11      BaDo   4.43   5.51
trzu   3.97   5.20      GK5.   4.43   5.54
ncsB   3.98   4.99      cali   4.47   5.49
taco   4.13   5.30      jg34   4.49   5.62
kmed   4.14   5.20      Crow   4.55   5.84
lnqi   4.16   5.26      yoop   4.62   5.64
Mike   4.17   5.49      nrfo   4.95   6.04

[kmedved]

They may have put predictions elsewhere, but I haven't gone looking, and no one has brought them here.
Also last year, two other participants were putting up tables with lots more detail than the ones I'm posting. They're in the contest this year but not saying much.

I have not had time to setup a spreadsheet to automatically update for this like last year, but will try and do so over the next week and be back with colorful charts.

Motivation may also be affected by my place in the standings!

[Mike G]

Suddenly lots of contenders.

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
RyRi   3.69   4.57      kmed   4.15   5.31
amp5   3.73   4.64      ncsD   4.21   5.01
(avg   3.76   4.69)     taco   4.25   5.16
ATCt   3.78   5.06      Mike   4.25   5.32
ncsB   3.81   4.91      sndi   4.26   5.12
BaDo   3.86   5.14      Crow   4.29   5.56
trzu   3.87   4.90      jg34   4.40   5.60
shad   3.87   5.13      cali   4.45   5.41
lnqi   3.92   4.86      yoop   4.76   5.78
sbs.   3.93   4.87      GK5.   4.94   5.87
vegas  4.06   5.07     16PyR   5.67   6.76
nrfo   4.11   5.36            

Just 3 days ago, only ampersand was within .20 of the lead; now there are 6.

[Mike G]

Again a new low and more separation from the pack.

Code: Select all

avg abs err   RMSE      avg abs err   RMSE
RyRi   3.45   4.49      ncsD   4.07   4.93
avg    3.59   4.63      nrfo   4.08   5.43
amp5   3.62   4.63      cali   4.12   5.35
ncsB   3.68   4.77      sndi   4.14   5.07
ATCt   3.70   4.96      kmed   4.15   5.27
shad   3.73   5.02      taco   4.15   5.11
sbs.   3.74   4.81      jg34   4.35   5.53
lnqi   3.75   4.83      Crow   4.44   5.63
trzu   3.83   4.93      GK5.   4.73   5.72
BaDo   3.87   5.22      yoop   4.74   5.60
vegas  3.96   5.11     16PyR   5.50   6.62
Mike   4.00   5.24            

Just past 3/8 of the season.

[shadow]

Not trying to be a sore loser or anything, but neither RyanRiot's decimal or whole number projections sum to exactly 1230. The decimal version sums to 1231.7 and the whole number version sums to 1232. Although his wasn't the only submission whose wins didn't sum to 1230; Tarrazu's and Crow's didn't either.

[Mike G]

That's true, and it's not really necessary to do well. If he'd scaled to 1230 wins, he'd likely be doing better.
I think Hollinger won it one year without bothering to hit 41 per team.

And nobody is close to losing or winning anything yet. In a couple of days your error may gain or lose .20 or more, relative to anyone else here.
In a week, you've gone from .50 out of 1st, to .18, to .28 . The volatility is pretty amazing.

[shadow]

I only noticed because I recently setup a spreadsheet with everyone's projections so I could track it in real-time. I scaled the ones that weren't 1230 to be exactly 1230 and he's still leading at 3.51, so the difference is negligible and probably won't impact the final standings. I was more curious how you were handling it when doing your calculations - using his raw numbers or scaling them to 1230. I'm not that upset about not being in first place, I know it's just a casual competition and I wasn't trying to ruffle feathers.

[Mike G]

In some years, we've asked people if they want their numbers scaled to 1230. When it's just a rounding error, one might say to give the Heat one more win, something like that. This year I didn't get the numbers together until the season had started, so ...

Anyway, my sheet uses the decimal fractions, where they were provided. That too makes a negligible contribution, compared to luck, say.
Don't worry about feather ruffling, it's all good and fun here.

[Crow]

I musta moved something at last minute without offsetting properly. But I am way back, so not really a troubling issue.

[Mike G]

Average absolute errors

Code: Select all

RyRi   3.43      BaDo   3.87      cali   4.17
avg    3.58      trzu   3.88      sndi   4.18
.                                 taco   4.18
amp5   3.66      Mike   3.97         
shad   3.71      vegas  4.00      jg34   4.39
ncsB   3.72      nrfo   4.03      Crow   4.48
sbs.   3.73                         
ATCt   3.73      ncsD   4.12      GK5.   4.70
lnqi   3.77      kmed   4.13      yoop   4.77

[Mike G]

Code: Select all

RyRi   3.46      vegas  4.07
amp5   3.51      kmed   4.08
(avg   3.56)     ncsD   4.11
-                sndi   4.20
shad   3.74      cali   4.24
sbs.   3.75      taco   4.27
ATCt   3.77      jg34   4.31
ncsB   3.79         
lnqi   3.80      Crow   4.48
BaDo   3.85         
nrfo   3.89         
Mike   3.95      GK5.   4.74
trzu   4.02      yoop   4.74