2018-19 team win projection contest

[eminence]

Mmkay, so Pacers wins and Grizzlies losses, I can get behind it.

Overall quite happy with how the season has gone in terms of these, held steady with the front pack and that's all you can really ask for. Would like to pull it out over Pelton

[eminence]

Grizz appear to have missed the memo.

[Mike G]

It has been a rough week, esp. for Cica, J-TF, and trzu.

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.71   6.71      gold   6.42   8.01
AnJo   5.86   7.51      Jinx   6.45   7.84
cali   5.96   7.21      Crow   6.55   8.32
KPel   6.15   7.62      538_   6.67   8.20
emin   6.22   7.67      sndi   6.69   8.12
shad   6.26   7.58      kmed   7.16   8.88
RyRi   6.33   7.72      J-TF   7.17   8.74
vegas  6.34   7.69      Rd11   7.87   9.86
bbst   6.38   7.76      Cica   7.89   9.16

The group avg of 6.56 is worst since Dec. 13.
AnJo and KPel had the least-bad week, as trzu lead is cut in half.

update Mar. 31

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.77   6.81      gold   6.51   8.18
AnJo   6.02   7.73      Jinx   6.59   8.04
cali   6.13   7.42      Crow   6.65   8.44
KPel   6.25   7.84      sndi   6.78   8.29
emin   6.40   7.87      538_   6.81   8.34
bbst   6.42   7.93      kmed   7.26   8.95
shad   6.46   7.84      J-TF   7.28   8.86
RyRi   6.47   7.92      Rd11   7.89   9.97
vegas  6.47   7.94      Cica   8.04   9.36

[Mike G]

Avg error is now 6.65
A blank line represents a separation of .10 or more.

Code: Select all

trzu   5.71     .    
.               sndi   6.85
.               538_   6.88
AnJo   5.96     .    
.               .
cali   6.15     .  
.               J-TF   7.24
KPel   6.27     kmed   7.31
emin   6.33     .   
shad   6.41     .    
bbst   6.44     .    
vegas  6.45     .    
RyRi   6.51     .    
gold   6.52     .    
Crow   6.61     Cica   7.93
Jinx   6.62     Rd11   7.93
.

I find it kind of impressive, in a way, that the season can't be predicted any better than this. It isn't football; there are 5 times as many games -- each with 5x as many possessions?
For every team that you guessed within 2 or 3, there's another that almost halved/doubled their expected W or L.

[eminence]

I'd really like to eke out an edge over Pelton, always feels good to finish ahead of a professional

[Mike G]

You could hope Hornets, Pacers, Raptors, Rockets win; Warriors & Grizzlies must lose. A couple of other teams could make a difference with a pronounced win or lose streak.

[Mike G]

Half the field is looking better than Vegas now. Contention at 2-3 and at 4-5.

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.70   6.74      RyRi   6.49   7.88
AnJo   5.96   7.71      Jinx   6.55   8.01
cali   6.03   7.37      Crow   6.60   8.37
emin   6.177  7.794     sndi   6.76   8.24
KPel   6.183  7.789     538_   6.82   8.31
bbst   6.37   7.87      J-TF   7.16   8.86
shad   6.43   7.78      kmed   7.21   8.88
gold   6.47   8.08      Rd11   7.86   9.99
vegas  6.48   7.89      Cica   7.87   9.29

Currently the Clipps are 47-32, but their MOV /Pythagorean W-L is 42-37. Still above all expectations, but much closer.
Conversely, the Pels are 32-47 but should be 37-42: still below our guesses but better in Pyth.
When I "correct" their b-r.com projections by + or - 5, everyone's error improves. In general, the "luck factor" (if that's what it is) has made predictions worse.
Making these adjustments for all teams, our group shakes out like this.

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   4.88   6.17      vegas  5.97   7.12
cali   5.25   6.75      gold   5.97   7.54
KPel   5.34   6.97      Crow   6.03   8.02
emin   5.40   7.09      538_   6.08   7.68
AnJo   5.54   6.95      sndi   6.16   7.70
shad   5.78   7.00      kmed   6.59   8.35
Jinx   5.79   7.34      J-TF   7.07   8.72
RyRi   5.81   7.24      Cica   7.10   8.74
bbst   5.89   7.28      Rd11   7.31   9.66

AnJo avg err improves by 0.42, but he drops to 5th. Cali is in 2nd, but trzu margin is still larger.

Subtracting everyone's errors in the lower table from those in the upper table, we get (roughly) the amount by which errors are inflated by bad luck:

Code: Select all

av abs err   rmse
KPel   .84    .82
trzu   .82    .57
cali   .78    .62
emin   .78    .70
Cica   .77    .55
Jinx   .76    .67
538_   .74    .63
RyRi   .68    .64
shad   .65    .78
kmed   .62    .53
sndi   .60    .54
Crow   .57    .35
Rd11   .55    .33
vegas  .51    .77
gold   .50    .54
bbst   .48    .59
AnJo   .42    .76
J-TF   .09    .14

avg    .62    .59

Those most dinged by bad luck are among the leaders, which means that luck has compressed the field, generally favoring the trailers.

[Crow]

I've never used decimals. I've wondered how much of an advantage that could be. How would one of the leaders using decimals average error change if rounded to nearest win? Caliban for example.

[Mike G]

Both caliban and Andrew Johnson avg errors are unchanged by rounding them to nearest integer.

The micro does not create advantage over the macro.
West teams are projecting to win 42.7 games on avg, East teams 39.3
Our 18 entries average guesses totaled 42.4 vs 39.6
The top 6 in the race all predicted a smaller E-W difference. Correlation between predicted E/W disparity and Avg Err is -0.35

One year the winner did not even avg 41 wins per team in his forecast.

[tarrazu]

Yup, 2 years ago, not good at adding sometimes.

[Mike G]

Some positions have flipped again.

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.61   6.74      Jinx   6.44   7.98
cali   5.91   7.37      vegas  6.45   7.89
AnJo   5.95   7.71      Crow   6.55   8.41
KPel   6.07   7.78      sndi   6.67   8.23
emin   6.15   7.85      538_   6.75   8.27
bbst   6.29   7.86      J-TF   7.06   8.83
RyRi   6.41   7.86      kmed   7.12   8.84
shad   6.41   7.78      Rd11   7.86   9.89
gold   6.43   8.07      Cica   7.90   9.34

update Monday:

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.70   6.79      RyRi   6.47   7.86
cali   5.91   7.39      vegas  6.49   7.93
AnJo   6.00   7.69      Crow   6.55   8.44
KPel   6.18   7.78      sndi   6.68   8.24
emin   6.24   7.91      538_   6.79   8.26
bbst   6.33   7.89      J-TF   7.12   8.86
shad   6.40   7.79      kmed   7.14   8.85
gold   6.44   8.08      Rd11   7.80   9.92
Jinx   6.45   7.97      Cica   7.94   9.37

[Mike G]

Final Standings

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.87   6.97      RyRi   6.61   8.02
cali   6.07   7.53      vegas  6.70   8.09
AnJo   6.19   7.85      sndi   6.80   8.38
KPel   6.35   7.93      Crow   6.80   8.61
emin   6.40   8.07      538_   6.93   8.42
bbst   6.54   8.07      kmed   7.26   9.00
gold   6.60   8.22      J-TF   7.30   9.01
Jinx   6.60   8.13      Rd11   7.88  10.10
shad   6.61   7.94      Cica   8.10   9.55

Congrats to tarrazu, who took the lead at end of Nov. and never relinquished it.
The margin is even more pronounced with exponent = 2 (RMSE)
Winners at various exponent:

Code: Select all

exponent
from - to
10.7 - ...   shad
0.43 - 10.6  trzu
0.39 - 0.42  AnJo
0.01 - 0.38  emin

eminence came within 1 win on Cha, Chi, Phl, GSW, and Mem; within 3 on 13 teams.
AJ had 6 within 1, and 10 of 3 or less.
shadow takes the Smallest Biggest Error Award, missing Sac by 14. Everyone else had at least a 15, except bbstats with three 14ers.

We had 10 entries better than Vegas!

[Jinxed]

Thanks for doing all the scoring Mike. I'm just happy I beat Vegas.

[RyanRiot]

Phew, narrowly better than Vegas.

[shadow]

Congrats to tarrazu for nearly going wire to wire.

Thanks to Mike G for taking the time to provide updates throughout the season.

Looking forward to next season.