2015-16 Team win projections

[bbstats]

TY Portland TY Boston

[bbstats]

Yoooooooo caliban....
Where you at

[Mike G]

He's at the bottom of the top 1/3

Code: Select all

KF   4.79      DSM  5.42     itca  6.54
km   4.87      rsm  5.67      Dan  6.63
bbs  4.98     Crow  5.73     yoop  6.67
DF   4.99     fpli  5.94      DrP  6.69
AJ   5.04      snd  6.17      nr   6.79
tzu  5.10      BD   6.35      EZ   7.19
Cal  5.29      MG   6.44     taco  7.67

[bbstats]

haha - I just meant I missed the frequent colorful chart updates.

thanks for the wins update Mike!

EDIT: Though - this looks like average error. After winning the average error but "losing" the contest last year I'd like to see Squared error if possible - let's stick to it

[Mike G]

Take your pick.

Code: Select all

square root of avg squared error
km   6.03      DSM  6.73      BD   7.72
AJ   6.13      rsm  6.88      itca 8.13
Cal  6.18      Crow 7.10      nr   8.21
DF   6.29      fpli 7.25      Dan  8.31
bbs  6.42      yoop 7.57      EZ   8.46
KF   6.49      snd  7.64      DrP  8.72
tzu  6.72      MG   7.68      taco 9.07

avg error
KF   4.82      DSM  5.39      itca 6.51
km   4.85      Crow 5.72      yoop 6.65
bbs  4.96      rsm  5.80      DrP  6.71
AJ   5.08      fpli 6.04      Dan  6.72
tzu  5.09      snd  6.20      nr   6.82
DF   5.10      BD   6.29      EZ   7.24
Cal  5.27      MG   6.32      taco 7.69

avg square root of error, squared
KF   3.70      Cal  4.74      MG   5.49
bbs  4.02      Crow 4.87      DrP  5.50
km   4.13      rsm  5.20      Dan  5.82
tzu  4.18      BD   5.23      yoop 5.96
DF   4.31      snd  5.25      nr   6.04
AJ   4.37      fpli 5.28      EZ   6.45
DSM  4.43      itca 5.47      taco 6.72

This last one shrinks large errors and rewards very-close guesses.

[kmedved]

Colorful update:



Mike G, what are your projections for KF btw? I feel like our numbers aren't converging on some guys, so I'm wondering if I fat fingered something somewhere. Can you link to where you got KF's? That's the one that jumps out to me.

[nbacouchside]

Take your pick.

Code: Select all

square root of avg squared error
km   6.03      DSM  6.73      BD   7.72
AJ   6.13      rsm  6.88      itca 8.13
Cal  6.18      Crow 7.10      nr   8.21
DF   6.29      fpli 7.25      Dan  8.31
bbs  6.42      yoop 7.57      EZ   8.46
KF   6.49      snd  7.64      DrP  8.72
tzu  6.72      MG   7.68      taco 9.07

avg error
KF   4.82      DSM  5.39      itca 6.51
km   4.85      Crow 5.72      yoop 6.65
bbs  4.96      rsm  5.80      DrP  6.71
AJ   5.08      fpli 6.04      Dan  6.72
tzu  5.09      snd  6.20      nr   6.82
DF   5.10      BD   6.29      EZ   7.24
Cal  5.27      MG   6.32      taco 7.69

avg square root of error, squared
KF   3.70      Cal  4.74      MG   5.49
bbs  4.02      Crow 4.87      DrP  5.50
km   4.13      rsm  5.20      Dan  5.82
tzu  4.18      BD   5.23      yoop 5.96
DF   4.31      snd  5.25      nr   6.04
AJ   4.37      fpli 5.28      EZ   6.45
DSM  4.43      itca 5.47      taco 6.72

This last one shrinks large errors and rewards very-close guesses.

2 out of 3 ain't bad.

[Mike G]

kmedved, see the 6th post in this thread.
KF predictions were submitted by ampersand.

[sndesai1]

Mike G, what are your projections for KF btw? I feel like our numbers aren't converging on some guys, so I'm wondering if I fat fingered something somewhere. Can you link to where you got KF's? That's the one that jumps out to me.

that's probably my fault (if you copied what i had compiled earlier). i had a typo for his new york knicks entry - i typed in 24 when he had 28. that should bump him up near the top for mean absolute error.
sorry!

[J.E.]

I seems (manually) bumping down the ratings of players with "off-court issues" would have been a good idea this season, and it may be beneficial to do so in the future, as well

Specifically, I'm talking about Lawson (multiple DUIs), Harden (Kardashian) and Markieff Morris (assault charges, twin brother traded)

That would have helped with 2 of the handful of teams responsible for average errors of 10+.

NOP can (only partially) be explained by the Tyreke Evans injury. Spurs are Spurs. Warriors have been mildly lucky with injuries, I think. Doesn't fully explain their great performance, though



On a completely different topic:
I'm pretty sure everyone in this contest is using a different factor for "regression to the mean". People who got the general direction of teams correct, but regressed to the mean too much or not enough could look worse than those who were generally worse at getting the direction of teams correct, but used a better factor for mean regression.

I'm thus suggesting to z-score everyone's predictions and expected wins, and compute error on the z-scores instead. That should put everyone on "equal ground"

[kmedved]

Mike G, what are your projections for KF btw? I feel like our numbers aren't converging on some guys, so I'm wondering if I fat fingered something somewhere. Can you link to where you got KF's? That's the one that jumps out to me.

that's probably my fault (if you copied what i had compiled earlier). i had a typo for his new york knicks entry - i typed in 24 when he had 28. that should bump him up near the top for mean absolute error.
sorry!

That fixed it. Thanks! Now ranks 2nd.

[kmedved]

On a completely different topic:
I'm pretty sure everyone in this contest is using a different factor for "regression to the mean". People who got the general direction of teams correct, but regressed to the mean too much or not enough could look worse than those who were generally worse at getting the direction of teams correct, but used a better factor for mean regression.

I'm thus suggesting to z-score everyone's predictions and expected wins, and compute error on the z-scores instead. That should put everyone on "equal ground"

Isn't calibrating the correct "regression to the mean" something we care about as well? It's a bit arbitrary obviously, but as an extreme example, if you generate super regressed projections which have every team between 35 and 47 wins, that doesn't feel like a very accurate prediction set, even if you nailed that Phoenix would be the 3rd worst team in the NBA at 37 wins or something.

Obviously this contest is whatever people want it to be of course.

[DSMok1]

On a completely different topic:
I'm pretty sure everyone in this contest is using a different factor for "regression to the mean". People who got the general direction of teams correct, but regressed to the mean too much or not enough could look worse than those who were generally worse at getting the direction of teams correct, but used a better factor for mean regression.

I'm thus suggesting to z-score everyone's predictions and expected wins, and compute error on the z-scores instead. That should put everyone on "equal ground"

Isn't calibrating the correct "regression to the mean" something we care about as well? It's a bit arbitrary obviously, but as an extreme example, if you generate super regressed projections which have every team between 35 and 47 wins, that doesn't feel like a very accurate prediction set, even if you nailed that Phoenix would be the 3rd worst team in the NBA at 37 wins or something.

Obviously this contest is whatever people want it to be of course.

Just another thing to consider, not the "official" measure. That way we can split out "correct regression to the mean" from "player projections" to some extent.

[Dr Positivity]

When it comes to getting the direction right, how about just tracking who was on the right end of over/under for the most teams?

I understand some people may have done a better job regressing to the mean than others, but isn't that true of other parts that went into a prediction as well? For example if two predictors used aging curves, but one of them used a better method, is that much different than one predictor regressing better than the other? Same goes for if someone's prediction used coaching well, and another's either did a less accurate job of rating coaching, or not at all

[permaximum]

I seems (manually) bumping down the ratings of players with "off-court issues" would have been a good idea this season, and it may be beneficial to do so in the future, as well

Specifically, I'm talking about Lawson (multiple DUIs), Harden (Kardashian) and Markieff Morris (assault charges, twin brother traded)

That would have helped with 2 of the handful of teams responsible for average errors of 10+.

NOP can (only partially) be explained by the Tyreke Evans injury. Spurs are Spurs. Warriors have been mildly lucky with injuries, I think. Doesn't fully explain their great performance, though



On a completely different topic:
I'm pretty sure everyone in this contest is using a different factor for "regression to the mean". People who got the general direction of teams correct, but regressed to the mean too much or not enough could look worse than those who were generally worse at getting the direction of teams correct, but used a better factor for mean regression.

I'm thus suggesting to z-score everyone's predictions and expected wins, and compute error on the z-scores instead. That should put everyone on "equal ground"

This is one of the worst posts I have ever read in this forum. You guys are going backwards... Don't make me agree with Charles Barkley at the end of the day.