2014-15 Hawks

[Mike G]

Over in the Predictions thread, it is seen that those of us making predictions guessed Atlanta would win 41 to 46 games. Pretty close agreement among 11 of us at least. They're 34-8, heading toward what looks like about 61 wins. Where did they get another 15 to 20 wins?

Their top 6 guys are veterans of at least 6 NBA seasons. They've trended thus:

Code: Select all

PER       2010   2011   2012   2013   2014   2015
Millsap   16.7   19.8   21.8   19.8   19.8   19.6
Korver    13.9   13.0   13.6   13.9   13.5   16.2
Horford   19.4   20.7   19.0   19.8   22.0   20.7
Teague    11.0   14.6   15.8   16.8   17.1   22.6
Carroll    8.5    8.7   11.1   15.5   13.9   13.8
Sefolosha  9.8   10.4    9.8   12.8   10.4   13.9
. all 6   13.2   14.5   15.2   16.4   16.1   17.8
                  
WS/48     2010   2011   2012   2013   2014   2015
Millsap   .151   .143   .179   .154   .129   .167
Korver    .130   .149   .172   .137   .118   .178
Horford   .183   .179   .180   .153   .141   .181
Teague    .042   .083   .131   .111   .098   .212
Carroll   .031   .066   .069   .134   .119   .138
Sefolosha .082   .097   .095   .146   .092   .135
. all 6   .103   .120   .138   .139   .116   .169
                  
BPM       2010   2011   2012   2013   2014   2015
Millsap    2.0    2.0    4.6    3.4    2.9    3.8
Korver      .3     .0    2.4    2.4    2.0    4.1
Horford    3.1    3.8    2.5    2.7    2.3    2.9
Teague    -3.9   -1.8    1.0    -.7   -1.0    3.3
Carroll   -3.8   -5.6   -1.0    1.9    1.9    2.1
Sefolosha  1.3    1.9    1.0    3.4     .9    2.1
. all 6    -.2     .1    1.8    2.2    1.5    3.1
                  
xyz       2010   2011   2012   2013   2014   2015
Millsap    5.5    5.8    8.2    6.7    6.0    7.2
Korver     4.0    3.9    5.5    5.1    4.6    6.9
Horford    7.1    7.6    6.7    6.4    6.3    7.2
Teague      .5    2.3    4.6    3.5    3.2    8.4
Carroll     .4    -.2    2.3    5.0    4.6    5.0
Sefolosha  3.5    4.0    3.4    5.6    3.4    4.9
. all 6    3.5    3.9    5.1    5.4    4.7    6.6

xyz = PER*(WS/48) + BPM/2 +2
The all 6 line is just a straight average, not weighted. These 6 account for about 71% of minutes.
Note that all 6 players declined, according to xyz, from 2013 to 2014. Some for 2 or 3 years straight.

Here's my own stat for corroboration:

Code: Select all

eW/484    2010   2011   2012   2013   2014   2015
Millsap   1.35   1.63   1.83   1.55   1.65   1.69
Korver     .86    .68    .68    .87    .79   1.09
Horford   1.52   1.77   1.50   1.65   1.81   1.62
Teague     .52    .68   1.03   1.18   1.34   1.89
Carroll    .27    .25    .39    .85    .79    .90
Sefolosha  .35    .46    .35    .76    .38    .84
. all 6    .81    .91    .96   1.14   1.13   1.34

[Mike G]

Here's how I predicted the Hawks would win 44 games. Showing last year's mpg (total minutes / 82) and eWins/484 min; this year's predicted minutes and eWins from last year's e484; actual mpg, eW/484 and projected eW totals for the year.

Code: Select all

Atlanta     - 2013-14 -     '15 pred    thru 38      
Hawks      tm   mpg  e484   mpg   eW   mpg  e484   e82
Millsap   Atl   30   1.65   28   7.8   33   1.69   9.4
Korver    Atl   29    .79   24   3.2   33   1.09   6.1
Horford   Atl   12   1.81   28   8.6   30   1.62   8.2
Teague    Atl   31   1.34   28   6.3   29   1.89   9.3
Carroll   Atl   29    .79   26   3.5   28    .90   4.3

Sefolosha Okl   19    .38   14    .9   18    .84   2.6
Schroder  Atl    8    .12    2    .0   17   1.10   3.2
Scott     Atl   18   1.07   20   3.6   14   1.12   2.7
Antic     Atl   11    .83   20   2.8   13    .62   1.4
Mack      Atl   18    .83   18   2.5   10    .72   1.2

Bazemore  LAL   11    .67    8    .9    8    .35    .5
Brand     Atl   17    .81   14   1.9    4    .38    .3
Muscala   Atl    3    .58    2    .2    2   1.40    .5
Jenkins   Atl    2   -.03    2    .0    1    .29    .0
Payne                 .50    8    .7    0   -.21    .0

totals         251         242  43.1  240         49.6

A season total of 49.6 eW is 8.6 above the avg of 41.0
So expected wins would be 17.2 above that avg, or 58.2

[Crow]

Revised-

So PER increased by about 10% for all6, ws/48 by 50%, bpm by 100%, xyz by 30%, eW by 15%. Why is BPM so different from the rest? Or given the team win change, why are the rest so different from BPM? Is the answer mainly the integral to the regression, team adjustment? How much of the change is the result of that element vs. all of the player's individual stat weights?

( PER's extreme by comparison lack of win value change may be affected heavily by usage levels that probably didn't change much and a change in shot defense would be completely missed due to absence.)

Is blending prudent in the face of such variation? Maybe not here but need to see more examples. Thinking about the reasons for the variation can be helpful.

Should the other metrics get a team adjustment or a bigger one? It improves explanatory match but does it improve explanation?

[DSMok1]

Revised-

So PER increased by about 10% for all6, ws/48 by 50%, bpm by 100%, xyz by 30%, eW by 15%. Why is BPM so different from the rest? Or given the team win change, why are the rest so different from BPM? Is the answer mainly the integral to the regression, team adjustment? How much of the change is the result of that element vs. all of the player's individual stat weights?

( PER's extreme by comparison lack of win value change may be affected heavily by usage levels that probably didn't change much and a change in shot defense would be completely missed due to absence.)

Is blending prudent in the face of such variation? Maybe not here but need to see more examples. Thinking about the reasons for the variation can be helpful.

Should the other metrics get a team adjustment or a bigger one? It improves explanatory match but does it improve explanation?

Percentage changes don't make sense if you don't have the same 0 line for each stat. If you add 15 to all BPM values (like PER, where average is 15), the "percentage change" would be much different.

[Crow]

BPM is win rate based so adding 15 makes no real world sense. Changing all other metrics to win rate basis makes sense. Then, I would say, a % change comparison would make sense. But this isn't going to change the yr to yr % changes for the other metrics where both values are above neutral, which they are in over 70% of the cases (all but 8 of 30 cases. 6 players * 5 metric ratings). So the degree of % change from previous % change basis will be modest, I think.