Correlations between player minutes and: e484, PER, WS/48, BPM

[Mike G]

Without internet for most of the last month, I've dug up last season's playoffs to gather correlations between some player summary stats and the minutes the player received.
Here are the 2 Finalist teams, round by round. Correlations are taken among top-5 (by minutes), top 6, etc. to top 10 (unless they only went 8 or 9 deep). So this is the average of 6 separate correlations; the 10th man is counted once, the 9th man twice, ... the top 5 are counted 6 times.

So here is an example resulting in the correlation line for Miami in their one series vs Boston. Butler and Rozier weren't playing, and their coach had to improvise.

Code: Select all

min   e484   PER   WS/48   BPM   
32    .82   11.8   .006    0.4   Love
52   -.06    8.0  -.038   -2.1   Mills
60   -.36    1.7  -.144   -9.7  Robinson
107   .61   15.9   .155    3.9   Wright
123   .56   10.1  -.079   -3.9   Jaquez 

126  -.25    3.4  -.065   -6.4  Highsmith
128  1.17   15.5   .081    3.9   Jović
176   .30   10.9   .056   -0.9   Martin
185   .62   10.5  -.059   -1.4   Herro
192  1.49   18.7   .050    2.1   Adebayo

#       - - correlations - -     who
10    .42   .41    .25    .27    1-10 
9     .64   .54    .32    .42    1-9
8     .59   .55    .30    .48    1-8
7     .33   .20   -.16    .02    1-7
6     .37   .46    .30    .36    1-6
5     .39   .46    .06    .23    1-5

Mia   .46   .44    .18    .30    avg

The player stats were recorded after the 1st round, 2nd round, etc.
All 4 stat correlations are weakest at 1-7, where #7 Delon Wright looks like one of their best players.
All are strongest thru 8 or 9 , with really weak contributions in the 8-9 spots.

The 2024 Finalists, round by round:

Code: Select all

rd   G    tm   opp    e484   PER   WS/48   BPM     avg
1    5    Bos  Mia    .50    .29    .01    .12     .14
2    5    Bos  Cle    .66    .47    .21    .55     .41
3    4    Bos  Ind    .92    .80    .34    .44     .53
4    5    Bos  Dal    .29    .10   -.29   -.23    -.14

1    6    Dal  LAC    .66    .41    .11    .44     .32
2    6    Dal  OKC    .38    .12   -.09    .39     .14
3    5    Dal  Min    .21    .15    .00    .07     .07
4    5    Dal  Bos    .47    .26   -.56    .23    -.02
.         averages:   .51    .32   -.03    .25     .18

https://www.apbr.org/metrics/viewtopic. ... 9&start=15
Win Shares had Doncic and Irving among the worst Mavs in the Finals. Lively, Gafford, and Exum were "great" and in the bottom 1/2 of minutes.
Celtics were almost as upended. Tatum, Brown and Pritchard were their weak links.

Conference finals losers:

Code: Select all

rd   G    tm   opp    e484   PER   WS/48   BPM     avg
1    4    Min  Phx    .49    .58    .49    .62     .56
2    7    Min  Den    .69    .74    .68    .72     .71
3    5    Min  Dal    .59    .45   -.07    .15     .18

1    6    Ind  Mil    .22    .15    .23    .34     .24
2    7    Ind  NYK    .36    .21    .19    .19     .20
3    4    Ind  Bos    .14    .09    .08    .05     .07
.         averages:   .42    .37    .27    .34     .33

https://www.apbr.org/metrics/viewtopic.php?t=10104
Pacers' Haliburton was their best player but missed more than half the ECF series.

2nd-round eliminees:

Code: Select all

rd   G    tm   opp    e484   PER   WS/48   BPM     avg
1    6    NYK  Phl    .36    .19    .00    .38     .19
2    7    NYK  Ind    .25    .01   -.17    .07    -.03

1    5    Den  LAL    .66    .65    .54    .65     .61
2    7    Den  Min    .60    .59    .38    .55     .51

1    4    OKC  NOP    .64    .56    .27    .17     .34
2    6    OKC  Dal    .65    .56    .48    .50     .51

1    7    Cle  Orl    .44    .42    .32    .41     .38
2    5    Cle  Bos    .12    .04    .09   -.03     .04
.         averages:   .46    .38    .24    .34     .32

west https://www.apbr.org/metrics/viewtopic. ... 0&start=15
east https://www.apbr.org/metrics/viewtopic.php?t=10101
Cavs with superman Mitchell down midway thru their last series, Strus became their 'leader'.

Also-rans:

Code: Select all

rd   G    tm   opp    e484   PER   WS/48   BPM     avg                               
1    7    Orl  Cle    .73    .68    .37    .80     .62
1    6    Phl  NYK    .59    .58    .58    .53     .57
1    6    Mil  Ind    .37    .28    .13    .06     .15
1    6    LAC  Dal    .65    .69    .73    .79     .74

1    5    LAL  Den    .82    .79    .64    .72     .72
1    5    Mia  Bos    .46    .44    .18    .30     .30
1    4    Phx  Min    .65    .63    .48    .61     .57
1    4    NOP  OKC   -.24   -.36   -.30   -.28    -.32
.         averages:   .50    .47    .35    .44     .42

Trey Murphy did not lead the Pels anywhere.
Top-heavy Lakers without much bottom support.

[Mike G]

2023 Playoffs:

Code: Select all

rd   G    tm   opp   e484    PER   WS/48   BPM     avg
1    5    Den  Min    .78    .70    .63    .67     .67
2    6    Den  Phx    .57    .54    .41    .50     .49
3    4    Den  LAL    .84    .84    .75    .85     .81
4    5    Den  Mia    .66    .55    .44    .48     .49

1    5    Mia  Mil    .50    .49    .27    .35     .37
2    6    Mia  NYK    .62    .59    .43    .33     .45
3    7    Mia  Bos    .62    .51    .32    .40     .41
4    5    Mia  Den    .71    .48    .09    .24     .27
correll. avg          .66    .59    .42    .48     .49

rd   G    tm   opp   e484    PER   WS/48   BPM     avg
1    6    LAL  Mem    .80    .62    .37    .54     .51
2    6    LAL  GSW    .76    .62    .46    .54     .54
3    4    LAL  Den    .87    .89    .85    .87     .87

1    6    Bos  Atl    .68    .35   -.37   -.02    -.01
2    7    Bos  Phl    .63    .53    .28    .31     .37
3    0    Bos  Mia    .60    .17   -.23    .00    -.02
correll. avg          .72    .53    .23    .37     .38

rd   G    tm   opp    e484    PER   WS/48   BPM     avg
1    4    Phl  Brk    .41    .23    .22    .51     .32
2    7    Phl  Bos    .60    .57    .41    .48     .48

1    7    GSW  Sac    .46    .27   -.10    .05     .07
2    6    GSW  LAL    .36    .13   -.02    .23     .11

1    5    NYK  Cle    .46    .32    .21    .33     .29
2    6    NYK  Mia    .79    .72    .59    .70     .67

1    5    Phx  LAC    .75    .66    .43    .53     .54
2    6    Phx  Den    .92    .85    .58    .84     .76
correll. avg          .60    .47    .29    .46     .41

rd   G    tm   opp    e484    PER   WS/48   BPM     avg
1    7    Sac  GSW    .74    .67   -.01    .59     .42
1    6    Mem  LAL    .52    .34    .09    .10     .18
1    5    LAC  Phx    .30    .24   -.17    .15     .07
1    5    Min  Den    .76    .70    .62    .52     .61

1    6    Atl  Bos    .43    .22   -.21    .05     .02
1    5    Cle  NYK    .74    .71    .26    .58     .51
1    5    Mil  Mia    .13    .18    .17    .00     .12
1    4    Brk  Phl    .57    .39    .12    .28     .26
correll. avg          .52    .43    .11    .28     .27

correll. 23 playoffs  .62    .50    .26    .40     .39

Minutes per game played would probably be better than total minutes in a series, as when a major player misses some games and is therefore much lower in the total minutes being correlated.
However, all the stats suffer the same issue, so they are still comparable to one another.

2022 Playoff correlations:

Code: Select all

rd   G     tm  opp    e484   PER   WS/48   BPM     avg
1    5    GSW  Den    .50    .35    .21    .32     .29
2    6    GSW  Mem    .43    .13   -.13    .22     .08
3    5    GSW  Dal    .28    .14   -.20   -.22    -.09
4    5    GSW  Bos    .17    .02   -.32   -.04    -.11
1    4    Bos  Brk    .59    .53   -.02    .37     .29
2    7    Bos  Mil    .61    .43    .07    .45     .32
3    7    Bos  Mia    .38    .27    .06    .16     .16
4    5    Bos  GSW    .31    .08   -.18    .06    -.01
Finalist correll.:    .41    .24   -.06    .17     .12

rd   G     tm  opp    e484   PER   WS/48   BPM     avg
1    5    Mia  Atl    .45    .36    .25    .32     .31
2    6    Mia  Phl    .80    .84    .92    .90     .89
3    7    Mia  Bos    .61    .60    .65    .59     .61
1    6    Dal  Uta   -.09   -.06    .17   -.02     .03
2    7    Dal  Phx    .50    .34    .08    .28     .23
3    5    Dal  GSW    .30    .26   -.02    .28     .17
Conf. finals          .43    .39    .34    .39     .37

rd   G     tm  opp    e484   PER   WS/48   BPM     avg
1    6    Phx  NOP    .45    .43    .50    .50     .48
2    7    Phx  Dal    .38    .07   -.19   -.01    -.04
1    5    Mil  Chi    .49    .30    .02    .10     .14
2    7    Mil  Bos    .80    .73    .03    .84     .53
1    6    Mem  Min    .46    .26    .03    .27     .18
2    6    Mem  GSW   -.01   -.01    .06   -.02     .01
1    6    Phl  Tor    .80    .53    .25    .19     .32
2    6    Phl  Mia    .61    .50    .15    .04     .23
2nd rounders          .50    .35    .11    .24     .23

rd   G     tm  opp    e484   PER   WS/48   BPM     avg
1    6    Min  Mem    .51    .18   -.20   -.11    -.05
1    6    NOP  Phx    .36    .14   -.25    .19     .02
1    6    Tor  Phl    .39    .28    .12    .11     .17
1    6    Uta  Dal    .60    .41    .13    .30     .28
1    5    Atl  Mia    .26   -.10   -.40   -.18    -.23
1    5    Chi  Mil    .84    .79    .30    .50     .53
1    5    Den  GSW    .59    .51    .31    .45     .42
1    4    Brk  Bos   -.06   -.20   -.34   -.33    -.29
1st rounders          .44    .25   -.04    .11     .11

[Crow]

PER and WS/48 are lacking in shot defense. BPM is very poor at assigning credit for shot defense to individuals because it is not play by play based.

I would like to see correlations to one of more RAPM based metrics as they at least measure shot defense while on the court, along with everything else with no subjective guidance for that part of a hybrid metric or the whole thing if pure RAPM.

[Mike G]

Where would I find RAPM or especially for playoffs? If it requires large numbers of games, it may not tell us much?

[Crow]

https://xrapm.com/

and other places.

[Mike G]

OK thanks, but that is updated yesterday and has Embiid 4th and Kawhi 9th, and they haven't even played this year.
Even real this-season RAPM isn't do-or-die mode like playoffs. A player in regular season may get a lot or little minutes for reasons other than Every Game Matters.

Whatever it is that these stats value or undervalue, the fact that head coaches regularly agree with some, more than with others, may mean something. They are likely not aware that they are 'agreeing', but they make decisions as if they do.

[Crow]

That front page displayed XRAPM may not have this season in it yet.

But under RAPM flavors tab, there is a 1 yr only RAPM.

Playoff only metrics are uncommon but there is some useful comparison to regular season stats on far right of pages like this one:

https://nbarapm-flask.replit.app/player/LeBron%20James

[Mike G]

Correlations thru the allstar break. Now it's minutes per game vs 4 summary stats.
The top 11 minutes guys for each team are then re-ranked according to MPG. Top-5 thru top-11 correlations are averaged.

Code: Select all

avg    tm    e484    PER   WS/48   BPM
.87    Cle    .94    .90    .72    .90
.82    LAC    .84    .87    .64    .92
.77    LAL    .77    .78    .76    .78
.76    Phx    .84    .89    .63    .70
.75    Mil    .69    .72    .79    .78

.67    Brk    .66    .73    .64    .65
.64    OKC    .69    .66    .58    .64
.57    Den    .60    .60    .56    .53
.55    Mia    .77    .63    .31    .50
.52    Dal    .64    .56    .30    .59

avg    tm    e484    PER   WS/48   BPM
.52    Tor    .70    .62    .17    .58
.51    Min    .66    .63    .30    .46
.51    GSW    .59    .50    .35    .60
.50    Ind    .63    .57    .38    .43
.49    Sac    .64    .59    .37    .36

.48    SAS    .63    .56    .13    .59
.47    Phl    .63    .57    .19    .47
.44    Uta    .48    .44    .41    .43
.43    NOP    .60    .51    .11    .48
.35    Mem    .61    .51   -.03    .31

avg    tm    e484    PER   WS/48   BPM
.33    Chi    .41    .41    .19    .30
.28    NYK    .30    .28    .22    .32
.25    Atl    .41    .30   -.11    .42
.24    Det    .37    .30   -.11    .38
.22    Bos    .51    .35   -.17    .20

.22    Orl    .34    .28   -.13    .39
.22    Cha    .32    .22   -.09    .41
-.07   Hou   -.06   -.15   -.09    .01
-.14   Was   -.20   -.17   -.15   -.03
-.23   Por   -.32   -.37   -.20   -.03

.43    avg    .52    .48    .26    .47

Some star players are not among their teams' top 11 in minutes: Holmgren, Embiid, Zion and Ingram.
Others are in the mix but not near the top in minutes: Porzingis, Valanciunas, Ayton, Mo Wagner, Towns, Sengun/Eason, Duren, Lively/Gafford, Capela/Okongwu.
This last group makes for weaker correlations.

[Crow]

Unless a metric has a RAPM component, starters should probably get an adjustment for facing a higher level of opponent. Maybe 10% or in that range.

[Mike G]

eWins contains a 'starter adjustment' which has diminished in importance in the last few seasons. Some players actually do better as starter, for whatever reason.
Westbrook :

Code: Select all

role   mpg   Reb   Ast    Pts   FG%   TS%   Usg%   ORt  DRt
sta   30.5   6.2   7.0   14.2  .534  .601   21.1   113  114
sub   24.1   3.7   5.6   11.3  .396  .492   24.8   101  113

Such players partially cancel the general tendency to get better numbers vs lesser opponents.
Kawhi Leonard has started all 15 of his appearances and averages only 26 mpg. I estimate that 82% of his opponents (and teammates) have been starters, and this would be one of the highest % in the league. He gets almost a 2% boost in all his 'per 36 rates', and his eW/484 rate is adjusted from 1.28 to 1.33

On the other end, players who never start and go low minutes are bumped downward about 5%.
If Payton Pritchard had his current numbers and was a starter, his e484 would go from current 1.23 to 1.32. I estimate his floor company to be only 53% starters.

League-wide, these small adjustments reduce the avg error (between team eWins and their pythagorean 'wins') only a smidge, from .97 to .96

[Crow]

Yes eWins has such an adjustment but at least 2 of the 3 others do not to my recollection. BPM might in some fashion. So the composite of 4 is half or more not adjusted for that..

EWins' adjustment has been mentioned before (long ago) and a reason for the comment. If this topic is brought up, it is appropriate to mention the adjustment.

Older metrics might still be "ok" for some purposes but newer ones with a RAPM component to adjust for specific / discrete play by play player quality differences and try to capture all impacts are preferable for greater accuracy.

Just sayin.

[Mike G]

Agree that in theory some form of RAPM should or could introduce accuracy; but in one season or 2 or 3, there always seem to be preposterous results. And do these metrics correlate to what coaches value? i.e., to player minutes?

I have considered tossing out all this eWins stuff, and just blending these other available numbers. But there's the persistent agreement with player minutes. In playoffs it's especially closer.
PER isn't too bad ranking players within a team but suffers from team to team, as it takes zero interest in opponent performance.
WS has the craziest results, but manages to hold up on the team level because no coach would take the floor with 5 low-usage / high-efficient guys.
BPM does stuff that I can't replicate, regarding opponent strength; and still doesn't jive as well with minutes given.

RAPM may be best at measuring a player when his career is several years along, or after it's over. The year to year 'noise' probably cancels and eventually agrees with perception and other production X efficiency stats. But for short intervals, like a playoff series, it doesn't even seem to exist.

[Crow]

Limits to knowledge, limits to getting expected results.

RAPM or other metrics against a specific team wouldn't have enough sample size for reliable "answers" but metric scores against top, middle and lower thirds of league by overall quality or another criteria would have some value to see, especially multi-year.

Some beat the best more than others others beat up on the meh or weak more than others. Could have similar overall metrics and have way different impact in playoffs.

[Crow]

Most team correlations of average of metrics to minutes or minutes to metrics are moderate or even low to my first impression. Some exceptions among the best teams in specific series and generally at least ok overall for top 15-20. Really weak for the bad teams. Bad management and / or tanking.

I see lots of low correlations in the playoffs.
I was given impression that Coaches' minutes to performance would shine there. Most don't.

2022 Warriors and 2024 Celtics below playoff average correlation and very low. 2023 Nuggets probably the highest that year or close? Different styles of coaching and / or far more change in predictability of performance in those 2 years? Room for further research.

Celtics up & down thru series and across years but never high.

Warriors almost always weak.

Concerns? Interest? Or "nah", zig zag all you want, just get it done?

Nuggets always at least relatively pretty high to high.

2 champs with higher playoff correlations in general but 2 champs with lower correlations in the finals. Inconclusive. But at least studied the data.

Thunder at leat average, maybe higher than average. First round opponent Pelicans with the by far lowest correlation in the study. 4-0 sweep.



E484 explicitly and separately credits minutes to my memory / understanding, the others do not.



If you have a way or ways to get the correlations up, do it.

[Crow]

"All are strongest thru 8 or 9 , with really weak contributions in the 8-9 spots."

And the 13th or 15th man?