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

[Mike G]

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

And the 13th or 15th man?

I found your quoted line in the opening post:

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.

They only used 12 players.

[Crow]

Used 12, so 13-15-18 not important, not helpful in the playoffs. As expected, in line with their relative performance in regular season. Negative impact likely if they had been used, so they weren't.

[Mike G]

Missing both Butler and Rozier, they were already digging deeper than preferred.
Teams stick with their better available players in postseason. But 2-4 of top 9 showed up as sub-zero.

[Crow]

In any use of the words average or below average, one should clarify mean or median. Mean is higher than many realize. Median is something of an achievement. Often paid well... but if you can buy near median performance for less or way less, you get good value... as an employer.

[Mike G]

How would one define median scoring rate, rebounding, or win production? Include players with very few minutes?
Average and mean are synonyms, but mean can mean a lot of things; even mean people know that.

[Crow]

Include or not include what you want. Mean what you want to mean.

[Mike G]

One persistent phenomenon that seems to suppress correlations between minutes and these stats: Separated by positions, SF and SG get the most minutes and are the least productive (or proficient).
Showing fraction of total minutes, and of 'wins' contributed:

Code: Select all

min   pos   eWin   perW    WS    bpmW
.188    C   .267   .262   .271   .229
.187   PF   .205   .197   .195   .195
.221   SF   .166   .159   .169   .169
.223   SG   .177   .180   .178   .200
.181   PG   .185   .202   .188   .207

This is last season's distribution, using b-r.com's position designations.
BPM I believe has a distinct position system defined by stats, and it measures players differently based on those positions.

Where the % of wins is more than the % of minutes, that position is seen as generating more than average wins per minute.
BPM diminishes the value of Centers, relative to the other 3; and Guards get relatively more credit.
Possibly a clearer form -- just dividing the %wins by the %minutes:

Code: Select all

min  pos   eWin    per    WS     bpm
.19    C   1.42   1.39   1.44   1.22
.19   PF   1.10   1.06   1.04   1.04
.22   SF    .75    .72    .76    .77
.22   SG    .79    .81    .80    .90
.18   PG   1.02   1.11   1.04   1.14

1.00 is average. There's a factor of 2 between some positions. SF get no respect!
I propose that the seeming contradiction is that PG and C are often not suited to slide to the adjacent position. Small guys tend to have defensive limitations that put them on the bench. Big guys are less suited to sprinting against small-ball lineups.
Playing 3 or 4 shooting wings may give a team an advantage; but those guys are sharing X number of shots, while a big collects plenty of rebounds, putbacks, alley-oops.

[Crow]

In a boxscore system, boxscore production is measured. Rebounds and assists substantially differentiate. Being the lead ballhandler / passer and the generally closest to the basket / biggest rebounder and often rim defender matters.

You probably need some wings for defensive matchups with some other wings. But how many? How often can you play extra "PGs" and "Centers", guys most advantaged with their physical advantages and learned skills? How many can adequately cover wings?

There has been occasional use of 2 PGs and twin towers and studies of those cases. I have done some small studies and reviews of such work but don't remember the findings in detail. Review and new work may help.

There are limits on time handling the ball
/ directly the offense and for rebounding and protecting rim, so duplication may see diminished returns but perhaps more returns than using wings at least in those categories.

Average impact by position using metrics with RAPM should come into this conversation. I basically recall PGs and / or especially Centers dominating here too but would want to re-confirm. It would make sense that playing larger than average roles on very important tasks would see that evidenced in RAPM ratings.
If anyone wants to dig up the best resources for that or do a new study, great.

Big PGs who can handle switches are of interest naturally, as are bigs who can participate in moving the ball with hands, feet and judgment.

Many still lust for wings, especially those who can do more than average on PG skills and big skills. And / or 3pt shooting and defense and switch based defense. That is reasonable. But apparently dual small / big skills above average are pretty rare.

Player acquisition should be driven by skill acquisition but sufficiently spread to all skills and cognizant of them in lineups and in proportion to importance. As measured by metrics or basketball eye & mind based. Winning based is best.

[Mike G]

Last year (2024-25) regular season correlations between min/G and 4 all-in-one stats.
Players 1 thru 15 in mpg, with at least 15 games for a team.
Playoff teams correlate better than the bottom 14, and those getting past round 1 were better yet.

Code: Select all

rd2+  PER   WS48  BPM   e484   avg
OKC   .60   .50   .57   .65    .58
Ind   .42   .33   .33   .58    .42
Min   .71   .48   .70   .79    .67
NYK   .53   .37   .60   .59    .52
Cle   .74   .60   .73   .83    .72
Den   .72   .65   .64   .72    .68
GSW   .55   .46   .58   .62    .56
Bos   .51   .06   .52   .66    .44
avg   .60   .43   .59   .68    .57

rd1   PER   WS48  BPM   e484   avg
LAC   .86   .62   .80   .86    .78
LAL   .75   .69   .81   .75    .75
Mia   .58   .32   .58   .63    .52
Mil   .48   .40   .46   .52    .46
Mem   .51   .11   .50   .61    .43
Det   .33  -.09   .40   .42    .27
Orl   .30  -.08   .36   .44    .26
Hou   .09   .06   .25   .32    .18
avg   .49   .25   .52   .57    .46

nope  PER   WS48  BPM   e484   avg
Brk   .69   .66   .61   .59    .64
Uta   .62   .49   .56   .64    .58
Phx   .72   .28   .51   .70    .55
Tor   .61   .26   .66   .69    .55
Atl   .55   .17   .58   .63    .48
Sac   .57   .31   .38   .57    .46
Cha   .42   .19   .58   .49    .42
SAS   .51  -.01   .37   .62    .37

no    PER   WS48  BPM   e484   avg
Dal   .40   .13   .47   .47    .37
Phl   .42   .12   .42   .45    .35
Chi   .28  -.04   .21   .40    .21
NOP   .17  -.10   .21   .23    .13
Por  -.20  -.26  -.21  -.03   -.18
Was  -.29  -.32  -.18  -.21   -.25
avg   .39   .13   .37   .45    .34

all   .48   .25   .47   .55    .44

eWins has zero mpg factor in this rendition.

[Crow]

Rockets lowest / worst performance to minutes correlation in the playoffs and nowhere close to what any 2nd round teams and beyond had.

Green and Brooks are gone; but still a very bad sign... including about coaching.


WS/48 with the lowest correlation in every case but one where it was modestly 2nd lowest. WS/48 is almost never used in the wild (for NBA) and probably should be left in past.

e484 generally the highest. What is the explanation?

Consideration of context of opponent scoring varies from none in PER to none direct and clearly apparent to me on BPM to generic and not play by play based in the others. Shot defense matters and is better evaluated in metrics with raw or adjusted plus minus data based on play by play instead of just team average for all minutes, on the court or not. They would make better choices for inclusion in a performance to minutes comparison. EPM actual, LeBron...

[Mike G]

Rockets' postseason correlations were higher than RS for all 4 stats, and above avg for the West in round 1.

Code: Select all

avg   West   PER   WS/48   BPM   e480      PER  WS/48  BPM  e480
.75   LAC   16.3   .106   2.18   1.02      .83   .73   .57   .88
.63   Den   15.6   .100   1.91    .98      .62   .61   .59   .69
.74   Min   15.8   .142   2.41   1.21      .77   .62   .72   .85
.62   LAL   13.4   .070   0.60    .79      .69   .51   .64   .63
.57   Hou   15.1   .111   1.42   1.00      .50   .51   .55   .70
.36   GSW   14.3   .094   1.49   1.00      .38   .10   .44   .53
.21   OKC   17.5   .206   4.70   1.41      .27   .10   .01   .47
-.46  Mem   11.2  -.010   0.25    .59     -.42  -.50  -.76  -.16

.43   avg   14.9   .102   1.87   1.00      .46   .33   .35   .57

OKC - Mem had a lot of garbage minutes, perhaps creating their low correlations.
Hou: Both Green and Brooks lost a couple mpg in the playoffs.

The GS-Hou series had zero factor for minutes but a substantial bonus factor to Starters. This is an easy component to add to any metric, but I don't know of any other that does it.
The correlation drops without it. With and without:

Code: Select all

Hou po e484:   min   wSt%   wo
Fred VanVleet  280   1.05   .99
Alperen Şengün 256   1.95  1.84
Amen Thompson  231   1.40  1.31
Jalen Green    219    .83   .76
Dillon Brooks  206    .58   .52
Steven Adams   155    .61   .80
Jabari Smith   143    .43   .59
Tari Eason     132    .76   .96
Aaron Holiday   27    .36   .52
Reed Sheppard   10    .22   .37
correlation w min:    .70   .59

The top 5 started all 7 games.

[Crow]

The sentence "Playoff teams correlate better than the bottom 14, and those getting past round 1 were better yet." right before the table made me think on quick read that you were sharing playoff data then but that was my mistake.

So it was worst regular season correlation and coaching. A less important time frame but a much longer case of "misallocation" and still quite concerning for the future. Even though playoff management was much better, it was only borderline and ultimately not good enough even against a lower seed with more misallocation and with Rockets homecourt advantage. Will want to review that series further.

In new playoff data, Rockets were 5th best on average. As low as 6th best, as high as 3rd best. Best mark on metric with weakest correlation rate.


Every metric with a RAPM component adjusts fully for quality of opponents, starters and otherwise, as shown by their specific play in the moment than than a set starter / bench rule.

[Crow]

In round 1 series loss, 4 of 5 Rockets starters were below playoff average efg%. 1 mildly, 1 moderately, 2 massively.

How does that compare on expected "shot quality"? I don't currently have a way to find that. That would be very important for judging player thinking, coaching of player choices in addition to quality of shot making. The size of the team efg% gap with Warriors and the deficit on own turnovers were both small overall. The large advantages on the other 2 factors should have been enough but they lost at the game level, especially in game 4 somehow.

Somehow, it appears was Sengun going to 36% usage but only shooting 43% efg% and 48% ts%. In addition to a bad Green performance. The game level combo was too much. The Rockets lost the 2 times Sengun's usage exceded 30% and efficiency underhelmed. The critical at the time game 5 victory came on his lowest usage/ best efficiency. The 3 Rockets wins came with / were partially caused by Sengun's best 3 passing games. Sengun the passer is an important thing to watch in future. Game 7 had pretty good Sengun assists but lousy efficiency from Sengun and Green and too much usage from Sengun, making it more like game 4 than 5. How much Sengun & Udoka "learned' / will change appears very important.

There is plenty to say / said at broader xs & os level by others, but the above is my main focus.

Using Oliver offfensive, defensive and net ratings, J Green performed the overall worst. Not a surprise to many including me but Rockets felt they had to try. Could / should have been more proactive there imo.

[Mike G]

Even though both GS and Hou avg'd 104.0 pts in their series, the Rockets still show a team WS/48 of .111.
I recorded the Dubs at .094 after that series. Something was wack and went unfixed.
Similarly, PER was Hou 15.2 - 14.3 GSW; BPM had it GSW 1.49 - 1.42 Hou.
(Still shows 1.4 for Hou, vs 1.1 in RS) --
https://www.basketball-reference.com/te ... l_advanced

These inequities were presumably carried to the conclusion for GS. I don't know if all players were docked equally or how it was allocated. The Hou .111 WS rate should win 4 or 5 games out of 7.

GS had a very NBA-avg opponent eFG% of .543/4 both before and after the Butler aquisition. They played mostly centerless in round 1, so it made some sense for Sengun to take it in against Draymond et al.

Jalen Green had eFG% of .505 in the season, but his median game was .478. This indicates a fairly high inconsistency: a few great games offset a lot more lousy ones.
In 7 playoff games, one was great (.680 and 38 pts), no more than 12 points otherwise, and 3 games < .400 eFG%.