Usage and efficiency stats

[Mike G]

It is an enduring mystery why ORtg and DRtg are still considered to be equally good approximations of what they purport to measure. They aren't. Full stop. ORtg is pretty good, whereas DRtg is really quite terrible...

I agree with all of this. See the OP. ORtg is almost meaningless without Usg%. It's also not very meaningful without DRtg.
DRtg alone doesn't tell us much without ORtg and Usg. Just as there's not much information in just a person's height, or just his weight, or just his fat content. Together, they tell us something.

Given that we can concoct a combination that in fact seems to concur with a player's known best season, maybe we are on to something.

If you aren't playing much D, or are part of a team that doesn't play much D, what does that say about your offensive stats? That they aren't translatable to a more competitive environment?
Yet if you have at your disposal something like DRtg -- basically your team's DRtg plus or minus a quantity that depends on your blocks and steals -- then you have something better than league ORtg to compare to your ORtg -- it's the avg league ORtg of the games in which you were actually playing.

DRtg doesn't have to be an "equally good approximation" to still be better than nothing.
How great is an ORtg of 113 with TS% of .565? Orlando Woolridge did this while scoring 26 pts/36 in 1991.
His 7.1 Reb/36 was above average in the league; but his Reb% of 9.3 was below avg. That's what happens when your team is out-rebounded by 3 per game.
His DRtg of 116 was just a bit worse than his team's 114.7

[schtevie]

Y'know, upon a moment's more of reflection, it occurs to me that I shouldn't have been so down on DRtg relative to ORtg. At least with the former, the overall range of estimates approximates what +/- data reveal reality to be (the point of proper apportionment of credit, for the box score statistical void, still applying).

The same simply cannot be said for ORtg, where the range of estimates (as revealed by these most excellent player seasons) significantly exceeds even that suggested by APM.

Given these facts, why should a metric with a built in offensive bias (i.e. USG%), reinforced by more offensive bias (ORtg^2), and rendered somewhat imprecise (1/DRtg^2) be relied upon to indicate overall player greatness?

[Mike G]

I don't know that I'd call Usg% a 'bias' for offense, when it simply distinguishes those who are more active on offense.
It's not a bias 'against defense'; it's neutral.

DRtg is largely the team DRtg.
It looks as if the proper exponent is 2.5

[schtevie]

But, of course, it is a biased measure. Consider the extreme hypothetical: a ham-handed defensive supah-genius, having the ability to prevent any and all from scoring in the paint, but who on offense would turn the ball over every time the ball was touched, were he given such an opportunity. Such a player would (hypothetically) be the greatest of all time, but would have a (rational) USG% of zero. Next, relax the assumption, but think of truly great defensive specialists. They are completely ill-served by this metric.

[Mike G]

Sigh, 'tis true. Ben Wallace had 10.5 Usg% -- and 10% is about as low as any major player goes -- along with O/D of 117/93 at best.
Thinking of an adjusted Usage% = (20+Usg)/2
This acknowledges that 1/2 of a player's presence is on defense, gives that an avg value of 20.
With that and a 2.5 exponent on O/D, Ben registers a 27.2 -- still below all those high Users.

[Mike G]

Another thread inspired me to take up this topic again.
A formula to approximate WS/48 when knowing just ORtg, DRtg, and Usg% is:
altWS/48 = (Usg+40)*(ORtg/DRtg)^(3.33)/334 - .087

This version does not seem to produce such extremely bad/weak WS/48 rates.
b-r.com lists 48 players with >100 min this year and negative WS/48. Their avg is -.046
altWS gives the same 48 players an avg of -.016; 12 are above zero.

Kobe has WS/48 of -.096 this year; in the alt version, he's -.006
Yet league WS add up to the same total. They've just been redistributed.

Here are the top 20 players in minutes played, comparing the 2:

Code: Select all

. Player        WS/48    alt         Player         WS/48    alt
Carmelo Anthony .169    .158        LeBron James    .265    .264
Kevin Durant    .306    .307        Josh Smith      .021    .043
DeMar DeRozan   .142    .144        Klay Thompson   .113    .112
Monta Ellis     .078    .082        Stephen Curry   .219    .207
Nicolas Batum   .127    .116        Kyle Lowry      .193    .182
Paul George     .181    .196       Lance Stephenson .130    .122
John Wall       .129    .129        Wesley Matthews .140    .138
Damian Lillard  .158    .147       Chandler Parsons .129    .126
DeAndre Jordan  .188    .237        Jeff Green      .056    .066
Blake Griffin   .205    .202        Joakim Noah     .191    .198

Not much difference for most players. Most are within the rounding error you get from lack of a decimal place in b-r.com's ORtg and DRtg.

[Mike G]

Last year's Finals summaries can be seen here -- http://www.basketball-reference.com/pla ... inals.html
They show ORtg and DRtg but not WS or WS/48. By the formula above, we get:

Code: Select all

Miami Heat     Min    TS%   Usg%  ORtg  DRtg   w/48    W
LeBron James   301   .529   29.7   113   105   .179  1.11
Chris Bosh     240   .491   18.5   108   104   .112   .55
Shane Battier   87   .689   15.4   141   110   .292   .52
Dwyane Wade    255   .505   28.5   104   108   .094   .49
Mike Miller    152   .841    7.7   131   112   .154   .48

Ray Allen      193   .715   15.2   121   116   .103   .41
Chris Andersen  71   .714   13.6   118   103   .165   .24
Joel Anthony    15   .500    6.8   146   117   .206   .06
Udonis Haslem   64   .477    8.1   102   114   .012   .02
James Jones     12   .600   25.0    96   121   .003   .00

Rashard Lewis   12   .333   33.0    65   125  -.062  -.02
Mario Chalmers 221   .509   19.7    87   112  -.010  -.05
Norris Cole     82   .328   16.8    75   115  -.046  -.08

. total                                              3.74
                                
. Spurs        Min    TS%   Usg%  ORtg  DRtg   w/48    W
Kawhi Leonard  255   .583   17.5   120   103   .199  1.04
Tim Duncan     254   .554   25.2   112   104   .163   .85
Danny Green    248   .649   15.8   115   108   .119   .61
Tony Parker    246   .472   25.4   103   113   .057   .29
Manu Ginobili  199   .560   22.4   101   112   .045   .19

Boris Diaw      94   .615   12.4   116   116   .070   .13
Nando De Colo    5     .0    0     226   118   .956   .10
Gary Neal      171   .552   18.6   100   115   .023   .08
Matt Bonner     36   .506   14.0   111   106   .101   .08
DeJuan Blair    17   .446   32.6    95    95   .130   .05

Tiago Splitter 107   .509   18.7    91   107   .015   .03
Tracy McGrady   14   .000    9.5    98   103   .039   .01
Patrick Mills   13   .400   17.3    93   115  -.002   .00
Cory Joseph     43   .473   18.0    85   106  -.004   .00

. total                                              3.45

The total Wins add to 7.19
De Colo had a DReb and an Ast in 5 minutes (no shots); throw out his garbage time, and we're closer.
Leonard also leads the Spurs in WS/48 for this season.