The debut and popularization of BPM

[Mike G]

Is there anything in BPM that forces all positions toward zero?
Here's how positions seem to break down for 2013-14:

Code: Select all

pos   %Min    BPM
PG    .203    .00
SG    .210   -.55
SF    .215    .19
PF    .210    .15
C     .162    .29

Relative positional strength looks similar to other metrics, but the magnitude seems smallish.
And the recurring question: Why does the strongest position (C) play the fewest minutes?

[colts18]

And the recurring question: Why does the strongest position (C) play the fewest minutes?

Small ball. Plus the alternative centers suck so the Center number is inflated.

[DSMok1]

Is there anything in BPM that forces all positions toward zero?
Here's how positions seem to break down for 2013-14:

Code: Select all

pos   %Min    BPM
PG    .203    .00
SG    .210   -.55
SF    .215    .19
PF    .210    .15
C     .162    .29

Relative positional strength looks similar to other metrics, but the magnitude seems smallish.
And the recurring question: Why does the strongest position (C) play the fewest minutes?

No, the positional averages are completely happenstance. Over the 35 years, they are all approximately 0.

The lack of centers is an artifact of how player's positions are designated, I think. Small, not good centers are probably called forwards...

[schtevie]

Is there anything in BPM that forces all positions toward zero?
Here's how positions seem to break down for 2013-14:

Code: Select all

pos   %Min    BPM
PG    .203    .00
SG    .210   -.55
SF    .215    .19
PF    .210    .15
C     .162    .29

Relative positional strength looks similar to other metrics, but the magnitude seems smallish.
And the recurring question: Why does the strongest position (C) play the fewest minutes?

No, the positional averages are completely happenstance. Over the 35 years, they are all approximately 0.

The lack of centers is an artifact of how player's positions are designated, I think. Small, not good centers are probably called forwards...

This, to me, is a very strange and counterintuitive result. My first reference is Eli Witus' APM results (for 2007-08) where you see a very plausible pattern for defensive numbers, with average performance increasing 1 to 5. Plausible because opportunity for defensive greatness is "unfairly" offered to those tasked with defending the paint, where the best scoring opportunities arise. And sifting through Jeremias' annual xRAPM tables, my impression has always been that this "center/power forward privilege" is a highly enduring/structural feature of NBA basketball.

And then a story of approximately offsetting offensive averages can be similarly told (offensive skill and value obtains to those more agile and less huge) but there is no particular reason to believe that be coincidence these would be complete. And indeed Eli's averages for 2007-08 don't show zeroing across positions.

Mike, might you have the O/D split for BPM to post?

[Mike G]

This year, it's

Code: Select all

Pos    OBPM     DBPM     BPM
C     -1.03     1.57     .55
PF     -.46      .47     .01
SF     -.28      .03    -.25
SG      .22     -.80    -.58
PG     1.27     -.83     .44

http://www.basketball-reference.com/lea ... anced.html
And last year:

Code: Select all

pos    OBPM    DBPM     BPM
C     -1.15    1.43     .28
PF     -.33     .49     .16
SF     -.06     .24     .19
SG      .34    -.90    -.56
PG      .95    -.95    -.01

[Crow]

Neil, here is a simple blend proposal for scoring:

65% OBPM plus 35% ORPM plus 35% DBPM plus 65% DRPM.

[permaximum]

Neil, here is a simple blend proposal for scoring:

65% OBPM plus 35% ORPM plus 35% DBPM plus 65% DRPM.

How many years of data RPM has?

[schtevie]

This year, it's

Code: Select all

Pos    OBPM     DBPM     BPM
C     -1.03     1.57     .55
PF     -.46      .47     .01
SF     -.28      .03    -.25
SG      .22     -.80    -.58
PG     1.27     -.83     .44

http://www.basketball-reference.com/lea ... anced.html
And last year:

Code: Select all

pos    OBPM    DBPM     BPM
C     -1.15    1.43     .28
PF     -.33     .49     .16
SF     -.06     .24     .19
SG      .34    -.90    -.56
PG      .95    -.95    -.01

2007-08, please, for an almost straight BPM/APM comparison (recognizing that position designations will be somewhat different)?

[Mike G]

For 2008

Code: Select all

pos    OBPM     DBPM     BPM          ows48    dws48   WS/48
C     -1.39     1.30    -.09          .051     .064     .115
PF     -.71      .66    -.06          .049     .058     .108
SF      .65      .05     .70          .057     .046     .104
SG      .76     -.74     .01          .053     .039     .093
PG      .53    -1.07    -.54          .054     .040     .094

There's a strong change over time, and also big differences by Win Share estimates.
And we are definitely describing the same sets of players for this season.

[colts18]

dsmok1, I'm not sure using MPG is right for this kind of exercise. Maybe MPG isn't really correlated to RAPM goodness. By that I mean players with high MPG get a lot of possessions. RAPM regresses players with high MPG less than other players. RAPM isn't saying that those players are better. It just reduces their RAPM values less.

[Crow]

Neil, here is a simple blend proposal for scoring:

65% OBPM plus 35% ORPM plus 35% DBPM plus 65% DRPM.

How many years of data RPM has?

Right, so use RAPM for before last season.

[Crow]

If I looked at it right, Scott's suppression of 3 pt rate by the BPM formula at team level is costing team about 0.5 pts per game from the interaction term.

Does the scoring term basically disappear at team level because of wtd sum of ts% equals tm ts% so ts - tm ts% for team goes to zero and so does the scoring term? Or am I looking at it wrong? If I am correct then the way that above league avg ts% gets credit is by relative performance compared to teammate ts% inflated / deflated by the team adjustment. Probably valid but kinda a weird way to back into the personal scoring impact.

The sqrt (assist% * totreb%) varies about 16% by team. But at team BPM level this cost the Lakers well less than a tenth of a point.

[DSMok1]

If I looked at it right, Scott's suppression of 3 pt rate by the BPM formula at team level is costing team about 0.5 pts per game from the interaction term.

Does the scoring term basically disappear at team level because of wtd sum of ts% equals tm ts% so ts - tm ts% for team goes to zero and so does the scoring term? Or am I looking at it wrong? If I am correct then the way that above league avg ts% gets credit is by relative performance compared to teammate ts% inflated / deflated by the team adjustment. Probably valid but kinda a weird way to back into the personal scoring impact.

The sqrt (assist% * totreb%) varies about 16% by team. But at team BPM level this cost the Lakers well less than a tenth of a point.

I would not apply BPM to the team level at all; it has no meaning at the team level. The weights are for individual impact, not team level impact. Also note that by definition, BPM will sum to the team's overall efficiency.

[DSMok1]

dsmok1, I'm not sure using MPG is right for this kind of exercise. Maybe MPG isn't really correlated to RAPM goodness. By that I mean players with high MPG get a lot of possessions. RAPM regresses players with high MPG less than other players. RAPM isn't saying that those players are better. It just reduces their RAPM values less.

When you've got as many possessions as the players in the regression have, the fact that it's RAPM hardly matters. APM would look basically the same for the players I'm using in the regression, most of whom have many seasons of data.

[DSMok1]

Right, so use RAPM for before last season.

The comparison Neil has been doing is over all seasons back to the 1970s, so RAPM doesn't exactly fit the bill. Perhaps he'll do a modified methodology to deal with those metrics separately?