Question About DBPM ?

[feyki]

For the 1975 , Wes Unseld has a higher DBPM than Elvin Hayes . But Elvin Hayes had both more steals and blocks than Unseld. Only advantage of Unseld is Defensive Rebounds.

How is this possible , statistically ?

I don't have any information about calculation of Defensive BPM . I wish
anyone help me here .

[Statman]

For the 1975 , Wes Unseld has a higher DBPM than Elvin Hayes . But Elvin Hayes had both more steals and blocks than Unseld. Only advantage of Unseld is Defensive Rebounds.

How is this possible , statistically ?

I don't have any information about calculation of Defensive BPM . I wish
anyone help me here .

I assume it's assist rate.

I'm not joking.

Someone will probably post with the formula.

[feyki]

For the 1975 , Wes Unseld has a higher DBPM than Elvin Hayes . But Elvin Hayes had both more steals and blocks than Unseld. Only advantage of Unseld is Defensive Rebounds.

How is this possible , statistically ?

I don't have any information about calculation of Defensive BPM . I wish
anyone help me here .

I assume it's assist rate.

I'm not joking.

Someone will probably post with the formula.

Really ?

I tought there's a good reason(a solid math combination/equation) for that .

[Crow]

DBPM doesn't give accurate individualized play by play shot defense, so it is missing around 1/3 of total defensive value. I'll pass by it and see what it says but I don't consider it good enough to really trust / use beyond a basic lean in most cases.

538 Carmelo projections took a big step back when it went from 50/50 RPM / BPM to all BPM, imo.

BPM may be pretty good on average and for many middle of the distribution players but it is highly suspect on the overall tails and specific stat tails, imo.

[AcrossTheCourt]

It's because of the influence of AST%*TRB%, which is pretty big overall in the formula, even on defense.

http://www.basketball-reference.com/about/bpm.html

I also think high USG% players get penalized in DBPM just for having high USG%.

[feyki]

DBPM doesn't give accurate individualized play by play shot defense, so it is missing around 1/3 of total defensive value. I'll pass by it and see what it says but I don't consider it good enough to really trust / use beyond a basic lean in most cases.

538 Carmelo projections took a big step back when it went from 50/50 RPM / BPM to all BPM, imo.

BPM may be pretty good on average and for many middle of the distribution players but it is highly suspect on the overall tails and specific stat tails, imo.

Is there any metric , which is includes shot defence ?

It's because of the influence of AST%*TRB%, which is pretty big overall in the formula, even on defense.

http://www.basketball-reference.com/about/bpm.html

I also think high USG% players get penalized in DBPM just for having high USG%.

Thanks . I'm not a fan of BPM , honestly . But DBPM was looking interesting . And probably best defensive metric to me . But it seems not good when you guys telling me about the formula .

[Crow]

Shot defense covered here viewtopic.php?f=2&t=9223

[Dr Positivity]

It's a good observation. As mentioned it looks like TRB% and Ast% combination

I think BPM from that long ago is somewhat sketchy anyways. I mean it's better than nothing, but it's built on +/- evaluating what's valuable in the modern era, whereas 1975 is a different game with no 3s, transition play having a whole different level of meaning, no turnovers tracked and steals/blocks before 1974, etc. The league has changed a ton in even just 15 years with the rise of pick and roll dominated 3pt shooting teams vs posting up and the rise of analytics/player tracking and game preparation etc. resources, let alone 40.

[feyki]

It's a good observation. As mentioned it looks like TRB% and Ast% combination

I think BPM from that long ago is somewhat sketchy anyways. I mean it's better than nothing, but it's built on +/- evaluating what's valuable in the modern era, whereas 1975 is a different game with no 3s, transition play having a whole different level of meaning, no turnovers tracked and steals/blocks before 1974, etc. The league has changed a ton in even just 15 years with the rise of pick and roll dominated 3pt shooting teams vs posting up and the rise of analytics/player tracking and game preparation etc. resources, let alone 40.

Of course , every era different than another one . But i don't understand rebound&assist rate make the defence better . It doesn't matter , how much changed the game ; assists isn't about the defence .

Shot defense covered here viewtopic.php?f=2&t=9223

Thanks . But i don't see any formula over there . If you have a good calculation(equation) about the shot defence anywhere , share with me if you want .

Actually , Playoffs are very solid source to measure the shot defence . And with a great calculation , results could be perfect .

[Crow]

If want to put shot defense into an equation, I'd look at counterpart shot defense and team shot defense while actually on court, convert the data from NBA.com to points allowed above or below average per x minutes or 100 possessions, taking account of both effectiveness and frequency of opponent possession usage, then blend the two at a ratio of 50-70% counterpart and 50-30% team shot defense. Put that in winshares / 48 (or BPM or DRE or a new metric by you) instead of the undifferentiated value by individual and time on court that is there.

Or use DRPM in a 50/50 blend with the suggested above or alone instead of any other stat / metric for defense.

Or hope for 4 factor RPM values again or calculate them yourself and use the shot defense factor found there instead of the shot defense estimate suggested above.

( Use a box score / rpm blend for offense, perhaps at 70/30 or 50/50.)

Or use PT-PM or Dredge.

The worse option is settling for the no shot defense of PER or the very poor estimates in ws/48 or BPM. DRE's globbed on shot defense to other defensive stats might be slightly better than ws/48 or BPM at least to the level of bigs vs. perimeters, but it is probably not as good as estimating by my method or as part of DRPM or by separate RPM factor.

[DSMok1]

For the 1975 , Wes Unseld has a higher DBPM than Elvin Hayes . But Elvin Hayes had both more steals and blocks than Unseld. Only advantage of Unseld is Defensive Rebounds.<br abp="1310"><br abp="1311"> How is this possible , statistically ?<br abp="1312"><br abp="1313"> I don't have any information about calculation of Defensive BPM . I wish<br abp="1314"> anyone help me here .

Nice find, feyki!

First of all, a quick background on how DBPM is calculated. http://www.basketball-reference.com/about/bpm.html

The variables and interaction terms in BPM were developed to best match long term RAPM.

OBPM and DBPM are based on the same secondary regression, which was tuned to split the total BPM into offensive and defensive components. So all of the variables in primary BPM were used in this secondary regression, with the target of the regression being to minimize squared error for OBPM and DBPM simultaneously. So every stat is used in DBPM, not just steals and blocks. Basically the BPM regression goes:
1. What is our estimate for this player's RAPM?
2. What is our best estimate for how to split that estimate into offense and defense?

Note: this approach had equivalent or lower error than directly running a regression on DRAPM.

So, now that we know that all of the stats are involved, what's causing this situation?

Here are their stats per 100 possessions:

Code: Select all

╔════╦═════════════╦═════╦════╦════╦══════╦═════╦══════╦═══════╦════╦═════╦═════╦═════╦══════╦═══════╦═════╦═════╦═══════╦═════╦══════╦══════╦═════╦═════╦═════╦═════╦═════╦══════╦══╦══════╦══════╗
║ Rk ║   Player    ║ Age ║ G  ║ GS ║  MP  ║ FG  ║ FGA  ║  FG%  ║ 3P ║ 3PA ║ 3P% ║ 2P  ║ 2PA  ║  2P%  ║ FT  ║ FTA ║  FT%  ║ ORB ║ DRB  ║ TRB  ║ AST ║ STL ║ BLK ║ TOV ║ PF  ║ PTS  ║  ║ ORtg ║ DRtg ║
╠════╬═════════════╬═════╬════╬════╬══════╬═════╬══════╬═══════╬════╬═════╬═════╬═════╬══════╬═══════╬═════╬═════╬═══════╬═════╬══════╬══════╬═════╬═════╬═════╬═════╬═════╬══════╬══╬══════╬══════╣
║  1 ║ Elvin Hayes ║  29 ║ 82 ║    ║ 3465 ║ 9.6 ║ 21.8 ║ 0.443 ║    ║     ║     ║ 9.6 ║ 21.8 ║ 0.443 ║ 5.3 ║ 7   ║ 0.766 ║ 2.9 ║ 10.2 ║ 13.1 ║ 2.7 ║ 2.1 ║ 2.4 ║     ║ 3.1 ║ 24.6 ║  ║      ║   88 ║
║  2 ║ Wes Unseld  ║  28 ║ 73 ║    ║ 2904 ║ 4.3 ║ 8.5  ║ 0.502 ║    ║     ║     ║ 4.3 ║ 8.5  ║ 0.502 ║ 2   ║ 2.9 ║ 0.685 ║ 5   ║ 11.8 ║ 16.8 ║ 4.6 ║ 1.8 ║ 1.1 ║     ║ 2.8 ║ 10.5 ║  ║      ║   88 ║
╚════╩═════════════╩═════╩════╩════╩══════╩═════╩══════╩═══════╩════╩═════╩═════╩═════╩══════╩═══════╩═════╩═════╩═══════╩═════╩══════╩══════╩═════╩═════╩═════╩═════╩═════╩══════╩══╩══════╩══════╝

And here are their "advanced" stats:

Code: Select all

╔════╦═════════════╦═════╦════╦══════╦═════╦═══════╦══════╦═══════╦══════╦══════╦══════╦══════╦══════╦══════╦══════╦══════╦══╦═════╦═════╦══════╦═══════╦══╦══════╦══════╦═════╦══════╗
║ Rk ║   Player    ║ Age ║ G  ║  MP  ║ PER ║  TS%  ║ 3PAr ║  FTr  ║ ORB% ║ DRB% ║ TRB% ║ AST% ║ STL% ║ BLK% ║ TOV% ║ USG% ║  ║ OWS ║ DWS ║  WS  ║ WS/48 ║  ║ OBPM ║ DBPM ║ BPM ║ VORP ║
╠════╬═════════════╬═════╬════╬══════╬═════╬═══════╬══════╬═══════╬══════╬══════╬══════╬══════╬══════╬══════╬══════╬══════╬══╬═════╬═════╬══════╬═══════╬══╬══════╬══════╬═════╬══════╣
║  1 ║ Elvin Hayes ║  29 ║ 82 ║ 3465 ║  19 ║ 0.496 ║      ║ 0.32  ║ 6.4  ║ 22.7 ║ 14.5 ║ 8.7  ║ 2.1  ║ 2.9  ║      ║      ║  ║ 4.3 ║ 8.2 ║ 12.5 ║ 0.173 ║  ║ 0.3  ║ 3    ║ 3.2 ║ 4.6  ║
║  2 ║ Wes Unseld  ║  28 ║ 73 ║ 2904 ║  15 ║ 0.538 ║      ║ 0.338 ║ 11   ║ 26.2 ║ 18.6 ║ 12.7 ║ 1.8  ║ 1.3  ║      ║      ║  ║ 4   ║ 6.6 ║ 10.6 ║ 0.175 ║  ║ 0.3  ║ 5.2  ║ 5.5 ║ 5.5  ║
╚════╩═════════════╩═════╩════╩══════╩═════╩═══════╩══════╩═══════╩══════╩══════╩══════╩══════╩══════╩══════╩══════╩══════╩══╩═════╩═════╩══════╩═══════╩══╩══════╩══════╩═════╩══════╝

Let's see--first of all, the BPM regression sees Wes Unseld as likely to have the better RAPM. Why? Hard to say without diving deep into the numbers, but players with versatility (able to assist, rebound, and score) typically do well. Also, volume scoring is only valuable in BPM if the player is a passing threat as well. There's a usage x ast interaction term.

So, after calculating the overall BPM, the secondary regression splits the BPM value into two halves. The big components in this are offensive volume, steals, and blocks. So Elvin Hayes had more steals and blocks, but RAPM shows that volume scorers on the offense are typically poor defenders on the other end, their defensive stats notwithstanding. And versatile players like Unseld typically are good defenders, their defensive stats notwithstanding.

So most of Wes Unseld's impact is assigned to the defensive end.

This is a very interesting case, and I appreciate you bringing it up! Did my explanation make sense?

[DSMok1]

If want to put shot defense into an equation, I'd look at counterpart shot defense and team shot defense while actually on court, convert the data from NBA.com to points allowed above or below average per x minutes or 100 possessions, taking account of both effectiveness and frequency of opponent possession usage, then blend the two at a ratio of 50-70% counterpart and 50-30% team shot defense. Put that in winshares / 48 (or BPM or DRE or a new metric by you) instead of the undifferentiated value by individual and time on court that is there.

Or use DRPM in a 50/50 blend with the suggested above or alone instead of any other stat / metric for defense.

Or hope for 4 factor RPM values again or calculate them yourself and use the shot defense factor found there instead of the shot defense estimate suggested above.

( Use a box score / rpm blend for offense, perhaps at 70/30 or 50/50.)

Or use PT-PM or Dredge.

The worse option is settling for the no shot defense of PER or the very poor estimates in ws/48 or BPM. DRE's globbed on shot defense to other defensive stats might be slightly better than ws/48 or BPM at least to the level of bigs vs. perimeters, but it is probably not as good as estimating by my method or as part of DRPM or by separate RPM factor.

The reason I chose to limit inputs into the BPM regression as I did was so it could be applied to historical cases like the 1975 Washington Bullets. If I intended BPM to be the best statistic currently, it would be constructed to take advantage of the additional modern inputs. There are several modern stats that measure defense better than BPM.

[feyki]

For the 1975 , Wes Unseld has a higher DBPM than Elvin Hayes . But Elvin Hayes had both more steals and blocks than Unseld. Only advantage of Unseld is Defensive Rebounds.<br abp="1310"><br abp="1311"> How is this possible , statistically ?<br abp="1312"><br abp="1313"> I don't have any information about calculation of Defensive BPM . I wish<br abp="1314"> anyone help me here .

Nice find, feyki!

First of all, a quick background on how DBPM is calculated. http://www.basketball-reference.com/about/bpm.html

The variables and interaction terms in BPM were developed to best match long term RAPM.

OBPM and DBPM are based on the same secondary regression, which was tuned to split the total BPM into offensive and defensive components. So all of the variables in primary BPM were used in this secondary regression, with the target of the regression being to minimize squared error for OBPM and DBPM simultaneously. So every stat is used in DBPM, not just steals and blocks. Basically the BPM regression goes:
1. What is our estimate for this player's RAPM?
2. What is our best estimate for how to split that estimate into offense and defense?

Note: this approach had equivalent or lower error than directly running a regression on DRAPM.

So, now that we know that all of the stats are involved, what's causing this situation?

Here are their stats per 100 possessions:

Code: Select all

╔════╦═════════════╦═════╦════╦════╦══════╦═════╦══════╦═══════╦════╦═════╦═════╦═════╦══════╦═══════╦═════╦═════╦═══════╦═════╦══════╦══════╦═════╦═════╦═════╦═════╦═════╦══════╦══╦══════╦══════╗
║ Rk ║   Player    ║ Age ║ G  ║ GS ║  MP  ║ FG  ║ FGA  ║  FG%  ║ 3P ║ 3PA ║ 3P% ║ 2P  ║ 2PA  ║  2P%  ║ FT  ║ FTA ║  FT%  ║ ORB ║ DRB  ║ TRB  ║ AST ║ STL ║ BLK ║ TOV ║ PF  ║ PTS  ║  ║ ORtg ║ DRtg ║
╠════╬═════════════╬═════╬════╬════╬══════╬═════╬══════╬═══════╬════╬═════╬═════╬═════╬══════╬═══════╬═════╬═════╬═══════╬═════╬══════╬══════╬═════╬═════╬═════╬═════╬═════╬══════╬══╬══════╬══════╣
║  1 ║ Elvin Hayes ║  29 ║ 82 ║    ║ 3465 ║ 9.6 ║ 21.8 ║ 0.443 ║    ║     ║     ║ 9.6 ║ 21.8 ║ 0.443 ║ 5.3 ║ 7   ║ 0.766 ║ 2.9 ║ 10.2 ║ 13.1 ║ 2.7 ║ 2.1 ║ 2.4 ║     ║ 3.1 ║ 24.6 ║  ║      ║   88 ║
║  2 ║ Wes Unseld  ║  28 ║ 73 ║    ║ 2904 ║ 4.3 ║ 8.5  ║ 0.502 ║    ║     ║     ║ 4.3 ║ 8.5  ║ 0.502 ║ 2   ║ 2.9 ║ 0.685 ║ 5   ║ 11.8 ║ 16.8 ║ 4.6 ║ 1.8 ║ 1.1 ║     ║ 2.8 ║ 10.5 ║  ║      ║   88 ║
╚════╩═════════════╩═════╩════╩════╩══════╩═════╩══════╩═══════╩════╩═════╩═════╩═════╩══════╩═══════╩═════╩═════╩═══════╩═════╩══════╩══════╩═════╩═════╩═════╩═════╩═════╩══════╩══╩══════╩══════╝

And here are their "advanced" stats:

Code: Select all

╔════╦═════════════╦═════╦════╦══════╦═════╦═══════╦══════╦═══════╦══════╦══════╦══════╦══════╦══════╦══════╦══════╦══════╦══╦═════╦═════╦══════╦═══════╦══╦══════╦══════╦═════╦══════╗
║ Rk ║   Player    ║ Age ║ G  ║  MP  ║ PER ║  TS%  ║ 3PAr ║  FTr  ║ ORB% ║ DRB% ║ TRB% ║ AST% ║ STL% ║ BLK% ║ TOV% ║ USG% ║  ║ OWS ║ DWS ║  WS  ║ WS/48 ║  ║ OBPM ║ DBPM ║ BPM ║ VORP ║
╠════╬═════════════╬═════╬════╬══════╬═════╬═══════╬══════╬═══════╬══════╬══════╬══════╬══════╬══════╬══════╬══════╬══════╬══╬═════╬═════╬══════╬═══════╬══╬══════╬══════╬═════╬══════╣
║  1 ║ Elvin Hayes ║  29 ║ 82 ║ 3465 ║  19 ║ 0.496 ║      ║ 0.32  ║ 6.4  ║ 22.7 ║ 14.5 ║ 8.7  ║ 2.1  ║ 2.9  ║      ║      ║  ║ 4.3 ║ 8.2 ║ 12.5 ║ 0.173 ║  ║ 0.3  ║ 3    ║ 3.2 ║ 4.6  ║
║  2 ║ Wes Unseld  ║  28 ║ 73 ║ 2904 ║  15 ║ 0.538 ║      ║ 0.338 ║ 11   ║ 26.2 ║ 18.6 ║ 12.7 ║ 1.8  ║ 1.3  ║      ║      ║  ║ 4   ║ 6.6 ║ 10.6 ║ 0.175 ║  ║ 0.3  ║ 5.2  ║ 5.5 ║ 5.5  ║
╚════╩═════════════╩═════╩════╩══════╩═════╩═══════╩══════╩═══════╩══════╩══════╩══════╩══════╩══════╩══════╩══════╩══════╩══╩═════╩═════╩══════╩═══════╩══╩══════╩══════╩═════╩══════╝

Let's see--first of all, the BPM regression sees Wes Unseld as likely to have the better RAPM. Why? Hard to say without diving deep into the numbers, but players with versatility (able to assist, rebound, and score) typically do well. Also, volume scoring is only valuable in BPM if the player is a passing threat as well. There's a usage x ast interaction term.

So, after calculating the overall BPM, the secondary regression splits the BPM value into two halves. The big components in this are offensive volume, steals, and blocks. So Elvin Hayes had more steals and blocks, but RAPM shows that volume scorers on the offense are typically poor defenders on the other end, their defensive stats notwithstanding. And versatile players like Unseld typically are good defenders, their defensive stats notwithstanding.

So most of Wes Unseld's impact is assigned to the defensive end.

This is a very interesting case, and I appreciate you bringing it up! Did my explanation make sense?

Yes , absolutely .

In my calculations , passing is definitely most important aspect of the offence . Not just feeding teammates , your score also coming from yourself more than who has less assists (or other named unassisted points ) . I'd agree with that on offence . And I'd rank Wes as good as Hayes on the offensive end , if not better .

I also agree with more versatile player , being a better defender . But I think on the defensive side , versatility can be measured by Steals/Blocks ( For example , Garnett/Draymond/Horry , those trio had 2 steals and 2 blocks on 100 poss in their primes . Or having more than 1.5 steals by Centers like Wes and Cowens ) .

Of course , it's my point and I'm glad to see that detailed answer .

If want to put shot defense into an equation, I'd look at counterpart shot defense and team shot defense while actually on court, convert the data from NBA.com to points allowed above or below average per x minutes or 100 possessions, taking account of both effectiveness and frequency of opponent possession usage, then blend the two at a ratio of 50-70% counterpart and 50-30% team shot defense. Put that in winshares / 48 (or BPM or DRE or a new metric by you) instead of the undifferentiated value by individual and time on court that is there.

Or use DRPM in a 50/50 blend with the suggested above or alone instead of any other stat / metric for defense.

Or hope for 4 factor RPM values again or calculate them yourself and use the shot defense factor found there instead of the shot defense estimate suggested above.

( Use a box score / rpm blend for offense, perhaps at 70/30 or 50/50.)

Or use PT-PM or Dredge.

The worse option is settling for the no shot defense of PER or the very poor estimates in ws/48 or BPM. DRE's globbed on shot defense to other defensive stats might be slightly better than ws/48 or BPM at least to the level of bigs vs. perimeters, but it is probably not as good as estimating by my method or as part of DRPM or by separate RPM factor.

Of course , datas of modern days makes it easy(with play by play tech) . And also , there are many shot tracking defence works along with past 2-3 years . And those are very useful . But I'm looking for , only statistical (to measure historical players) .

Thanks for advices . I can't determine on/off court shot defence . But I think separating area's of the court and judging players by their shot defences in the playoffs . Regular season isn't looking possible , but playoffs already a better choice with a few reasons ( players are always be more concantrated for the playoffs . And shot defence quality determined based on regular season level of offensive player . )