The debut and popularization of BPM

[Crow]

Interesting article about roles and persistance of performance of different players in those roles.

I have looked a fair amount at maximum usage on good and title winning teams. Probably should look for patterns in usage amongst starting five positions and rotation 10 given this information.

[Statman]

By the way--RPM agrees that Jeff Green is at or below replacement level: http://espn.go.com/nba/statistics/rpm/_ ... position/5

My work had a team of Jeff Green 13-14 caliber players at winning just over 22 games. Not that anyone cares, I just felt like looking it up for fun.

[Statman]

I have 2 simple questions... Not just for me but everyone else.

1. Where should we use the metric?
2. Why should we use it instead of RAPM or something else?

1. Best box score metric, has the advantages and disadvantages of box score metrics.
2. When a good quality RAPM is not available or doesn't measure what you want. RAPM is good for a specific situation: estimating true talent level over a period of months and years. If you want a stat that will stabilize quickly over smaller periods of time, BPM is far, far better than RAPM. For instance--who has been playing best this year? 5 games in, BPM can give you a good idea, RAPM cannot. Also, if you don't have access to lineup data (history or other leagues), then BPM is the best option. Even with multi-year samples for estimating true talent level, RAPM can still struggle with multicollinearity issues, hence the need for RPM/xRAPM. Even at the end of this season--if you asked me who played better this year, I think BPM is more accurate than RAPM, and perhaps even xRAPM/RPM (since prior years are included). (Remember, though, BPM doesn't capture defense extremely well.)

I still believe I have the best box score metrics - but I am biased. I would put mine up against anybody's fwiw.

[Crow]

It is a little late for a fair entry into this season's friendly competition, but if you feel this way you could still fashion team projections from your metric and run that test. Or do so next season. You haven't entered the contest here in at least the last 2 seasons, if I read it right. Have you made team projections from your metrics at your site previously? If so how did they do?

[Statman]

It is a little late for a fair entry into this season's friendly competition, but if you feel this way you could still fashion team projections from your metric and run that test. Or do so next season. You haven't entered the contest here in at least the last 2 seasons, if I read it right. Have you made team projections from your metrics at your site previously? If so how did they do?

I will do it Crow, I'm just so far behind on projects right now. I am confident the results will be quite solid.

I may be throwing my hat in the ring on this attempting to quantify defense based solely on conventional box score data. I'm not sure I'm comfortable with some of the things I read out there - like including height into the regressions. It seems too much like trying to quantify the best power hitters in baseball, and and including player weight into the power number/field/opposing pitcher regressions. Yes, heavy baseball players usually hit more homers, BUT that doesn't mean every heavy player should automatically be given credit for power performance beyond the actual results.

I have what I think are some very good ideas rattling around in my brain in that respect, we'll see. When I get that idea ironed out, I'll probably run some WAY too late projections.

BTW Crow, since I very well may have missed it and you seem to catch everything - is there a comparison somewhere of many/all of these metrics that takes the actual metric's player results, and plugs them back into the season using actual minutes played of every player to make sure all the team totals compile correctly to a very solid team ranking comparison? PER did a horrible job of this years ago when I did it. If this simple thing can't be done correctly (player results compile correctly to mimic proper team strength), then I believe it should be back to the drawing board. I'd do these comparisons myself - but I can't even keep up with the stat demons in my brain to even check other people's work.

Also, maybe someone can explain to me why I should completely trust a metric that uses an approach that works best on HUGE amounts of data (which we don't have), uses multiple previous season data, none performance data (player heights, age?), etc that results in some guys ranking near the top of the league in which their own fans think is completely crazy. Real plus minus had a top 20 last season that included Nick Collison, Frye, Amir Johnson, Vince Carter, & Patrick Beverley. So, I should believe a team with a starting lineup of those guys would be one of the very best in the league? If not, what is the metric really measuring outside of best team fit? Also, how can I trust any metric that has MANY players every year who play big minutes in the NBA as BELOW replacement level? Are most NBA coaches just complete and utter idiots - completely fooled by their own eyes?

Isn't this the same board that vilifies Dave Berri (justifiably I might add imo) because of how "off" his results seem/are? I mean - he and his disciples SWEAR that the regressions they run at the team level prove that their weights at the player level are correct. How can I really trust ANYBODY'S regressions?

[Crow]

Alex the sports skeptic looked at how metrics did at explaining actual (reference in threads worth reading thread) but many of the big name metrics don't change. ASPM /BPM and RPM have.

[DSMok1]

I still believe I have the best box score metrics - but I am biased. I would put mine up against anybody's fwiw.

Perhaps they are better; I just haven't seen them tested. BPM/ASPM is the best I have seen tested publicly, though it looks like Neil's SPM may be equal or superior.

[Neil Paine]

I still believe I have the best box score metrics - but I am biased. I would put mine up against anybody's fwiw.

The proof is in the pudding. Send me your data (preferably going back to the merger, or at least the turnover era), and I'll see how well it predicts out of sample. My email is neil.paine@fivethirtyeight.com.

[DSMok1]

I still believe I have the best box score metrics - but I am biased. I would put mine up against anybody's fwiw.

The proof is in the pudding. Send me your data (preferably going back to the merger, or at least the turnover era), and I'll see how well it predicts out of sample. My email is neil.paine@fivethirtyeight.com.

Keep adding the metrics in! Any others of note that are missing?

I'd like to see a very simple "ReMPG" metric somewhere--basically, just a simple minutes per game metric x a coefficient.

[jbrocato23]

I still believe I have the best box score metrics - but I am biased. I would put mine up against anybody's fwiw.

The proof is in the pudding. Send me your data (preferably going back to the merger, or at least the turnover era), and I'll see how well it predicts out of sample. My email is neil.paine@fivethirtyeight.com.

Keep adding the metrics in! Any others of note that are missing?

I'd like to see a very simple "ReMPG" metric somewhere--basically, just a simple minutes per game metric x a coefficient.

Might as well send my boxscore metric (estimated impact) along just because I'd like to see it tested by someone else...I'll email you neil

[kmedved]

I still believe I have the best box score metrics - but I am biased. I would put mine up against anybody's fwiw.

The proof is in the pudding. Send me your data (preferably going back to the merger, or at least the turnover era), and I'll see how well it predicts out of sample. My email is neil.paine@fivethirtyeight.com.

Can you explain how you're going to test out of sample? The rest of the season?

For example, using Minitab's ease of calculating PRESS, I've built some box score metrics with an R^2 out .90+ out of sample, but that's just extreme data mining. I'm pretty sure it's just generating junk, but grades out incredibly well, depending how you look at it.

[DSMok1]

I still believe I have the best box score metrics - but I am biased. I would put mine up against anybody's fwiw.

The proof is in the pudding. Send me your data (preferably going back to the merger, or at least the turnover era), and I'll see how well it predicts out of sample. My email is neil.paine@fivethirtyeight.com.

Can you explain how you're going to test out of sample? The rest of the season?

For example, using Minitab's ease of calculating PRESS, I've built some box score metrics with an R^2 out .90+ out of sample, but that's just extreme data mining. I'm pretty sure it's just generating junk, but grades out incredibly well, depending how you look at it.

Here's how Neil did it last time: http://www.apbr.org/metrics/viewtopic.p ... 343#p15334

[Crow]

Neil,it would be great if you include a metric blend.

From these threads http://sportskeptic.wordpress.com/2012/ ... ect-blend/
http://sportskeptic.wordpress.com/2012/ ... nd-update/

or new.

[Mike G]

The coefficients:
ORB% 0.137600 100.0
DRB% -0.151938 100.0
sqrt(AST%*TRB%) 0.691501 100.0*100.0

The total value from these terms would be:
ORB%: .137*12.4 = +1.7
DRB%: -.152*23.5 = -3.6
sqrt(AST%*TRB%): 0.691*sqrt(9.0*17.6) = +8.7

So the total of these rebounding terms for Anthony Davis, despite his mediocre assist percentage, is +6.8.
.

Following this example, let us compare Moses Malone's 1983 (MVP) season with Charles Barkley in 1987. Here's how they look:
http://bkref.com/tiny/IUYtV
They have identical PER (25.1); Moses with a big edge according to WS/48 -- .248 vs .210
Some other strong similarities:

Code: Select all

player  year   ORb%   DRb%   TRb%   Ast%    ORb    DRb  Reb*Ast   T    obpm  dbpm   BPM
Moses   1983   16.8   25.9   21.6    5.1    2.3   -3.9    7.3    5.6    2.6   0.8   3.4
Barkley 1987   16.7   24.8   20.8   19.0    2.3   -3.8   13.7   12.3    6.2   2.6   8.9

Charles leads Moses by 6.7 in the Reb/Ast terms, and by 5.5 in the overall BPM. Was Moses better by 1.2 in other areas?

The 1983 leader board was dominated by Moses, before these latest stats:

Code: Select all

Win Shares        BPM             VORP
Moses    15.1    Bird     7.1    Bird     6.8
Bird     14.0    Magic    7.0    Magic    6.6
...              Erving   5.9    Moncrief 5.0
.                Moncrief 5.3    English  4.8
WS/48           Marques J 4.6    Erving   4.8
Moses    .248    English  4.4   Marques J 4.7
Moncrief .233    Nance    4.3    Nance    4.6
...              Ruland   4.1    Ruland   4.4
.                B Jones  4.1    Moses    4.0
PER              Parish   3.6
Moses    25.1    Moses    3.4
English  24.1
Bird     24.1

http://www.basketball-reference.com/lea ... aders.html

Moses hit 4.1 and 4.0 BPM in earlier seasons.
Brad Miller also had 2 years at 4.0
Vlade Divac had 4.5, 4.8, and 4.9

[Statman]

I still believe I have the best box score metrics - but I am biased. I would put mine up against anybody's fwiw.

The proof is in the pudding. Send me your data (preferably going back to the merger, or at least the turnover era), and I'll see how well it predicts out of sample. My email is neil.paine@fivethirtyeight.com.

Cool - will do Neil, I'd love to have you pick my work apart. Someone else looking at it might "see" things that maybe I've been blind to.

I can send you all NBA player ratings back to 1980, season by season. I'll send you may general rating age curve adjustments also.

The average NBA player is 100 - so it should be super easy for you to work with. I'll send the HnI ratings, those are the ones I base my WAR off of, and they ignore missed games, so it's (theoretically) my best projection rating.

My work is based off of straight full season conventional stats (obviously adjusted for pace/league/etc) - so as best to compare past and current seasons and create player career curves. There isn't anything I work into the rating that isn't garnered from conventional player & team stats.