Can players learn from BPM?

[ampersand5]

(side note - I continued to be amazed by the elegance of statistical plus minus. Thank you so much for everyone who contributed to the development and publishing of BPM)

Last night, both Jonas Valanciunas and Enes Kanter put up excellent box score stats as their team's picked up opening night victories.
Jonas: 32:17 min, 8-11 shooting, 5-6 from the line,+8 , 15 rebounds, 1 block, 21 points
Kanter: 24:05 minutes, 7-11 shooting, 6 offensive and 16 total rebounds, 15 points.

Both players have struggled so far in their careers and have an unclear promise of growth. To fans of both teams, yesterday was a great sign of things to come. I mean come on! look at those great stats! how could they not have played great.
However, with only boxscore stats, BPM is sophisticated enough to tell us that despite the prima facie great numbers, both players played to the exact same level they did in prior years, and did not make a positive impact on yesterday's game.

The reason I make this thread is in regards to how BPM knows they did not make a positive contribution. BPM doesn't actually know anything as it is just a formula that merely guesses how much of an impact a player made based on a series of variables (based on a history of strong prediction power). From a cursory glance, it appears as if these two players have subpar BPM numbers due to a low steal% and assist%. My question is... if these two players were instructed to prioritize getting steals and assists, just to improve their BPM number (imagine that this actually took place), would the players be materially better? or just have a materially higher BPM number?


To put it in another way - is BPM revealing a basketball truth to us by showing poor numbers for JV and Kanter? Are these coefficients a proxy for an entire style of play, or are they just a proxy for the actions that lead to the stats?

Would these two be better basketball players if they adjusted their game solely to appease BPM?

[Mike G]

Code: Select all

Would these two be better basketball players if they adjusted their game solely to appease BPM?

That would indeed be a perfect stat. You could pay players based on it, make trades based on it, etc.
However, we know that BPM has a sometimes enormous team adjustment. If your team doesn't benefit in scoring differential, you don't get as much credit. All 16-8-4 lines are not equal.

Some stats are harsh on turnovers; many players would get lots more TO if they tried to get more Ast. You raise one stat at the expense of another. The tradeoff may land you on the bench.

[Statman]

Would these two be better basketball players if they adjusted their game solely to appease BPM?

All they have to do is get assists with their solid rebound totals to rank MUCH higher. LOW assist totals for rebounders KILLS their BPM. It's my biggest caveat with the rating. Moses Malone, Mutambo, Mourning, etc. rank MUCH lower by BPM than the general fan or scout would rank them.

So, get an extra assist a game - viola! Much better basketball player!

[Nate]

...
Would these two be better basketball players if they adjusted their game solely to appease BPM?

IMO people read way too much into these composite box score stats. The box scores themselves are a manifestation of the human tendency to focus on the ball and little on everything else, and anything driven by them will suffer from the same bias. We see a lot of criticism of people who over-focus on scoring, but any box-score driven individual metric will have similar issues from over-focusing on the ball.

Centers contribute a lot to the team winning by controlling very valuable key space. If they leave the low post open to chase the ball, that's almost certainly worse basketball. For example, compared to other NBA centers, Dwight Howard discourages high-value shots by occupying the space under the basket, but that's not going to be captured by his box scores at all so that kind of contribution ends up (I think) as part of the team adjustment.

[Mike G]

Any player who gets Minutes will have a bigger box score presence. MPG is the coach's estimate of the player's value. Some overall metrics actually incorporate mpg in their formulae.

Those which do not appreciate mpg are deliberately ignoring the obvious. They may enjoy the position of feeling superior to the coaching minds entrusted to lineup construction. Yet there are many supporting arguments : A player may go all-out for 5-10 minutes, and not sustain that productivity over 30-40. He may have defensive or other 'non boxscore' shortcomings that limit his effectiveness, and hence his minutes. His productive numbers may cut into his teammates' productions.

Rather than saying, "this guy should be playing more", we should wonder why he doesn't. Instead of declaring, "this player hurts his team", we can look for the ways he helps his team.

[DSMok1]

(side note - I continued to be amazed by the elegance of statistical plus minus. Thank you so much for everyone who contributed to the development and publishing of BPM)

Last night, both Jonas Valanciunas and Enes Kanter put up excellent box score stats as their team's picked up opening night victories.
Jonas: 32:17 min, 8-11 shooting, 5-6 from the line,+8 , 15 rebounds, 1 block, 21 points
Kanter: 24:05 minutes, 7-11 shooting, 6 offensive and 16 total rebounds, 15 points.

Both players have struggled so far in their careers and have an unclear promise of growth. To fans of both teams, yesterday was a great sign of things to come. I mean come on! look at those great stats! how could they not have played great.
However, with only boxscore stats, BPM is sophisticated enough to tell us that despite the prima facie great numbers, both players played to the exact same level they did in prior years, and did not make a positive impact on yesterday's game.

The reason I make this thread is in regards to how BPM knows they did not make a positive contribution. BPM doesn't actually know anything as it is just a formula that merely guesses how much of an impact a player made based on a series of variables (based on a history of strong prediction power). From a cursory glance, it appears as if these two players have subpar BPM numbers due to a low steal% and assist%. My question is... if these two players were instructed to prioritize getting steals and assists, just to improve their BPM number (imagine that this actually took place), would the players be materially better? or just have a materially higher BPM number?


To put it in another way - is BPM revealing a basketball truth to us by showing poor numbers for JV and Kanter? Are these coefficients a proxy for an entire style of play, or are they just a proxy for the actions that lead to the stats?

Would these two be better basketball players if they adjusted their game solely to appease BPM?

BPM looks at the correlation of box score stats to productivity. If a player specifically targets the box score stats that lead to BPM, I would assume there would likely be SOME improvement in actual productivity, but less, and perhaps much less, than that reflected by the associated improvement in BPM. If a player targets steals, there could actually be a negative impact on actual productivity, as there is a well-recognized trade off to gambling for steals. Targeting assists would likely not be so problematic, as there is no direct negative trade off potential.

[Crow]

DSMok1, have you considered

1) using a team assist rate while on the court instead of or in addition to assist rate given that directing the ball to the scorer is mostly a perimeter player task? Screeners, cutters, post threats and spacers contribute to team assist making but in a strict box score system get none of the indirect credit.

2) using role adjustments for assists (for PG, wings and bigs)?
And perhaps the same for rebounding.

The extreme penalties for doing little assisting or rebounding remains a reason why I am not as comfortable using BPM as I would want to be. Versatility, completeness are good things, especially if you are in the middle of the positional range and your role calls for it / offers good opportunity for it but maximum versatility is not as necessary at the ends of the position / role spectrum, at least not if the team numbers for assists and rebounds don't show signs of suffering when a certain player is on the court.

RPM in its holistic rating approach addresses these concerns but I think BPM could too to at least a degree, if the decision was made to try. The team adjustment already takes it beyond the individual box score so to me further adjusting wouldn't in itself change BPM from pure box score to hybrid because it is already hybrid.

[Crow]

How much are public metrics actually used in player salary negotiations? NBA offices are now pretty deep in analytic staff. Some may presume that they look at public metrics in addition to all the discrete stats and visual based new data but how much do they use roll-up metrics for evaluation and separately for negotiations? Do they use a private metric or one or more public metric or both? I have been told that agents are up to speed in awareness and even usage of public roll-up metrics in negotiations but I have no basis to confirm this. I gave never heard a team or player agent publicly site a rollup metric in the public part of negotiations or negotiation hype. Some of the media has started referring to WinShares, RPM like PER before it but will we get to the point where GMs and agents will use them in public? Do they negotiate in private which metrics to validate for discussion? I can see that in most discussions one side or the other would have an incentive to try to dismiss a metric that didn't advance their cause but a lot of the talk is just talk and the level of acknowledgement of its credence is a side show to the siding of bottom-line contract salary figures back and forth

Will we get to the point where rabid fans recognize the archaic distorted nature of 9-10 category fantasy basketball and opt for a cleaner metric based system? Or is the complicated system's flaws the fundamental attraction, to learn how to game the system regardless of the real values of the categories relative to each other?

[xkonk]

Targeting assists would likely not be so problematic, as there is no direct negative trade off potential.

As Mike noted above, what about turnovers?

Also, it would certainly be more difficult for some players to get assists because they would have to get the ball in their hands in the first place. Big men in particular only have so many opportunities with the ball, and would have to give up shot attempts to hunt assists. On the other hand, anyone who's willing to freelance can go for a steal.

Unrelated, ampersand, where did you find BPM for individual games this season? And are we sure Jonas really had a bad (or irrelevant) game? He had the 15th best game score out of 300 players that night (http://www.basketball-reference.com/fri ... &year=2015). He had the highest offensive rating and second-lowest/best defensive rating of anyone in the game. His true shooting was 77% ! That all sounds pretty good to me.

[ampersand5]

Any player who gets Minutes will have a bigger box score presence. MPG is the coach's estimate of the player's value. Some overall metrics actually incorporate mpg in their formulae.

Those which do not appreciate mpg are deliberately ignoring the obvious. They may enjoy the position of feeling superior to the coaching minds entrusted to lineup construction. Yet there are many supporting arguments : A player may go all-out for 5-10 minutes, and not sustain that productivity over 30-40. He may have defensive or other 'non boxscore' shortcomings that limit his effectiveness, and hence his minutes. His productive numbers may cut into his teammates' productions.

Rather than saying, "this guy should be playing more", we should wonder why he doesn't. Instead of declaring, "this player hurts his team", we can look for the ways he helps his team.

I don't disagree. It's also worth noting that MPG is a component of BPM.

[ampersand5]

Targeting assists would likely not be so problematic, as there is no direct negative trade off potential.

As Mike noted above, what about turnovers?

Also, it would certainly be more difficult for some players to get assists because they would have to get the ball in their hands in the first place. Big men in particular only have so many opportunities with the ball, and would have to give up shot attempts to hunt assists. On the other hand, anyone who's willing to freelance can go for a steal.

Unrelated, ampersand, where did you find BPM for individual games this season? And are we sure Jonas really had a bad (or irrelevant) game? He had the 15th best game score out of 300 players that night (http://www.basketball-reference.com/fri ... &year=2015). He had the highest offensive rating and second-lowest/best defensive rating of anyone in the game. His true shooting was 77% ! That all sounds pretty good to me.

Since it was the first game of the season, I just looked at his season total BPM...

In regards to the turnover-assist issue as it pertains to JV. Jonas does not pass the ball. It's not that he gets close to 0 assists per game, its that he gets close to 0 passes per game. I cannot stress this enough, JV will not look to pass at all, he will just drive. This is likely a specific strategy of the Raptors because Jonas can individually score at a more efficient rate than the team as a whole. What I am ultimately curious about is that despite Jonas scoring personally at an efficient rate, if his presence on the floor makes the lineup as a whole worse because he cannot play in the flow of the offence.

The other four players are now much more limited in what they can do.

[BasketDork]

Quick question, when BPM was released, I tinkered with the formula and team adjustments in of all places, the fantasy realm, not H2H or Roto, but rather my created player from NBA2K14.. My player, year in and year out scored 30+ ppg, about 8 boards per game, had an AST% that hovered around 45-50%, always a STL % above 5 (unheard of to do in the real NBA) and as a point/2-guard mostly, drained 200+ 3's per season with a few 100+ block years on top of it. Obviously, no surprise, my yearly BPM's were always in the 20's(!!). It got me to thinking, obviously I was collecting data from a video game, but I had molded this player into a very versatile one. My question is what, exactly are the coorelations between BPM, Versatility Index and MVI ? My guess would be pretty high, right ?

[ampersand5]

The other factor to consider (specifically in regards to JV/centres) is that for every season, there are only between 10-15 with positive ORAPM values to begin with.

[Crow]

According to BPM Zach Randolph is currently -11.5 on impact, 2nd worst to Bargnani. Middleton -10.2. D Rose -7.6. PER and WS/48 agree they are bad, but the scales are different and the intensity of the negative impact appears different too.

Derrick Williams though is -6 on BPM, slightly above mean performance on ws/48 and well above average on PER. Kanter is a similar case of divergent boxscore metrics. Will want to see the RPM estimates say before really trying to make any sense of this.



The top 50 on BPM right now average +8.4. The bottom 50, -6.8. By comparison the top 50 on BPM average about 21 on PER and .204 on ws/48. Bottom 50, 10.1 on PER and -0.02 on ws/48.

[Mike G]

After 4 games, Steph Curry has PER = 50.4, WS/48 = .648, BPM = 23.5
He's like 2 MVP candidates in one player.