APBRmetrics

Mike G wrote:

Statman wrote:Hassan was below average bpm in 2015. In 2016 he had a big uptick in BPM to merely a meh 2.5 (from -0.9). .. He went from well below average to a decent player pretty much because of 1/2 an assist per 48 - as far as I can tell.

Another possibility is that the BPM team adjustment changed significantly from one year to the next.

Without a team adjustment, BPM would have to try harder to make the stat rates work.

The whole approach to BPM was to START with all players getting the same credit (i.e. the team adjustment), and then add/subtract credit based on the individual stats only as warranted by the regression onto APM/RAPM. In other words, having the stats required to sum up to the team's production may artificially change how credit is split up because we are requiring those limited box scores to capture all of the production. Which we know they don't. So BPM only does "what it can" and then splits the rest of the credit up evenly to balance out.

Not balancing is a problem but so are equal team adjustments. Wins produced was highly criticized for its team adjustment.

But which is it, team adjustment first or last? I thought it was last. In the post above it sounds first and last. Technically it has to be one. The BR "about" explanation of BPM says it comes last.

The team results uncut comes first, not the team adjustment.

In other words, having the stats required to sum up to the team's production may artificially change how credit is split up because we are requiring those limited box scores to capture all of the production. Which we know they don't. So BPM only does "what it can" and then splits the rest of the credit up evenly to balance out.

such as defense outside of def rebs, steals, and blocks...

the publicly available sportvu data at stats.nba.com (data now for 4 seasons) not only shows FG% allowed by defenders, but also a FG% difference between what a defender allows and the average FG% of the players they guarded...

so right now draymond green is allowing a 45.8% FG% within 10' of the basket, and that is 10.2% lower than the average FG% of the players he's guarded. rudy gobert is a 46.2% allowed (within 10') and that is 10.6% lower than the average FG% of the players he's guarded...

on the other hand nikola jokic allows a 60.2% FG% (within 10') and that is 2.4% above the average of those he's guarded (all FGA defended and he's at almost +5% above the average of those he's guarded)...

not perfect data as it's extracted from a 2-D representation of the court action, but very telling data nonetheless...

you can also extract from the data FGA/40min allowed per defender, such that some defenders limit opponent FGAs. some defenders also have low numbers of FGAs allowed but only because they rarely guard 1st, 2nd, or 3rd offensive options on opponents. you watch enough games and you can tell for the most part who these players are...

"If a player is shooting 50% on a terrible team, he's a better player than if he's shooting 50% on a great team."

Wouldn't this be more accurate if it referred to a player's on court lineups only, not the entire team on & off? That value is available from several sources. Nothing about a player's shooting percentage is directly affected by the quality of his team or its shooting when he is off court. But that goes into "quality of team" and goes into his shooting adjustment. And is a distortion of the player. Now team and team without player quality are related and may be similar in many cases but they are not the same.

Same for team defense when not on court. Part of team but that part of team is not directly to that player.

Team adjustment for team behavior when ahead or behind? Always have been jittery about that part (not all teams that post similar rates & sizes of leads or deficits are playing to the same level of their potential) but now I can at least say it should be based on lineup behavior with that player on court not team behavior with and without. If starters play way ahead and bench not the same way then starters and bench players shouldn't be treated the same. But they are. The frequency a specific player is ahead or behind by x could be found from play by play database or estimated from raw on / off or crudely estimated from aggregates of mainly starter lineups (3-5 starters) and mainly bench lineups.

"a player's assists are worth more if he is also finishing possessions" Logical supposition but where is the analysis of how much, how much it varies and why ? Doesn't the data on efg% or ts% or offensive efficiency of teammates vary for players with similar assist and efficiency rates? I think it does and that relates to value of player not captured by this generic formula (and to quality of teammates which regretfully is about those players and the quality of their teammates and that one). Conceivably this term could be roughly adjusted by assists at rim or assists on wide open shots. Should Assists on threes by regressed some? Adjustments aren't going to be perfect but there is better and worse and adjustments to adjustments might be better for individual accuracy than simple adjustments same for everybody with relevant differences left unrecognized and accounted for.

Hopefully these comments will encourage adjustments in BPM 2.0.

I'd probably change the rebs / assists dynamic term too but I don't have a specific recommend at this time.

If 4 factors of RAPM or RPM where available I'd test the fit of various dynamic term formulas. How the BPM component values fit with the factors or factor values minus the direct individual parts of the factor to try to get at the interactive element by itself.

Crow wrote:Not balancing is a problem but so are equal team adjustments. Wins produced was highly criticized for its team adjustment.

But which is it, team adjustment first or last? I thought it was last. In the post above it sounds first and last. Technically it has to be one. The BR "about" explanation of BPM says it comes last.

The team results uncut comes first, not the team adjustment.

Effectively, the team adjustment comes last, but when I developed BPM, in the regression, the iterative regression fit was determined after the team adjustment on each fitting step.

The big issue with WP was not the team adjustment, it was first valuing everything directly based on possessions at the team level (i.e. a defensive rebound = 1 possession) and then taking those poor values and assuming that stats at a team level could be directly translated to value at a player level.

Ok, WP had more than one major issue. But team adjustments are still signs of weakness, work finished generically, imprecisely.

So no response on the other points or not yet?

Crow wrote:Ok, WP had more than one major issue. But team adjustments are still signs of weakness, work finished generically, imprecisely.

So no response on the other points or not yet?

I strongly disagree with that first statement. If the box score was truly all-encompassing, then maybe it could be supported. But it's not. Trying to make the box score account for everything (which is what happens without a team adjustment) will cause problems.

On the other points, I welcome all input as I look at revising BPM! Any comments are good comments. Please recall, though, I am intentionally limiting BPM to statistics that are available back into the 1970s. Only the basic stats.

DSMok1 wrote:.. I welcome all input as I look at revising BPM! Any comments are good comments. Please recall, though, I am intentionally limiting BPM to statistics that are available back into the 1970s. Only the basic stats.

Do we consider ORtg and DRtg to be 'basic' stats? They're in the boxscores at b-r.com for single games, seasons (or any span of games), playoff series.
They're derived from player and team rates, so they act as a sort of built-in team adjustment. And they're shown from 1977-78 onward (DRtg from 1973-74)
Somehow Offensive Win Shares are given from the beginning of time.

A player's ORtg/DRtg gives some information; and factors like (TmORtg/OppDRtg) and (OppORtg/LeagueORtg) all cancel out, so there's no need to complicate it beyond O/D.
Usage is also derived from the most basic stats, and it can combine with O/D to give a basis for player proficiency.

Ast% in combination with Reb%, and/or with (1-Usg) further refines the number. Min/G and GS/G are positively related to RAPM but independent of O/R, so add yet more info. And so on.

Mike G wrote:

DSMok1 wrote:.. I welcome all input as I look at revising BPM! Any comments are good comments. Please recall, though, I am intentionally limiting BPM to statistics that are available back into the 1970s. Only the basic stats.

Do we consider ORtg and DRtg to be 'basic' stats? They're in the boxscores at b-r.com for single games, seasons (or any span of games), playoff series.
They're derived from player and team rates, so they act as a sort of built-in team adjustment. And they're shown from 1977-78 onward (DRtg from 1973-74)
Somehow Offensive Win Shares are given from the beginning of time.

A player's ORtg/DRtg gives some information; and factors like (TmORtg/OppDRtg) and (OppORtg/LeagueORtg) all cancel out, so there's no need to complicate it beyond O/D.
Usage is also derived from the most basic stats, and it can combine with O/D to give a basis for player proficiency.

Ast% in combination with Reb%, and/or with (1-Usg) further refines the number. Min/G and GS/G are positively related to RAPM but independent of O/R, so add yet more info. And so on.

ORtg and DRtg are just all inclusive stats like BPM. They are the basic building blocks of the modern version of Win Shares. http://www.basketball-reference.com/about/ratings.html http://www.basketball-reference.com/about/ws.html

I'm going to start with the basic stats, like AST% or USG%. I will also use minutes, mpg, things like that.

Win Shares does not strive to match players' APM or RAPM or RPM. ORtg only gets at 'individual efficiency', which does not mean much in a team concept. ORtg/DRtg is inclusive of several more 'basic' stats, but also includes team and opponent adjustments.

You could just use those components of ORtg and DRtg, or you could shortcut the process; adjust it further with some form of Usg%, Ast%, etc.
And very little additional team adjust is necessary when O/D is a component. Perhaps if Ast% is corrected by a home/away factor, it's mostly extraneous.
We can do this independently and see which better fits with RPM or RAPM.