APBRmetrics

Aiming for implementation before this season starts or hard to say yet?

Looking for any more input on draft changes?

Italian Stallion wrote: ↑Wed Aug 14, 2019 3:49 pm ...
I am saying people are so worried about the accounting looking right, they are getting the reality of the values wrong.

To be clear, I don't have some kind of fool proof way to account for these things and get the values right. I simply think a LOT of things of value (both positive and negative) are going on in every possession, many of which are not being accounted for, and some that are being given too much value to make the accounting look neat. IMO, the OREB/Putback Score combination is overvalued.

I agree with the sentiment. The thing is that, in some sense "the accounting" is all we have. Without it, there's not much to go on. (Heck, even keeping track of which team scored the most points in a game is a form of accounting.)

Suppose that, instead of worrying about whether the value is exactly right, we're only trying to work out which of two values (a or b) is better or more correct. Do you have some sense of how you would do that?

It's hard to have constructive conversations when people are just saying "I think the value is too high" or "I think the value is too low."

Crow wrote: ↑Thu Aug 29, 2019 7:03 pm Aiming for implementation before this season starts or hard to say yet?

Looking for any more input on draft changes?

I'm aiming for it. We shall see.

Coefficients of GmBPM:

Compared to BPM it appears these weights for GmBPM massively increase the value of defensive rebounds. Offensive rebounds diminished to same status as defensive rebounds or down to only half of defensive rebounds? If equal weighting of offensive and defensive rebounds, that would be like a certain other older, much criticized for this, metric. Or... is it defensive rebounds now twice as valuable? How is that finding possible / rational? I see seemingly conflicting values for types of rebounds in the tabs and need a clarification.

I can't tell from surface due to BPM complexity on this stat whether assists got increased in total value, diminished or stayed roughly the same.

PFouls re-enter as a negative after being absent in BPM. And more b=negative than in other metrics? What is the argument for doing this?

Blocks go up a bit?

Ready to explain position breakpoint and slope yet? How do these terms impact player scores by position on average and in extreme cases compared to all the interactive terms in BPM?

If in a single possession there is a 2pt fga, a make and an assist , the total value of those are less than 2 points. Does the rest of the value get split up among everybody? Equally or by variably by position?

TOs less negative than in other metrics? To the individual... so more blame to other players?

Crow wrote: ↑Tue Oct 01, 2019 4:31 pm Compared to BPM it appears these weights for GmBPM massively increase the value of defensive rebounds. Offensive rebounds diminished to same status as defensive rebounds or down to only half of defensive rebounds? If equal weighting of offensive and defensive rebounds, that would be like a certain other older, much criticized for this, metric. Or... is it defensive rebounds now twice as valuable? How is that finding possible / rational? I see seemingly conflicting values for types of rebounds in the tabs and need a clarification.

I can't tell from surface due to BPM complexity on this stat whether assists got increased in total value, diminished or stayed roughly the same.

PFouls re-enter as a negative after being absent in BPM. And more negative than in other metrics? What is the argument for doing this?

Blocks go up a bit?

Ready to explain position breakpoint and slope yet? How do these terms impact player scores by position on average and in extreme cases compared to all the interactive terms in BPM?

If in a single possession there is a 2pt fga, a make and an assist , the total value of those are less than 2 points. Does the rest of the value get split up among everybody? Equally or by variably by position?

TOs less negative than in other metrics? To the individual... so more blame to other players?

BPM had many of these statistics in multiple terms, so comparing is not at all straightforward. The value of rebounding as a whole is somewhat decreased vs. BPM, and it no longer has an interaction with AST%. Rebounds are only mildly positive at all. Most of the value of a rebound is actually divided up among the players on the floor (if we do this analysis at the play level).

The biggest winners vs. BPM are players with many good points but lacking rebounding or assists (i.e. the interaction terms). Players like Steve Nash and John Stockton. Biggest losers tend to be players with a lot of minutes but few stats, because I'm not using an MPG term at this point. Players like Tyson Chandler and Kendrick Perkins.

Oh yeah... the biggest loser was MVP Russell Westbrook.

Crow wrote: ↑Tue Oct 01, 2019 4:31 pm If in a single possession there is a 2pt fga, a make and an assist , the total value of those are less than 2 points. Does the rest of the value get split up among everybody? Equally or by variably by position?

I don't think trying to make a metric accurate down to a possession level is an achievable goal. If the coefficients (roughly) add up over the course of a full game that's a huge success in my book.

eminence wrote: ↑Tue Oct 01, 2019 6:21 pm

Crow wrote: ↑Tue Oct 01, 2019 4:31 pm If in a single possession there is a 2pt fga, a make and an assist , the total value of those are less than 2 points. Does the rest of the value get split up among everybody? Equally or by variably by position?

I don't think trying to make a metric accurate down to a possession level is an achievable goal. If the coefficients (roughly) add up over the course of a full game that's a huge success in my book.

With a truly linear metric, it should generally make sense even down to possession level, with a few exceptions. Steals, for instance, may in some cases be worth more than a possession to the player, since they indicate defensive activity on other possessions as well.

In most cases, though, there is surplus credit to divide between the team on the floor (if you're looking at it that way).

DSMok1 wrote: ↑Tue Oct 01, 2019 6:53 pm

eminence wrote: ↑Tue Oct 01, 2019 6:21 pm

Crow wrote: ↑Tue Oct 01, 2019 4:31 pm If in a single possession there is a 2pt fga, a make and an assist , the total value of those are less than 2 points. Does the rest of the value get split up among everybody? Equally or by variably by position?

I don't think trying to make a metric accurate down to a possession level is an achievable goal. If the coefficients (roughly) add up over the course of a full game that's a huge success in my book.

With a truly linear metric, it should generally make sense even down to possession level, with a few exceptions. Steals, for instance, may in some cases be worth more than a possession to the player, since they indicate defensive activity on other possessions as well.

In most cases, though, there is surplus credit to divide between the team on the floor (if you're looking at it that way).

I agree it should make sense in that they all 'point' in the intuitive directions (and your values seem to), I just meant 'accurate' in a sense that your stat adding up nearly exactly to +2 for the offense (or +1.5 O and -0.5 D or whatever) on a possession where a team makes a layup isn't a realistic goal.

The only spot that strikes me as just a tad odd on 1st glance is FTA and FT being separate with FTA being negative, maybe I'm just blanking on scenarios where that makes sense, but it would seem to me you could get very very nearly the same results without including FTA.

"The value of rebounding as a whole is somewhat decreased vs. BPM"

Ok.


But for the part that is allocated out to individuals is it equal for offense and defensive rebounds or 2-1 more for defensive rebounds? Or both or neither?

And when total rebounding value is divided, what is the comparison between OR and DR? Are offensive rebounds considered more of a team event for some reason?


I am still interested in about 5 other questions raised; but you'll either answer them, now or later or not.

I see this:

https://twitter.com/Tim_NBA/status/1179178236812627968

Any comments on agreement / disagreements with your weights / findings?

Crow wrote: ↑Wed Oct 02, 2019 1:12 am I see this:



Any comments on agreement / disagreements with your weights / findings?

No disagreement in particular. Offensive rebounding at a player level is so much dependent on what is asked of that player, not on the actual player ability.

Crow wrote: ↑Tue Oct 01, 2019 4:31 pm Compared to BPM it appears these weights for GmBPM massively increase the value of defensive rebounds. Offensive rebounds diminished to same status as defensive rebounds or down to only half of defensive rebounds? If equal weighting of offensive and defensive rebounds, that would be like a certain other older, much criticized for this, metric. Or... is it defensive rebounds now twice as valuable? How is that finding possible / rational? I see seemingly conflicting values for types of rebounds in the tabs and need a clarification.

Rebounds are funny. Offensive rebounding is very dependent on team context and philosophy--at a team level they are quite beneficial, but not so much at the individual level. I found no benefit to splitting offensive and defensive rebounds up. As a whole, rebounds are less important than in the original BPM.

Crow wrote: ↑Tue Oct 01, 2019 4:31 pmI can't tell from surface due to BPM complexity on this stat whether assists got increased in total value, diminished or stayed roughly the same.

Assists are similar in value, but the removal of the rebound interaction term helps certain players get more credit for their assists (Stockton, Nash).

Crow wrote: ↑Tue Oct 01, 2019 4:31 pmPFouls re-enter as a negative after being absent in BPM. And more b=negative than in other metrics? What is the argument for doing this?

It fit well...

Crow wrote: ↑Tue Oct 01, 2019 4:31 pmBlocks go up a bit?

Again, this is how it fit the best.

Crow wrote: ↑Tue Oct 01, 2019 4:31 pmReady to explain position breakpoint and slope yet? How do these terms impact player scores by position on average and in extreme cases compared to all the interactive terms in BPM?

Position adjustment is mostly just a penalty on PGs. Each player is assigned a position from 1 to 5 (on a continuous scale) and then the position adjustment is calculated. Above the "breakpoint" there is no adjustment. The slope runs down from that break point back towards 1. So a PG gets a penalty of over 2 points. SG a small penalty. Everyone else no penalty.

Basically, it's saying--if you're a small player, you better show impact in the stats, because your non-stat impact (i.e. defense, picks, etc) is not as good.

Crow wrote: ↑Tue Oct 01, 2019 4:31 pmIf in a single possession there is a 2pt fga, a make and an assist , the total value of those are less than 2 points. Does the rest of the value get split up among everybody? Equally or by variably by position?

The remaining credit does get divided up, equally.

Crow wrote: ↑Tue Oct 01, 2019 4:31 pmTOs less negative than in other metrics? To the individual... so more blame to other players?

Yes, that is a reasonable interpretation.

Thanks for the replies.

DSMok1 wrote: ↑Tue Oct 01, 2019 11:54 am Coefficients of GmBPM:

I'm interested in input: should any of these stats be in any way context dependent? I.E. relative to league or team averages?

Thanks!