APBRmetrics

Thinking out loud here...

I could create two dummy variables for each player on the floor, one for the player as a "passer" and the other as a "shooter". When the player shoots the ball, he gets a 1 for the latter and a zero for the former. His teammates are assigned exactly the opposite (1 for passer, and 0 for shooter).

(There would be no need to split this out on the defensive side.)

I think, if this works, we'll see that, for example, Korver might have a high "shooter" rating, but his "passer" rating will be closer to average.

Thoughts?

Actually, thinking some more...I guess the "shooter" will always perfectly correlate with his own assisted shots. But this might still subtract him out from the passer part of the equation.

Yeah. I don't think you can tease it out that way. What I was thinking of was something more along the lines of this:

Rose on: 60% of team baskets assisted. 25% of Rose's baskets are assisted. I don't know how many FGs when he is on the court are his, but assume 30%. Getting that 30% number is the only tough number to get from 82games btw. This would mean that 75% of his teammate's baskets are assisted when he's on the court.

Rose off: 62% of team baskets are assisted.

Raw +/-: Rose increases teammate assists by +13%.

Basically, he's doing something before the pass that leads to a basket that is in part responsible for the assist, because his assist totals alone don't generate this type of impact.

Do that for enough players and you should be able to pick up who the real facilitators are. Digging deeper and looking at lineups would be even better.

The problem with assigning dummy variables is that while it would remove Korver from an "assistor" for his own shots, it's still not telling you who is really contributing among his 4 teammates. Is it the assistor, the hockey assistor, the guy on the opposite corner three side picking his nose while acting as a spacer?

Chicago76 wrote:...
Value of assist at rim from high post should result in higher efg%. ..
Assists in the NBA are overwhelmingly the result of hitting the open man after drawing a double team. Drawing a double occurs in one of two ways (generally): dribble penetration or doubles on bigs. ..
Why I purposely chose a 5-6 ast/36 min "caretaker" PG like Kevin Ollie is that the bulk of his assisting is done on simple rotation. He's not doing a lot of dribble penetration ...

There must be something Ollie contributed, beyond what a guy off the street could do, to average 15 minutes over 13 years. As you note, he didn't do much else that showed up in the boxscores, other than his 5.2 Ast/36.

The occasional center who may avg 5 Ast/36 probably does a lot less of bringing the ball up the floor, for one thing. This is an essential skill to have on the floor at all times, and there's no statistical reward for it. There's the occasional TO.

Walton before the wheels completely fell off during the 2 yr layoff didn't have huge turnover rates compared to lg avg (+1 or +2% above). If you want an example of someone who did, look no further than one of the best passers of all time: Magic Johnson...

Magic averaged 19 pts and 11 ast per 36. Such productions can be expected to come with about 3.7 TO, and he got 3.85.
Walton's averages of 16 and 4 would suggest about 3 TO/36, but he averaged 3.6 -- not including his first 3 seasons.
So Magic got about .15 excess TO per 36, and Bill got about 4 times that much, .60 .

Just as TO cancel some of the effectiveness of Ast totals, missed FG do not produce Ast; hitting shots create both Points and Ast, proportionate to the FG%*.

* Now we could argue that an Ast on a 3pt FG is worth more than one for a 2pt FG.

Has anyone done a multiple regression on lineup vs lineup rebounding data to see if having more guys than your opponent on the court 6-11+ or 6-9+ tall or 2+ inches over position average or if a cumulative height or weight advantage of the whole lineup or just the front line or other variables are statistically significant?

Mike G wrote:There must be something Ollie contributed, beyond what a guy off the street could do, to average 15 minutes over 13 years. As you note, he didn't do much else that showed up in the boxscores, other than his 5.2 Ast/36.

My theory on Ollie is that he managed to stay in the league for so long not because he did anything well (or even average), other than as a defender. He just didn't do anything awfully. These guys have a tongue in cheek profile on him that I think really describes this better than I can:

http://www.shamsports.com/content/pages ... jsp?id=308

Mike G wrote:The occasional center who may avg 5 Ast/36 probably does a lot less of bringing the ball up the floor, for one thing. This is an essential skill to have on the floor at all times, and there's no statistical reward for it. There's the occasional TO.

But bringing the ball up the floor in and of itself has nothing to do w/ assisting. I suppose if you really want to connect the dots, you could argue that bringing the ball up the court successfully keeps a possession alive, which in turn could lead to an assistable opportunity. That specific action is almost a non-event in 95% of all scenarios though. Kind of like if you and I decided to pass the ball back and forth 40 feet from the basket. Seems like a huge stretch to bring this up in the context of discussing the creation of assistable FGs or FGAs

If you were to credit Ollie with that activity (which gets successfully converted 99% of the time by all NBA PGs), then you probably need to credit bigs with higher ORBs, which are much more variable than the conversion rate of bringing the ball up the court.

Mike G wrote:Magic averaged 19 pts and 11 ast per 36. Such productions can be expected to come with about 3.7 TO, and he got 3.85.
Walton's averages of 16 and 4 would suggest about 3 TO/36, but he averaged 3.6 -- not including his first 3 seasons.
So Magic got about .15 excess TO per 36, and Bill got about 4 times that much, .60 .

Seems like a dubious claim to me.
-First, I was talking about pre-injury Walton, not 16+ and 4+ Walton over his entire career.
-Secondly, where is this "estimate" coming from? Are you regressing something? Controlling for era? Including reb in your regression, which I assume would also have a positive correlation to TOVs? Controlling for lack of 3 pt line? What is the standard error, because the difference looks pretty small.
-Finally, comparing excess TOVs is about the only way you can make these two look significantly different. It would be like if the avg temp on a given day was 70. Monday it was 71. Tuesday it was 74. And then you tell me that Tuesday's excess over normal was 4 times greater than Monday's. It sounds like a lot that way, but in reality, you aren't changing how you dress from Monday to Tuesday.

Regardless, this is way off tangent from the original point, which is that marginal production is really important, because marginal production (at a team level) is what wins and loses games.

5 ast/36 is pretty baseline while 1-2 ast/36 may be for a big. As you increase above these #s for their respective positions, you are getting into positive territory that is likely to produce more assisted baskets at a team level--and team level is all that really matters. Similarly, I would never argue that a C getting 5 defensive rebounds/36 is worth as much as a PG who is staying home on his opposing PG defensively producing the same 5 drb/36. The C is likely reducing his team's DRBs at a team level, while the PG is improving the DRB situation for his team.

Chicago76 wrote:... I would never argue that a C getting 5 defensive rebounds/36 is worth as much as a PG who is staying home on his opposing PG defensively producing the same 5 drb/36. The C is likely reducing his team's DRBs at a team level, while the PG is improving the DRB situation for his team.

In 2001-02, the Nets started Todd MacCulloch at center, and he averaged 5.3 DReb/36.
Backups Aaron Williams and Jason Collins got 5.2 and 4.3, respectively.
Jason Kidd, at 5.5, was 2nd on the team to Keith Van Horn's 6.9 .

We understand that Kidd > Todd, but I'm not sure it's possible to argue that any C was "reducing his team's DRBs" in any practical sense.
Their starting SF got 73% more DReb than their PF (KMart), and the PG got more than the C.
They were just above avg in both DRb% and ORb%, and their DRtg led the league.

Including reb in your regression, which I assume would also have a positive correlation to TOVs? ... What is the standard error, because the difference looks pretty small.

Yes, Reb are also correlated to TO, and that's a large part of the reason the difference is pretty small.

xkonk wrote: An interesting claim

Not a claim, it is a fact that Kidd led the league in WP in 2007.

xkonk wrote: , since you list that his WP48 went down a tiny bit if anything (as did his WS48) and he played pretty much the same number of minutes those years. So it seems like Kidd's standing in the rest of the league might be due more to what other people did (which apparently must have been get worse) as opposed to what Kidd did.

Well, that is trivial. It should be rather obvious that the rank of a player depends on the performance of the other players. Kidd did not led the league in WP48 in 2007. David Lee had the highest WP48 in 2007 with 0.344 (as good as LeBron James in 2009).
His PER, btw., went even in up from 2006 to 2007, so, it is not just an issue with WP48.

Chicago76 wrote: I see your point better now. Thanks for that. I think the fact that WP48 goes up while RAPM and team D goes down is a standard hazard of box score metrics. PER would have the same issue. And this isn't a defense of WP48, because as far as box score metrics goes it leaves a lot to be desired. That said, there is some place for adj +/- and a box score metric like PER. One is more prone to errors of misattribution (box score) while the other is prone to greater standard error (adj +/-).

I agree. That's why I think a mesh between boxscore based metrics and +/- based metrics is a better way to go.

Chicago76 wrote:If this is your criticism (and I think it's a valid one), and you feel that this corrupts the metric, then don't look at any box score metric, because this isn't unique to WP48.

Well, my metric actually captured it. So, it is not an issue which is in fact related to the issues of a boxscore, but rather of the not "accurate" measured coefficients. If the weights are picked better, we should see a better predictive performance. And with my SPM or Dsmok's ASPM we see a better predictive power than for WP48 or PER for example.

Chicago76 wrote: Personally, I think we need to look at some form of a box score metric just to balance out the weaknesses of adj +/-, so we might as well refine that metric as best we can.

We seem to agree on that.

... it is a fact that Kidd led the league in WP in 2007.

Careful how you use the word "fact".
Even if WP was around in 2007 -- and this very thread starts out regarding some change since that time in WP -- what if one or more of the obviously bigger problems with WP should be corrected? What if the 'position adjustment' is revised or discarded, and Kidd then is not the 2007 leader in WP?

It's an unstable metric.

Mike G wrote:

... it is a fact that Kidd led the league in WP in 2007.

Careful how you use the word "fact".

Not quite sure what I should be "careful" about in that case. With the current version Kidd led the league in WP in 2007. That is a fact.

http://wagesofwins.com/wins-produced/wi ... uced-2007/

There is no discussion about that at all.

Mike G wrote: Even if WP was around in 2007 -- and this very thread starts out regarding some change since that time in WP -- what if one or more of the obviously bigger problems with WP should be corrected? What if the 'position adjustment' is revised or discarded, and Kidd then is not the 2007 leader in WP?

It's an unstable metric.

Sometimes I wish people would just talk about something they really have knowledge about and not just pull stuff out of thin air. Mike, Berri has a weblog, on that page (which was linked already multiple times) you can easily see how WP is calculated and you can check WP numbers since 2001.

And yes, Berri can make more adjustments to his metric, it just adds further proof that Berri is not trusting his original hypothesis. Which just confirms what most smart people said in the first place.

mystic wrote: And yes, Berri can make more adjustments to his metric, it just adds further proof that Berri is not trusting his original hypothesis. Which just confirms what most smart people said in the first place.

Really? Raise your hand if you created a metric of some kind and then never changed it because it was perfect the first time. On the list of reasons to not like Berri/WP, 'changed it in an attempt to make it better' seems like it should be awfully low. Offhand, the only metric I'm familiar with that I don't know has changed over time is PER, and that's probably just because the changes happened before I was paying attention.

Xkonk,

Most of us concede this when creating a stat.

Berri, on the other hand, just sort of got an attitude of "they don't know what they're talking about" in WOW, and took the same stance upon rebuttals that further clarified the information he brushed aside as irrelevant. Lots of antagonism, honestly, with no supporting evidence, until FINALLY changing his results.

Kevin Love was an obvious error.

xkonk wrote: Really? Raise your hand if you created a metric of some kind and then never changed it because it was perfect the first time. On the list of reasons to not like Berri/WP, 'changed it in an attempt to make it better' seems like it should be awfully low. Offhand, the only metric I'm familiar with that I don't know has changed over time is PER, and that's probably just because the changes happened before I was paying attention.

What has that to do with anything I wrote? Do you really not understand that Berri's original hypothesis is that boxscore entries have intrinsic values and those values can be found via regression? If the result of that regression does not lead to anything useful and heavy adjustments or changes have to be made (he is making adjustments, btw, he is still using the same PA, PE equations and the same regression results at the start), he is going further away from that original hypothesis. That is all I wrote about that. Improving the metric by making changes is nothing bad at all, it is just not inline with science to make adjustments after the marginal values are calculated and still claim that the original hypothesis would be correct.

mystic wrote:

Mike G wrote:

... it is a fact that Kidd led the league in WP in 2007.

Careful how you use the word "fact".

Not quite sure what I should be "careful" about in that case. With the current version Kidd led the league in WP in 2007. That is a fact.
http://wagesofwins.com/wins-produced/wi ... uced-2007/
There is no discussion about that at all.

So, was Kidd "in fact" the 2007 WP leader in 2007, as well as now?
Will he be in fact the 2007 WP leader in the 2015 version?

I'm not particularly interested in how WP is currently calculated, any more than I want to figure out how to start an engine that's never started before, on a car with square wheels. There's no possible reward.

Mike G wrote: Will he be in fact the 2007 WP leader in the 2015 version?

How should I know? But that doesn't change my point at all, because I was talking about the most recent version of WP anyway.

Mike G wrote: I'm not particularly interested in how WP is currently calculated, any more than I want to figure out how to start an engine that's never started before, on a car with square wheels. There's no possible reward.

That sounds pretty ignorant to me. I read Berri's papers and his first book. And everyone can also learn from the mistakes made by others.