All Purpose Stats Garbage Dump Thread

[benji]

Key piece? Vince Carter played lots of minutes for the Mavericks but he was just average. Note that his 15.9 PER came with 23.1 usage rate.

But that's the point I'm trying to make and that I contend WP doesn't account for well enough.

Minutes AND possessions are limited. If we take away Carter and Ellis, even if replacing them with more efficient players, those players can't necessarily use as many possessions AND maintain their efficiency.

Just to use a hypothetical, assume I have this nine man team (made up numbers and I'm using the Cats/Hornets because I copy-pasted from the chart above), lets also assume there are exactly 100 possessions to distribute:
If I use this "five" as my lineup and they do something like this:
jefferson,al 30 poss 110 ortg
walker,kemba 25 poss 100 ortg
henderson,gerald 20 poss 105 ortg
kidd-gilchrist,m 10 poss 95 ortg
mcroberts,josh 15 poss 105 ortg

That's 100 poss with 104.3 ortg.

And on my bench I have guys with ortgs of:
sessions,ramon 102
adrien,jeff 110
taylor,jeff 105
biyombo,bismack 115

The "obvious" thing to do is to replace everyone but Henderson (since I don't have another SG) since each at the other positions are better per possession.

Now let's make the five man lineup, ASSUME they keep their efficiency but also assume they keep their possession usage:
biyombo,bismack 5 poss 115 ortg
sessions,ramon 25 poss 102 ortg
henderson,gerald 20 poss 105 ortg
taylor,jeff 15 poss 105 ortg
adrien,jeff 10 poss 110 ortg

I don't have 100 possessions accounted for, only 75. That means there are 25 possessions I need to add to this lineup, and while WP kinda rejects it, pretty much everyone else agrees that increasing usage decreases efficiency to some extent. If we assume for maths ease that for every additional 10% of possessions we lose one point of ortg, just distributing an extra five possessions to everyone we wind up with:
biyombo,bismack 10 poss 105 ortg
sessions,ramon 30 poss 100 ortg
henderson,gerald 25 poss 102 ortg
taylor,jeff 20 poss 102 ortg
adrien,jeff 15 poss 105 ortg

So now I have my 100 possessions but at 102.2 ortg. Even though all four players I moved into the lineup had better efficiency numbers.

Does benji have blog or something? Or even thread where he makes up or at least try to explain his system (or even his Live rating)? WTF is benji system?

And now we do.

[NovU]

First of all, I wholeheartedly love the way you used 100 possessions theory. It is actually quite logical. But only issue is that ORtg's use in context is just not the same as likes or WS, WP, PIE, etc. Your way of prediction is nice but is just another discussion.

That said...

Key piece? Vince Carter played lots of minutes for the Mavericks but he was just average. Note that his 15.9 PER came with 23.1 usage rate.

But that's the point I'm trying to make and that I contend WP doesn't account for well enough.

Minutes AND possessions are limited. If we take away Carter and Ellis, even if replacing them with more efficient players, those players can't necessarily use as many possessions AND maintain their efficiency.

That is obvious. But aren't you unfairly penalizing WP when none of the other metrics also fail to capture that aspect? That is my problem since your criticism also applies to WS, PER, and rest.

Tyson Chandler is still extremely valuable despite the fact that he can hardly score 10 points or grab 10 rebounds. He produces much WinShare at great rate as he maximizes his limited usage and by minimizing making mistakes. But you still can't win with 5 Tyson Chandlers(high eff low usg%). WinShare at the end of the day suffers from a similar setback as WP, just at smaller scale. I'm also WS biased but at the end of the day don't you have to consider it just one of the useful tools. But if most (or all) tools lead to a similar conclusion, then I'd consider it concrete and sound.

PER obviously has higher usage players at advantage.

I don't have 100 possessions accounted for, only 75. That means there are 25 possessions I need to add to this lineup, and while WP kinda rejects it, pretty much everyone else agrees that increasing usage decreases efficiency to some extent.

And WinShare doesn't?

They both are designed to do the same but their values and variables vary. One may overvalue/penalize certain aspects of game more than the other, and one's position based. That I thought was the major difference.

[benji]

WS simply tries to allocate team wins credit to the individual players, it's absolutely not supposed to be a predictive metric, although Justin and Neil found out that it actually was somewhat decent. PER is the same in terms of not being predictive, which is why Hollinger wound up creating EWA.

WP claims to be able to predict because of the consistency of the way it allocates wins. But all systems of this manner do it, which is why you can be somewhat predictive with WS.

I used ORtg because it's an effective replacement for points per possession and the impact of possession usage is easiest to show on the offense end. You can take those ORtg changes and convert them to WS (which is made out of ORtg and DRtg) or do something similar using my simulator. If we assume the defense unchanged at say 103. Then the lineup goes from 45-37 to 39-43. You can then split up those wins based on whatever formula you want.

[NovU]

I do not deny the effectiveness of ORtg. I believe it's quite sound and concrete thus became a conventional analytic stats rightfully. It also has quite the fun relations to rest of the stats such as WS and usage% as well. And this makes analysis easily comprehensible. If CP3 can raise his usage% by 4-5% without sacrifice in efficiency, you are probably looking at LBJ equal in productivity.

One major issue WP has imo though stems from their so called 'scientific approach'. By that, I mean they added positional element to their formula as you also noted in the other thread. This tries to define the right way of playing the position and entirely ignores the system how their offense is operated. It also confines the players to a single position. But more or less the flaws, I think this is a progressive attempt for the world of statistics, and clearly has cons, but also pros in many ways. Hey, it can be argued that WinShare is extremely raw, un-scientific, and outdated to be used as a measurement since you are entirely ignoring different roles and putting everyone in a same basket.

[shadowgrin]

it can be argued that WinShare is extremely raw, un-scientific, and outdated to be used as a measurement since you are entirely ignoring different roles and putting everyone in a same basket.

Does it matter if a C scored 20 pts yet only shot 45 FG% compared to a PG that also scored 20 pts and shot 45 FG%?

Considering the rest of the stats to be equal for both, a basket made is a basket made.

Just because the C shot only 45%, even if bigs are expected to shoot above 50%, doesn't diminish the 20 pts he contributed to the team.

Say the C and PG both scored 20 pts and shot 50 FG%, same again with other stats being equal. Even if the PG shot 50% it doesn't magically increase his 20 pts contribution to the team into 30 pts, it's still 20 pts.

[benji]

Well it does matter comparing within "positions" the problem is properly defining the role you're short-handing for with the positions.

C-PF-SF-SG-PG as an established position is less important than a role composed lineup like Rebounder-Post Scorer-Ball Handler-Scorer-Shooter. Mario Chalmers ability as a ball handler or distributor matters a lot less when I have LeBron and Wade. Similarly for any teams interior defense when they have Joel Anthony.

I want to say it was Ed Kupfier(sp?) who identified effectively eight or nine role types rather than five positions.

[benji]

Not Ed who I think did it first, but a similar study from Sloan:
http://www.sloansportsconference.com/wp ... 012PPT.pdf

[NovU]

Those 8 and 13 ways to differentiate the roles sound quite reasonable though those images look chaotic and I have no idea what the heck they represent.

shadowgrin, don't get me wrong. What you said is why I prefer WS over WP. I just don't think WP entirely bases on faulty premise. It may be weak which I admit.

Position based idea however is not entirely about the players' roles. It's actually weak because it's less about the roles and more about their counterparts. Most players will play accordingly against their position. PF will mostly play against PF most of times throughout the season. PG against PG and SF against SF. WP is created based on the idea that since players play against their counterparts accordingly to the position most of times, they can be analyzed as a group. Of course this creates another sets of issues that we've been discussing.

[benji]

Just a little testy thing:
https://dl.dropboxusercontent.com/u/820 ... eltics.htm
https://dl.dropboxusercontent.com/u/820 ... 5/nets.htm
https://dl.dropboxusercontent.com/u/820 ... knicks.htm

[shadowgrin]

That projection fails for not being a realistic model that would account if Sullinger, Olynyk, or Faverani played as a SG.

[NovU]

The Knicks are nearly 50 wins team?

Do one for the Heat please.

[benji]

https://dl.dropboxusercontent.com/u/820 ... 5/heat.htm

I found that I can save as one file and it auto puts tabs to each team, so I'll just do that from now on.

[NovU]

I don't mean to nitpick but I don't see how Luol Deng is a bare minimum 4.2 wins guy. I gotta say he's a solid 5 wins guy in 70 gp. And he's a two way player.

McRoberts also might be 5 wins guy under heavy load. Wade is about right if taken into count that he'll play more with more on his shoulder. Bosh is probably 9 wins guy with more possessions going through him. Chalmers probably a 5 guy, solid one at that unless there was LBJ effect on him. Also Chris Anderson should be rated little higher even though he was probably the one that was most LBJ boosted.

Perhaps I am being little too optimistic but I think this team is solid 46-50 wins team for a year or two.

[benji]

I don't mean to explain this again but the "wins" aren't some kind of static value, they're a relative value.

Here for example, using nearly the same (can't use the exact same since the "simulator" is an even greater estimate) distribution metric is last year's Heat:

Code: Select all

james,lebron      19.3
bosh,chris      8.3
wade,dwyane      7.6
chalmers,mario      5.4
allen,ray      3.6
cole,norris      2.8
andersen,chris      1.8
battier,shane      1.4
beasley,michael      1.3
lewis,rashard      1.0
haslem,udonis      0.6
douglas,toney      0.4
jones,james      0.2
mason,roger      0.2
oden,greg      0.2
hamilton,justin      0.1

I have a different one that operates somewhat of a bit more team independent:

Code: Select all

james,lebron      +29.26
wade,dwyane      +8.78
bosh,chris      +5.69
chalmers,mario      +3.83
andersen,chris      +3.56
allen,ray      +2.31
beasley,michael      +0.78
jones,james      +0.69
battier,shane      +0.33
lewis,rashard      -0.03
mason,roger      -0.51
cole,norris      -0.80
oden,greg      -1.00
douglas,toney      -1.08
haslem,udonis      -1.31
hamilton,justin      -1.93

[shadowgrin]

lol Heat sucks.

If the Knicks win more than 45 games and the Heat win less than 40 games this season, fuck benji.

[benji]

Trying to figure out rookie comparisons to potentially use for guys who will get major minutes...for example the most common players statistically to Andrew Wiggins seasons are:
NCAA (but current NBA player): Terrance Ross
NBA (post 2002): Josh Howard

Guess I could create some kind of blend or just use one fill-in and then note any major changes in the asterisk, for example, using Howard's all-star season will probably give better results for the Cavs than Ross' rookie season...

If the Knicks win more than 45 games and the Heat win less than 40 games this season, fuck benji.

Why do you always have to put conditions on it?

[shadowgrin]

Your sex therapist aka Jackal said you like playing games like that.

[benji]

https://dl.dropboxusercontent.com/u/820 ... eaders.htm

ew gross

[benji]

Updated that with the C/PF, SF/SG, PG splits, use the tabs at the bottom to switch.

[benji]

https://dl.dropboxusercontent.com/u/820 ... ctions.htm

putting all projections into this now, Celtics updated, good news for Raptors? great news for Sixers

[NovU]

Glossary? They are just meaningless numbers otherwise.

[benji]

They're all explained in the titles?

Except xW which is just a random name I gave it. It somewhat correlates to win and it's an eXtreme calculation, but basically imagine it like WARP or WORP or QORP or if I replaced a Sixers level starter with this guy how many games is this team going to improve? (And it's like half the amount.)

[shadowgrin]

Projection fails again for using the word 'extreme' in the title.

[NovU]

For instance, what the fuck is ball handling metric. Are we supposed to say oh that guy's got a great handle because he has a ball in his hand or somewhere. Or is this a fancy name for some sort of TO per ast/pt related metric?

[benji]

For instance, what the fuck is ball handling metric. Are we supposed to say oh that guy's got a great handle because he has a ball in his hand or somewhere.

Why would you assume this instead of assuming it's some kind of metric that suggests who the best ballhandlers are?