APBRmetrics

[Neil Paine]

I was fooling around with Dougstats and Dan's old statistical +/- formula today, so I thought I'd share the results. Basically I forced the weighted sum (not the weighted average!) of each team's individual offensive and defensive +/- scores to equal the team's (ORtg - LgRtg) and (DRtg - LgRtg), respectively. At first glance, the statistical +/- numbers also suffer from the "all PGs are defensive liabilities" quirk of pure APM. Nonetheless, here's the data, make of it what you will:

[Mountain]

Thanks.

Not that many surprises with statistical. Here are some marks that caught my eye:

Derrick Rose at -1.5 and Thaddeus Young at -1 are a bit disappointing but not alarming.

Iverson at 0. Nast at just +1. The PG thing, in part due to lower rebounding opportunity / captures, lots of mid-range shots and turnovers. I guess the assists don't balance it out.

Kidd at +5.3 even on statistical, 7th best among guys who play much. Harris +3.6. Rondo 6th best here.

Artest leading the Rockets, Yao right behind. The only major 2 way notably helpful players on the team.

Noah leads on the Bulls.

Mo Williams 85th, still about 75th if you remove the real small minute guys. Probably somewhere near 100th on 2 year adjusted as well. But congrats on the team success and all-Star nod.

[fundamentallysound]

[fundamentallysound]

well, that didn't format the way I wanted it to. whoops.

[Mountain]

Acknowledging that adjusted values are 2 yr and as of today I still checked how they roughly compared to the statistical from last month:

Paul almost +20 on adjusted, just +10 on statistical. Is his non-boxscore contribution really equal to the boxscore contributions of pts, assists, steal, rebounds, etc.? I doubt it.

James, it seems more possible but still the adjusted I'd guess is somewhat higher than true.

Is Wade's non boxscore value twice the boxscore? By adjusted's attempt to fit the whole league that is the answer being given. Seems stretched too high again to me, in an attempt to rank everybody and best explain the league results.

Garnett's non boxscore twice his boxscore contributions? Maybe this season. Iggy's non boxscore three times his boxscore contributions? Maybe but have to wonder a bit.
Nash's non boxscore 9 times his boxscore contributions which includes assists? Odom's 4 times? Ray Allen's double? Nowitski's 5 times? J Johnson's 5 times? RFernandez's 7 times the size?

And on the other side of things Ginobili at almost -4 on nonboxscore? Ben Wallace -3 0r 4? Billups -4 or 5? Maybe.

Hard to say still.

[fundamentallysound]

so, I just realized that I screwed up a bit in using the wrong form of the data. I tried to fudge a bit with the Bulls numbers by using per 36 numbers and converting them to per 40 numbers my multiplying by (10/9), but this seems to have screwed up the results some. so I redid the calculation and came up with this.

[Neil Paine]

[fundamentallysound]

Thanks, Neil!

[erivera7]

Nice work, Neil. Thanks.

Looking at the Magic .. seems like every player checks out fine, statistically, which is good when trying to measure the accuracy of how good/bad a player is. I think the SPM numbers do a great job of showing how balanced Orlando is (clearly there are other ways of showing this trend), as a team. Obviously Dwight Howard is what makes the squad go (and when healthy, Jameer Nelson too) but the contributions of Rashard Lewis and Hedo Turkoglu can't be understated. Ditto with role players like Mickael Pietrus and Marcin Gortat.
_________________
@erivera7
I cover the Orlando Magic - Magic Basketball

[gabefarkas]

[Mountain]

No. given his adjusted of +9 and statistical +/- of just +3 at the time I said that- I meant his non-boxscore contributions were rating as twice as valuable (i.e. in +6 range) and that ratio I think you have to pause and think or perhaps question. The current adjusted has moved up, not sure how far the statistical has changed too but his non-boxscore to statistical (or boxscore) impact may have moved to 2.5-3 / 1. Non-boxscore equal to boxscore impact I can like you believe but when it is several or many times I wonder, though use of "multiples" is a way of dramatizing it and just using the linear +/- scale might make it seem less dramatic and perhaps more believable.

[THWilson]

[Neil Paine]

[Ilardi]

[Mountain]

Effective FG% or TS% allowed is the key topic.

I'd suggest trying some blend 82 games counterpart eFG% and team eFG% while on the court. 50/50 blend or maybe 2/3rds / 1/3rd either way as you see fit. Assists allowed could be used to tweak who gets how much blame for shots that fall for other guys counterparts and shows up in the team eFG% if you want to push it that far.

Taking the data to per minute basis might involve for simplicity sake taking the average shots against for that position per minute or maybe the reported actual rate though if you do that it should be pace adjusted and there other considerations.

TS% might be the way to go. Take it to some counterpart / team blend of points allowed per minute or above / below league average per minute.

Fouls are tricky. The low negative value in Dan's model seemed alright to me for his model. It was research determined but the non-boxscore stuff is fitting here or elsewhere for lack of other spots to manifest itself. It has to manifest somewhere but if eFG% allowed is added and was a lot of what was uncounted then maybe fouls get treated back to or at least somewhat back to the impact of that action basis only?