While I agree that adjusting for team-defense is an aspect of the utmost importance, I do believe there will always be a gigantic flaw when measuring the effect they have on players.
In no way are we immediately able to quantify that during the Laker's first 3-peat, defenses were built and assumed around Shaquille O'Neal, who saw most of the attention and majority of the doubles. They adjusted dependent on approach and performance. These numbers make an adjustment based on the team's defensive performance, but those numbers exist because the defense gave Shaq a hard time than they gave Kobe.
Likewise, these numbers don't show the Detroit Bad Boys being the best defensive team in the league. However, they were most certainly the best team and defending and containing Michael Jordan. They'd created and adapted the most comprehensive basketball scheme against on player in basketball history. 15 defensive schemes to contain Jordan. Surely, that defense has to have a higher weighting on Jordan's performance than the next.
Similarly, while Lenny Wilkins coached some pretty good defensive teams, he was unequivocally the worst coach I've ever seen at scheming against Jordan.
The one team that stands tall, much like the Boston C's, in this regard are the New York Knicks, circa '92. Top defensive team, and probably the best defensive approach I've ever seen against a single player.
Over the course of an 82-game season, it's quite possible that the '08 Celtics were better than the '92/'93 New York Knicks. But they certainly didn't use the amount of men to stop one man.
While not the same squad, I've always loved this clip:
http://www.youtube.com/watch?v=C4kmu4DS6vA
And Neil, I think you'll like this. Amongst YouTube's laymen, there has been a defensive argument going on quite heavily for quite a while now. One Kobe fan posted this video, following the '08 Finals loss.
Kobe Bryant vs Zone Defense:
http://www.youtube.com/watch?v=4kYBeNQdSCc
While another poster made a sarcastic, satirical reply with this
Michael Jordan vs. 1-1 isolation defense
http://www.youtube.com/watch?v=2s9_GKFNL9E
Things like this make stepping away from the stat-sheet quite fun, and actually keep my somewhat sane.
]]>Coincidentally LeBron has just over 25 win shares (regular and post-season) in a single season, slightly besting Jordan's peak. I don't think he has a problem against # 1 defenses, he destroyed the Magic last year and was still respectable against the Celtics this year.
]]>I believe everything you're looking for can be found there, Themo.
]]>On an unrelated matter, I'm assuming there has been much discussion here or at APBRmetrics about the topic in years past, but what is the motivation for the Version 1.1 change to WinShares which removes the sum to team Wins adjustment? Is a team level adjustment viewed as too ad hoc and arbitrary in nature? Should Win Shares be titled something less snappy such as Efficiency Differential Shares?
By the way I love Win Shares, so I'm just trying to understand the motivation and the implications of the refinements made to the methodology.
]]>R^2 are much higher, but I'm regressing against the low-noise 6 year average APM's of Ilardi, so there is no direct comparison.
I'm using only context-neutral stats in the new regression like AST% and ORB%. I'm liking how that' turning out so far.
I took the scoring/turnover term in a nonlinear direction, posing it as ((TS%*2*(1-C_1*TO%))-PPP_Threshold)*USG%*C_2, where C_1, PPP_Threshold, and C_2 are terms found by the regression. This contextualizes the turnovers to some extent. PPP_Threshold is in the .88 vicinity in my initial regression, and the C_1 term is about 1.35 (indicating that a turnover is worth 1.35 points). If I can figure out how to include assists both to and from the player within this same term, I think we're seriously on to a good regression.
I would love to have the additional data for additionally refining the regression.
A cut-down version of my spreadsheet is at: http://spreadsheets.google.com/ccc?key=0Ah1NfCUslJwxdEhreTd4OGN3SzU1b2U5YXNmTG1JeVE&hl=en
]]>But I guess you already sorta came up to that point in the last post..
- J
]]>DSMok - Yeah, I have no idea what's wrong w/ APBRmetrics, but I think the gist of my comment was that I'm interested in helping/collaborating with the new advanced-stat-based SPM it looked like you were working on (I think I missed your intro for that). I can send you the APM dataset I used for the most recent regression if you want (it's the Rosenbaum data + 2 yrs. of Lewin data + recent BBValue data; no Ilardi because Steve combined playoff numbers w/ regular-season ones). What was your R-squared on the first pass? Better than what we were getting w/ the Rosenbaum per-40 method?
]]>