--- In APBR_analysis@y..., "dlirag" <dlirag@h...> wrote:
> Prof. Oliver mentioned in his Jordan vs. Olajuwon article that a
> correlation range of 0.7-0.8 between team wins and linear weight
> ratings he studied suggested that the linear weight systems weren't
> that precise. I agree, but do those figures also mean that those
> formulas would give bad *approximations* of player performance?
Nope. They don't mean that. Those linear weight formulas are quite
good _approximate_ methods. They tell you who is good vs. bad and
give you some distinctions. Arguing over the details is foolish. I
have seen at least 10 different versions of these methods where
people have argued for the weights they have, but none have any real
rationale for what the weights should be. Normally, the author
doesn't know what the formula is supposed to represent. It's not
points scored -- so you can't compare to that. It's not wins or
winning percentage. What exactly is "performance" that these are
trying to approximate???
The methods do give an _approximate_ value of players that relates to
talent. And that approximate value is occasionally used in the NBA
(with the teams' own weight systems) for trades, for drafting, and
for contract negotiation. It's fine. It just doesn't give anyone an
edge in doing so. It also doesn't help predict what will happen to a
team. Will getting a player improve your offense? defense?
passing? Will it make you better than the Lakers? It doesn't answer
any of these questions. But it sure can help your fantasy team.
And, no, I'm not a professor.
Journal of Basketball Studies