Re: [APBR_analysis] Re: And while I'm at it...
----- Original Message -----
Sent: Wednesday, February 05, 2003 10:29 AM
Subject: [APBR_analysis] Re: And while I'm at it...
> Ed --
> Take a look back at Message 14. That message discusses some of the
> other methods for weighting stats. JohnH of this group wrote a book
> with different weights. There is an economist named Dave Berri who
> also has something effectively similar. There are a few more on the
> web, too. I've developed a matrix showing the relative weights that
> the developers put on the different stats. I can't put that matrix
> out here since it's in the book I've got coming out.
Hmm, mostly 0.5 - 1.0, which is what I guessed.
> Yours is slightly different, of course, because you don't include any
> defensive stats, making it purely an offensive tool. That makes it
> somewhat unique.
I focus only on offense (and only part of the offense, for that matter) because I'm
incapable of developing an all-encompassing system. I prefer to let you guys slug out a
more comprehensive model -- I've seen how difficult it is. I'll focus on the small
I think what makes it different than the others is it's theoretical basis. Most of the
methods you describe are linear weights-type systems, the
add-the-good-subtract-the-bad-to-arrive-at-a-rating models. Mine is a little different.
(First of all, I wanted it to look like a player's points total for a game, so you
would have an easy way of grasping what the rating means.) I assume those other ratings
used some kind of regression to arrive at weights that would produce results that jibe
with a qualitative assessment of player ability. I don't do that (I share your distaste
with linear weights). I only add and subtract stuff with regards to a teams expected
points per possession: Teams can expect 1 point per possession. If a player misses a
shot, he cost the team 1 point. If he turns the ball over, he cost the team 1 point.
Like I said, it's a pretty shaky basis on which to construct a rating system -- but it
seems to produce half decent results. I'd like it to be more accurate, but I can't seem
to do that without changing weights. I'm thinking that there may be a conceptual
problem with my approach that I'm overlooking. (Here's one: since missed shots usually
result in an offensive board 30% of the time, they can't possibly be worth -1 with
regards to a possession. But why does -1 give good results? I don't know.)
- --- In APBR_analysis@yahoogroups.com, <igorkupfer@r...> wrote:
> Like I said, it's a pretty shaky basis on which to construct arating system -- but it
> seems to produce half decent results. I'd like it to be moreaccurate, but I can't seem
> to do that without changing weights. I'm thinking that there may bea conceptual
> problem with my approach that I'm overlooking. (Here's one: sincemissed shots usually
> result in an offensive board 30% of the time, they can't possiblybe worth -1 with
> regards to a possession. But why does -1 give good results? I don'tknow.)
Would -0.7 per missed FGA give better results?
- --- In APBR_analysis@yahoogroups.com, "dlirag <dlirag@h...>" > a
> > problem with my approach that I'm overlooking. (Here's one: sincedon't
> missed shots usually
> > result in an offensive board 30% of the time, they can't possibly
> be worth -1 with
> > regards to a possession. But why does -1 give good results? I
> know.)There's not a lot of solid guidance on what the weights should be.
> Would -0.7 per missed FGA give better results?
If all you want are better results -- "better" is a personal
decision, so you can try it and see whether the results
are "better". Whether you also account for some teams rebounding
only 20% of their misses and other teams rebounding close to 40% is a
question. And if you modify the weight on missed FGA, should you
also modify the weight on offensive rebounds?