--- In APBR_analysis@y..., "dlirag" <dlirag@h...> wrote:
> My observations/questions-
> 3. Assuming his statistical tests are sound, he seems to
> that his linear model is better than the Cobb-Douglas framework. In
> page 3, he seems to associate the latter with the idea that the
> of a given stat depends on a team's other statistics. Did he intend
> to debunk this idea or just the Cobb-Douglas model that was
> associated with it? It seems to me that Berri's linear concept
> doesn't seem compatible with the concept of player performances
> interdependent with team performance. For example, wouldn't
> misses have more negative value if he were on a team with lousy
> offensive rebounding?
Berri has said that he does not think that the value of stats do not
depend on a team's other stats. And, yes, he does associate that
concept with Cobb-Douglas.
> 4. I noticed in the team tempo equation that FG attempts are
> more heavily than FT attempts by a ratio close to that found in Dr.
> Oliver's team possession formula. Is this notable?
Hadn't noticed that. I did pass along the possession formula to
Berri and he redeveloped his numbers. He said that it made little
difference in his results. Since possessions are a measurable thing
(not required to be estimated with the formula), I asked him whether
he would consider simply using the possessions total and he didn't
want to do that because it hurt his individual win calcs.
> 5. Berri's solution to the apparent statistical bias towards big
> players doesn't seem to fit well with the rest of his theory. It's
> hard for me to explain my unease since I don't have rigorous
> in statistics, but I feel that his framework seems to render such
> adjustments unnecessary. For all I know, his model could really be
> saying that big players are a lot more valuable than smaller
> (He writes about big players getting more rebounds and the smaller
> guards getting more turnovers in his section on positional
> adjustments). If so, maybe this suggests a deficiency in the
> statistics we keep, a flaw in his model, or that big players really
> are much better.
I think there is a little truth in all of these. There is a
deficiency in the stats we keep. Big men get credited for blocks,
but there is not measure associating Joe Dumars with his ability to
keep good shooters from getting the ball in good shooting position.
There are flaws in his model that are accentuated by this deficiency
in stats. Big players are, by my reckoning, more important to a
defense than small players. Big players do seem to have an advantage
overall in that height helps get you to the NBA. The average height
of NBA players is higher than in college than is higher than in HS.
The average height of NBA players is higher than those in the NBDL.
> 6. I don't know if this is one of the formula's intended functions,
> but what does the formula say about the performance of the San
> Antonio Spurs in the season when David Robinson had that long
> due to injury?
A lot of things suggest Robinson's value to the Spurs prior to
> 7. Do the parentheses around some numbers mean that they're
> I'm having trouble with the figures in Tables 6-8.
Yes, parens mean negative. Position averages for everyone but
Centers, I think, were negative.
> 8. The paper deals with "wins producers." What might a study
> producers," given this framework, yield?
Berri is thinking about this. He doesn't have a good answer for how
to estimate losses.
I sent him my list of individual win-loss records for his top win
producers. Here are the approx #s
These are approximate. I originally calculated them with an old
formula, which makes a difference for the big offensive rebound guys
only (Rodman, Williams, Outlaw) and these are my recollections of how
the numbers I have written down (from old formula) actually change
with the new formula.
Fundamentally, the point I made to Berri is that it's very unclear
how many "games" a player is responsible for. My sense is that
players like Rodman, Williams, Outlaw aren't responsible for a lot
of "games". This is because we don't typically see big drop offs in
team performance when these guys miss games. As an example, I had
Doug Steele calculate for me the Bulls record with and without Rodman
over the 3 years he was there. They were 40-7 (85%) without him and
163-36 (82%) with him. This suggests that the difference between him
and his replacements (Kukoc, Simpkins, Salley, Wennington,
B.Williams, Buechler, and Caffey) wasn't very big and/or that
his "game" impact was small.
Journal of Basketball STudies