- Mar 31, 2002--- In APBR_analysis@y..., "Michael K. Tamada" <tamada@o...> wrote:
>

games in

> >MacCulloch's

> >curve is really good, but it is based on limited data (not many

> >the last couple yrs).

and

>

> Better than Kobe's and close to Shaq's if I'm reading it correctly.

> Astounding both for how far to the right it goes (offensive rating)

> even more so for how far up it goes (high ratings maintained at high

on

> possessions per minute).

>

> Given that you're using statistical significance to find the points

> the curve, I would've expected the limited data on MacCulloch to

fix this

> problem automatically: wouldn't his small sample size mean lack of

off"

> significance at most cutoff points, and thus his graph should "die

> pretty quickly?

Yeah, it should, but the differences are pretty big, more easily

>

suggesting differences between low and high possession games. What

is in MacCulloch's numbers that is weird is that he doesn't play many

games of a lot of minutes. So he scores a bunch in a few minutes.

As I say, I dunno. Yet. I'll keep an eye on him.

> >(I've also

a level

> >thrown in Kobe Bryant's and Keith Van Horn's with lower statistical

> >significance, signified by the "4".)

>

> The graphs change a fair amount, which makes me leery ... choosing

> of significance usually shouldn't change one's results per se, but

instead

> one's interpretation of the results.

Except that I use the level of significance to create the charts. If

>

we are actually confident that the curve is changing, I then define a

cutoff. If we are not confident that the curve is changing, I use

the lowest previous rating where it was significant. So, because I

use level of significance a little differently here (to say whether

the underlying raw curve is changing), it does show some

sensitivity. It's a weird application of significance testing, I

admit.

> Instead of basing the graphs on at what point one finds statistical

player's

> significance, maybe they should be based on at what point the

> performance (either possesions per minute or offensive rating)

changes by

> some set amount (which could differ at the different offensive

rating

> levels).

Really gets too noisy. I also wanted to be robust to the different

number of points over which I was using the moving average (fewer at

the higher and lower percentiles and for fewer number of games).

Absolute differences are too sensitive to this.

There really could be a lot of econometrics in evaluating

basketball. For anyone interested, especially those in the Seattle

area, the Western Economics Association is having a meeting in

Seattle June 29 - July 3 this year and they do have a couple sessions

on sports economics. Could be very boring or tedious if they get

into details of the math, but could also be very interesting and

educashenal. I've never gone, but will be going this year. Let me

know if any of you want to go and want to meet, more importantly, for

beer after the sessions...

Dean Oliver - << Previous post in topic Next post in topic >>