Re: So did "luck" prevail in the Monty vs. M. Petch match?

On May 11, 3:06=A0pm, Michael Petch <mpe...@[EMAIL PROTECTED]
> wrote:
> On 5/11/08 12:41 PM, in article 65HVj.261147$pM4.187166@[EMAIL PROTECTED]
 "Kees
v=
an
>
> den Doel" <u...@[EMAIL PROTECTED]
> wrote:
> > Here's a =A0rating analysis of the =A0series, using the =A0same gnubg
se=
ttings
> > used to =A0calibrate the rating =A0estimate formulae built in
=A0gnubg. =
Listed
> > are =A0the =A0 estimated =A0rating, =A0the =A05% =A0confidence =A0
inter=
val, =A0and =A0the
> > contributions from =A0checker play and =A0cube play towards the
=A0lost =
rating
> > points (2000 being calibrated as "perfect play").
>
> > mpetch =A0 =A0 =A0 =A0 =A0(Average of 7 =A0) 1852 +- 9 =A0 =A0
=A0Moves:=
 133 =A0 =A0Cubes: 15
> > Monty =A0 =A0 =A0 =A0 =A0 (Average of 7 =A0) 1824 +- 17 =A0 =A0 Moves:
1=
46 =A0 =A0Cubes: 30
>
> Thanks Kees. I have one question, in GnuBG doesn't it use 2050 as
perfect
> play. The numbers pretty much jive except they are about 50 lower than
wha=
t
> Gnu would generally show.
>
> I understand the Gnubg rating calculation is an estimate based on
Checker
> and Cube play. With that being said - An interesting thing to note is
that=

> if you take 1824+17 you get 1841, and 1852-9=3D1843. Within the expected
e=
rror
> range we could be seen as pretty much even, with a slight edge in my
favor=
..


  It should say 95% CI instead of 5%. A common misinderstanding with
confidence intervals may be at play here. Suppose that your lower
amount and Monty's upper amount actually did meet. For example, you
were at 1850 instead of 1852. That does not mean that there is only
95% confidence that your rating is higher. It is actually higher than
that as two unlikely things would have needed to happen. Your true
rating needs to be much lower than the prediction and alos Monty's
much higher than predicted.

Bob Koca