Re: [Bug-gnubg] std and j.s.d

[Jim Segrave]

On Mon, Jul 28, 2003 at 03:22:10PM +0200, Jim Segrave wrote
> On Mon 28 Jul 2003 (11:21 +0000), Joern Thyssen wrote:
> > On Mon, Jul 28, 2003 at 12:43:08AM +0200, Jim Segrave wrote

snip an explanation of the use of joint standard deviations in
deciding to stop a rollout when move A appears to have a significantly
higher equity than move B. Assuming 

(equity(A) - equity(B)) == 1.96 * joint standard deviation of the equities

Then we can conlude that we are 95% confident that A is a better move
than B. I then added:

> > Or perhaps somewhat higher, as a normal distribution is symetrical and
> > of the 5% uncertainty, 2.5% represent cases where A has an equity more
> > than 1.96 j.s.d.s higher than B.

> Yes, good point. Since we're looking at the difference between
> equities we're not interested in the upper limit. Maybe you should post
> this to the list in order to get some feedback Zare or another math
> wizards.

And here I defer to those better trained than me - can one extend the
confidence to 97.5% in this case (if you believe that the rollouts
give reasonably good approximations to normal distributions)?

-- 
Jim Segrave           address@hidden