"David Jones" <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>...

snip good stuff

> The distribution for the sum is not exactly logistic
> but you can expect it to be somewhere between a logistic and a Normal.

I now believe that is going to be generally true. I haven't been able
to figure a way to make it look heavier-tailed than logistic (perhaps
not surprising given it's log-concave).

If the logistic is a bound, then that will give much better bounds 
than the a chebyshev-like bound for symmetric unimodal.

> The formulae will give you the kurtosis, which is some indication of
> shape, and of couse you know that the distribution is symmetric.
> 
> Johnson & Kotz suggest that others have found that a student-t
> distribution is a reasonable approximation to the logistic and you may

Matching kurtosis and scaling for variance (and shifting for location)
in the t might do reasonably as an approximation for the sum, as long 
as you don't go really far into the tails.

Glen
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to