"Ray Koopman" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
> "Christian JAMMES" <[EMAIL PROTECTED]> wrote in message
news:<[EMAIL PROTECTED]>...
>
> > Assuming a set of measurements:
> > (m(i); sd(i)) with u(i) = measured value and u(i) = standard uncertainty
> > If one could propose the overall mean value mtot as being the weighted
mean
> > value:
> >
> > mtot = sum(w.*m/sum(w)) with w(i)=1/u(i)^2,
> >
> > what the hell could be the overall variance vartot:
> >
> For any fixed weights w(i) and mutually independent measures m(i),
>
> var(sum(w(i)*m(i))/sum(w(i))) = sum(w(i)^2*var(m(i)))/(sum(w(i)))^2.
And if he wants to see why his original weights are those that give minimum
variance, see the last part ("Sketch of weighting by HDOP analysis") of:
http://users.erols.com/dlwilson/gpseqtns.htm
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
