Newsgroup sci.stat.math 14311

Forgive the intrusion if this is a FAQ.  Years ago someone
went to great pains to explain to me why, when reporting
calculated errors ("estimated standard deviations") you 
should go from two to one significant figure upon reaching
28.  i.e. 0.1234(28) was valid, but 0.1234(29) should be
reported as 0.123(3).  For the past 30 years I have been 
using this "rule of 28", but was recently told that I should
instead be using different criteria.  It turns out the the
International Union of Crystallography has specific 
recommendations (http://www.unige.ch/crystal/astat/recomm.html)
which suggest the standard deviation should be reported
between 2 and 19.  Can anyone point to a clear explaination
of a "correct" way to do it?
TIA
-- 
|===================================================|
| John C. Huffman                                   |
| huffman@indiana.edu  http://www.iumsc.indiana.edu |
|===================================================|

Jim C Chao (e0fdrdde@mail.erin.utoronto.ca) wrote:
> Hi, I understand it's easy to generate data for a univariate normal
> distribution from any math/stats software.  However, could anyone please
> tell me how to generate data for the multivariate normal distribution? 
> Any reference or book available?  I would appreciate it if you could
> help me on this.
	See Ernest M. Scheuer and David S. Stoller, "On the Generation
	of Normal Random Vectors," Technometrics, Vol. 4, No. 2, 
	May 1962, pp. 278-281.

In article <33963168.167E@bu.edu>, Ivo Grosse   wrote:
:I am looking for the generating function of ln(k+1), 
:i.e. what is:
:
:f(z) = \sum_{k=0}^\infty z^k / k! * ln(k+1).
:
:???
:
:Do you know a closed form expression??? Does this
:function have a name???
:
:
:Ivo
This one I don't know; but what is easy to find is
 g(z) = sum {k=1 to infinity} ln(k) / z^k}  (Real(z) > 1)
It is   - (d/dz) zeta(z).
Cheers, ZVK (Slavek).

 > From: Xelp95bpa@stoat.shef.ac.uk
 > Hi! I wonder if anyone could help. I've often seen texts which makes
 > use of
 > eigenvalues and eigenvectors. Also basic math books do tell us how to
 > derive these. But what does eigenvalues/vectors actually tell us
 > about
 > something, and how can we use the information?
What do you man "I've often seen texts" ? These texts should
have explained, what eigenvectors are used for.
You didn't bother to read them ? ;-)
What eigenvalues mean in a certain application of mathematical
methods can't be explained in general, you have to be more
specific.
E.g. the spectral lines in the light of stars are the eigenvalues
of certain equations of physics.
In solving least squares problems, you can make use of the
so called 'singular value decomposition'. The 'singular values'
are the eigenvalues of a matrix. These eigenvalues tell you
if the matrix is singular or not and 'how much' singular.
 > Also, what does the term regularisation mean? Many texts seem to make
 > use
 > of it, I think rather haphazardly.
The meaning of this term is domain specific. Authors could
invent artificial words, but then we would have to remember the
spellings ;-)