[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: RE: reprogramming survwgt ?

From   Nick Winter <>
Subject   Re: st: RE: reprogramming survwgt ?
Date   Fri, 05 Sep 2008 12:43:42 -0400

Indeed; I just don't know what the number of iterations would tell you, either as a comparison across algorithms, or as a comparison across datasets. So what if raking dataset X took more iterations than raking dataset Y, as long as both converged?

- NW

Nick Cox wrote:

Just to make a few banal remarks on the number of iterations:
Users would be able to compare #iterations on different data with the same program if that mattered to them.
Any comparisons with any results published elsewhere by what is supposedly the same algorithm should however be highly circumspect.
Results will depend a little on how the algorithm is initiated and a lot on the criteria for convergence.
Rosenblueth and Wiener pointed out that the best model of a cat is another cat -- no, preferably the same cat. In the same way, only -survwgt- can be compared with itself.
This is not, directly or obliquely, a criticism of -survwgt-, which I have every reason to believe to be an excellent program.
Nick Winter

As the author of the program in question, I've added an option to show the number of iterations completed (although I'm not really sure why one might care), along with an option to control the maximum number of iterations the program will go through before giving up.

I'll make an announcement on-list once the new versions are up on SSC.

* For searches and help try:
Nicholas Winter                                 434.924.6994 t
Assistant Professor                             434.924.3359 f
Department of Politics         e
University of Virginia w
PO Box 400787, 100 Cabell Hall
Charlottesville, VA 22904
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index