Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: replication weights for data with strata with single PSU

From   Steven Samuels <>
Subject   Re: st: replication weights for data with strata with single PSU
Date   Sun, 13 Feb 2011 11:55:18 -0500

Patrick, the FAQ ask that you reference contributed commands. -svr- is a package written by Nick Winter and can be found by typing "findit svr".

Graubard and Korn (Analysis of Health Surveys, 1999, Wiley, p. 204) have a section on your question. The best approach is to collapse strata "that are most alike with regard to presampled characteristics". To use all the data, I would split the largest of the single strata into two pseudo-strata. Collapsing will increase the apparent variance and splitting will decrease it, so the ultimate estimate will probably be conservative. If you want to analyze that split stratum stratum as a domain, then you should join the pieces again.


On Feb 12, 2011, at 2:58 PM, McCabe, Patrick J wrote:


I have a question about how to create replication weights using svr. I'm working with complex sampled nationally representative data. The data includes non-self representative strata that have two PSUs and self representative strata with only one PSU. To use svr I need two PSU for each strata, my question is is it better to collapse self representative strata with only one PSU together (unfortunately there is an odd number of these)

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index