Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Re: Stata's complex sampling - how deep does complex sampleanalysis have to go?

From   "Arnold Levinson" <[email protected]>
To   "Statalist" <[email protected]>
Subject   st: Re: Stata's complex sampling - how deep does complex sampleanalysis have to go?
Date   Sat, 7 Aug 2004 21:21:17 -0600

Leslie Kish (Survey Sampling, Wiley & Sons) shows that reliance on PSU-level
variance is conservative, i.e., it overestimates variance in multi-stage
sampling. When selected clusters comprise a small proportion of total
clusters, as is the case in all national surveys and most state-level
surveys, the overestimate is negiligible. Seems to me the advantages of
SUDAAN might be attractive when total sample size is small or when the
sampling fraction in a finite population is quite large (e.g., more than

Arnold H. Levinson, Ph.D.
Associate Scientist
AMC Cancer Research Center
Assistant Professor
University of Colorado School of Medicine
[email protected]
Date: Thu, 05 Aug 2004 11:34:32 -0700
From: "Thomas W BRUNDAGE" <[email protected]>
Subject: st: Stata's complex sampling - how deep does complex sampleanalysis
have to go?

People -

This came my way in  discussion on the relative merits of SPSS, Stata, and

"If you are using STATA as a replacement for SUDAAN you need to be aware
that only the first couple (I think it is two) levels of complex
sampling can be accounted for...  then you have to assume simple random
sampling...  this may work for some of your work.... SUDAAN accounts for
up to 8 levels of nested sampling.  So, if there are very complex sampling
strategies, you may have to resort to SUDAAN anyway.

I have come across a couple of CDC sponsored surveys that require this...
but not all."

>From my reading of the Stata V8 Manuals, esp [U]30, I find that the
variance estimators in the current version of Stata are suitable for use in
multistage sample designs even though they are based on computations at the
primary-sampling-unit level.  There is indication that other methods, not
yet implemented, account for secondary sampling.

Given that Stata uses the primary-sampling-unit level to compute the
variances, how much accuracy is lost by not going down to the other levels
of nesting (such as to the 8th level)?


thomas brundage
Department of Human  Services
Oregon Public Health
800 NE Oregon Street, #827
Portland OR 97232

phone: 503.731.3448
FAX:    503.872.5398

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index