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

Re: st: Re: svyset for stratified probability proportional tosizedesign

From   Steve <>
Subject   Re: st: Re: svyset for stratified probability proportional tosizedesign
Date   Thu, 11 Aug 2005 15:58:11 -0500


Thanks for your helpful response; it also helps reconcile the apparently contradictory advice of textbooks and Stata examples.

Can you confirm that at whatever stage of a complex sample design where sampling with replacement is used, all lower stages using sampling without replacement below that are ignored? Or is it only when the first stage setting up PSUs is sampled with replacement that the subsequent sampling stages without replacement are ignored? And, are they only ignored in Stata's svyset command or with any complex survey analysis package?

If the lower stages are ignored, how can one take into account possible clustering at those stages, such as with blocks within census tracts or households within blocks?


----- Original Message ----- From: "Nick Winter" <>
To: <>
Sent: Thursday, August 11, 2005 12:05 PM
Subject: Re: st: Re: svyset for stratified probability proportional tosizedesign

You only -svyset- one weight variable--which correspond to the ultimate probability of selection of each observation. With multiple stages of sampling, these weights will be a function of the probabilities of selection at each stage, possibly with additional corrections for non-response, etc. So the sample might be self-weighting at the first stage, but perhaps not at lower stages, resulting in non-constant weights for the ultimate observations.

So you would do something like:

. svyset psu1 [pw=weight] , str(strat1) || psu2 , fpc(fpc2) || _n , fpc(fpc3)

Note, however, that if the first stage is actually sampled with replacement, than lower-level without-replacement sampling is ignored. This will be conservative--your reported standard errors will be somewhat larger than they would otherwise be.

--Nick Winter

At 02:37 AM 8/11/2005, you wrote:

We run on Stata 9, updated all July 5.

We have a complex survey design data set with the first stage stratified
into two parts. The primary sampling units (census tracts) have been drawn
with replacement from each of the two strata with probability proportional
to population size (PPS) in each of the strata.

We have the total population, the population of each stratum, and the
population of each PSU. There are two further stages in the sampling
design, random sampling of 3 blocks within each PSU without replacement and
random sampling of 2 households within each block without replacement.

Our difficulty lies in specifying specifying svyset for this design,
particularly selecting (or not) the pw weights for the first level. On the
one hand, we read from texts that the glory of probability proportional to
population sample is that it doesn't need weights. On the other hand, we
see from svyset examples at the UCLA site, that considerable effort has
gone into calculating the values for the first stage weights with PPS
design, though without access to the text, we are unclear how to apply their

Could any suggest how to at least set up svyset for the first stage of the
above described design?

Many thanks,
Steve Rothenberg
Instituto Nacional de Salud Pública
Cuernavaca, México

* For searches and help try:
Nicholas J. G. Winter                     607.255.8819 t
Assistant Professor                       607.255.4530 f
Department of Government     e
Cornell University w
308 White Hall
Ithaca, NY 14853-4601

*   For searches and help try:

*   For searches and help try:

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