Bookmark and Share

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


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

Re: st: Strata, PSU, and Bootstrap Issues with svyset


From   Steve Samuels <[email protected]>
To   [email protected]
Subject   Re: st: Strata, PSU, and Bootstrap Issues with svyset
Date   Wed, 29 Jan 2014 21:50:19 -0500

Two levels of problems here:

1. To describe the sample dataset, don't use -svy- commands. Those
describe the population. Use ordinary non-survey descriptive commands
instead.

2. The single-unit- strata problem, though not relevant to your question
now, will not be solved by -bootstrap- or -jackknife- techniques. The
problem usually occurs because PSUs are dropped if one or more of the
variables is missing for all observations in a one of your "male" categories,
though it's barely conceivable you have all male and all female PSUs. Its up to you
to investigate what's missing and where.

The problem can be solved "automatically" by setting a -singleunit()
option in your -svyset- statement. The most conservative is
-singleunit(centered)-.  If consecutively numbered strata are geographic
neighbors, I tend to form new strata that amalgamate a sufficient number of neighbors to get 2+ PSUs.




Steve [email protected]



On Jan 29, 2014, at 3:52 PM, "George-Jackson, Casey Elizabeth" <[email protected]> wrote:

Hello,

I am analyzing a complex survey design that has 360 strata and approximately 2 PSUs per stratum. The main analysis for the study uses propensity score matching. To describe the dataset, I am attempting to generate a means table for the variables I am using for the PSM, but am encountering issues with the strata, PSU, and bootstrapping. 

The following syntax displays the means, but no standard errors and no confidence interval: 

    svyset PSU [pw=F2BYWT], strata(STRAT_ID)
    svy: mean v1 v2 v3…., over(male)

A note appears after that table that missing standard errors because of stratum with single sampling unit.

I found that 28 of the 360 strata have only 1 PSU. I then attempted to bootstrap the analysis: 

    svyset PSU [pw=F2BYWT], strata(STRAT_ID) vce(bootstrap) bsrweight(F2F1P1-F2F1P200) bsn(10)
    svy: mean v1 v2 v3…., over(male)

The resulting means table has no SEs and no confidence interval. The bootstrap command does not appear to work because the typical output generated when running bootstraps does not appear (replication dots are not suppressed). I receive the same note after the table that states missing standard errors because of stratum with single sampling unit. 

I then attempted to work around this by running:

    egen group=group(STRAT_ID PSU)
    svyset group [pw=F2BYWT], vce(jackknife)
    svy: mean v1 v2 v3…., over(male)

This provided the means, SEs, and confidence intervals from all of the PSUs (717), but only one stratum. 

Am I using the bootstrap syntax incorrectly with svyset? I also looked into using brrweights but am unsure what information to use to calculate Fay's adjustment. 

I am using Stata 13 on Windows XP (32bit)

Much appreciation for your time in advance!
Casey


Casey E. George-Jackson, Ph.D.
IES Postdoctoral Research Fellow
University of Illinois at Urbana-Champaign
1310 S. Sixth Street, MC 708
Champaign, IL 61820
[email protected]


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


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