Statalist


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

st: correcting standard errors due to correlated error terms in first differenced model


From   "Brent Fulton" <fultonb@berkeley.edu>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: correcting standard errors due to correlated error terms in first differenced model
Date   Mon, 11 Feb 2008 09:51:19 -0800

Dear Statalist,

I am using Stata 9.2 to analyze panel data that used a complex sampling
design and would like to choose the most appropriate svyset to handle
potentially correlated error terms.

The sampling design included multiple stages that were summarized into one
psu_id, one strata_id, and a pweight variable--due to privacy concerns, I
don't have any more transparency e.g., ssu_id does not exist in the data.
With -svy: reg-, I am estimating a first-difference model and am concerned
about serial correlated first-differenced error terms within individuals.
However, I am also concerned about first-differenced error-term correlation
across individuals within the same sampling unit (e.g., one sampling unit
level is a classroom of students; note students might change classrooms over
time). I'd like to use the following svyset:

(A) svyset psu_id [pw=weight], strata(strata_id) || child_id

but get this note: "stage 1 is sampled with replacement, all further stages
will be ignored." (Although the PSUs were sampled without replacement, I
can't specify without replacement in the first stage because I don't have an
fpc variable.)

So given the above, which svyset best handles the correlated error terms:
(B), (C), or (D), or something else? (B) seems to handle first-differenced
error-term correlation across individuals. (C) seems to handle
first-differenced error-term correlation within individuals. (Note I don't
see how (D) handles the issue, but it is an option.)


(B) svyset psu_id [pw=weight], strata(strata_id)

(C) svyset child_id [pw=weight], strata(strata_id)

(D) svyset [pw=weight], jkrweight(weight1-weight90, multiplier(1))
vce(jackknife) mse


Also, I was trying to figure out how Stata estimated the variances of the
parameter estimates. Under "variance estimation" in [SVY], it shows how the
survey design (e.g., strata, psu's) are incorporated into calculating the
variance of -svy: total-, but it doesn't show how the survey design is
incorporated into calculating the variance of B-hat's (where it uses a
Taylor series expansion)--but it must be incorporated, so was wondering what
I was missing (see pg. 258, 263)?

Thanks,
Brent

______________________________________________________
Brent D. Fulton, PhD
Health Services Researcher
The Nicholas C. Petris Center on Health Care Markets and Consumer Welfare
The Global Center for Health Economics and Policy Research
School of Public Health
50 University Hall, MC7360
University of California, Berkeley
Berkeley, California 94720
Phone: 510-643-4102 
Fax: 510-643-4281
Email: fultonb@berkeley.edu
www.petris.org
www.gchepr.org 



*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index