Re: st: correct for autoregression in log link function in survey data

[Stas Kolenikov <skolenik@gmail.com>]

What is the autocorrelation that you have in your survey? What exactly
do you mean by "accounting" for it?

The standard errors are arguably easier to fix. If your survey design
has measurement occasions nested in households nested in PSUs, then
your -svy- already controls for an arbitrary correlation patterns
within PSUs (including correlated observations within the household).
In other words, you don't need to do anything.

Theoretically, you can obtain more efficient point estimates if you
account for autocorrelation between observations nested in the
household. -glm- won't allow for that, so you would need to look
elsewhere -- for instance, at -xtgee-. However, the latter is neither
supported by -svy-, nor does it have -vce(cluster)- option. I am
guessing that you can use -bsweights- with -svy bootstrap- to run
this.

If it works, I would be quite curious to see how big your efficiency
gains are (by how much your standard errors change).

On Mon, Jan 9, 2012 at 12:57 PM, Li, Rui (CDC/ONDIEH/NCCDPHP)
<eok8@cdc.gov> wrote:
> I am using a log link fuction to estimate cost data from a survey, using
> svy:glm cost $char, link(log) family(gamma) le(95)
>
> How can I account for auto-correlation in this model?
>
> Thanks!
>
> Rui
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/



-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

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