st: Re: mixed effects model and autocorrelation

[Kit Baum <baum@bc.edu>]

Just as -newey- is an alternative to -prais- for a single time  
series, you might consider -xtivreg2- as a substitute for -xtgls-.  
You can either model the autocorrelation as being of unknown form  
with no more than L significant autocorrelations, using the bw() and  
kernel() options, which will deliver HAC standard errors, or you can  
use the cluster(country) option, which will allow for cross-panel  
heteroskedasticity and arbitrary correlation within each panel's  
error structure. See Baum, Schaffer, Stillman, BC WP 667 on my CV  
below. NB: you need not have any instrumental variables to use - 
xtivreg2- (or all regressors can instrument themselves).

This option does not handle multilevel mixed effects, but those  
models are essentially random-effects models, and RE models have  
stronger assumptions about the independence of the error components  
from the regressors than do the fixed-effects models supported by - 
xtivreg2-, available from SSC.


Kit Baum, Boston College Economics and DIW Berlin
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html


On Oct 12, 2007, at 2:33 AM, statalist-digest wrote:


> Now to my question: How should I think about, and understand, the
> autocorrelation in the mixed effects model? With my (somewhat limited)
> experience with time-series data, I need to somehow take the
> autocorrelation into account. How is this done in a mixed-effect  
> model?
> Since years are nested within countries, will the estimated standard
> errors be valid?
*
*   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/