Why not use -xtmixed- with random intercepts for the households and regions, then use dummy variables for the states as fixed effects? Also, what is the sampling design? If you're working with survey data, you should -svyset- your data and in doing so can avoid having to manually adjust things.
HTH,
Billy
Sent from my iPhone
On May 3, 2013, at 5:20, Rüdiger Vollmeier <ruediger.vollmeier@googlemail.com> wrote:
> Hello,
>
> Ich have an observational dataset at hand (Household Survey,
> cross-sectional, 2800 observations). I am wondering at which level I
> should cluster the standard errors (e.g. household level (around 1500
> observations), regional level (70 observations), state level (20
> observations)) in a linear regression.
>
> - reg depvar indepvar, cluster(cluster_level) -
>
> How do I test the appropriatness of the level at which I cluster?
>
> Kind regards,
> Ruediger
> *
> * 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/
