Re: st: Hierarchical model

[Evans Jadotte <evans.jadotte@uab.es>]

Maarten buis wrote:
> --- On Sat, 3/10/09, Evans Jadotte <evans.jadotte@uab.es> wrote:
>   
> > .xi: quietly xtmixed logequiv i.education i.mpsex
> > mpage* i.sector non_farm
> > small_cattle large_cattle s_s_cattle shock_cattle dumland1
> > tech_land
> > electricity landline piped_water road_access death d_rem
> > i.hhtype
> > ||cluster:, var
> >
> > estimates store model1
> >
> > .xi: quietly xtmixed logequiv i.education i.mpsex mpage*
> > i.sector non_farm
> > small_cattle large_cattle s_s_cattle shock_cattle dumland1
> > tech_land
> > electricity landline piped_water road_access death d_rem
> > i.hhtype ||strata:
> > ||cluster:, var
> >
> > . estimates store model2 
> > . lrtest model1 model2
> >     
>  
> This test is a test of a null hypothesis on the edge of the 
> parameter space (the strata level variance is >= 0, so 0 is
> on the edge of allowable values for that variance). Things 
> get weird when you perform such a test, so be careful.
> For more on this see: -help j_chibar-.
>  
>   
> > The last test strongly suggests that cluster is nested into
> > strata and that the latter should be in the model. I tried 
> > estimating the other way around (strata nested into cluster) 
> > just to check, and convergence failed. So, based on the
> > three-level model:
> >     
>
> It appears you are trying to empirically determine the nesting
> structure in your data. You should not need to do this as it
> should be a-priori clear from the design of your study. If the
> nesting is not strictly fixed by design you are probably 
> dealing with a crossed random effects model. In any case the
> nesting structure is determined by the design of your study and
> is not an empirical question. 
>
> Hope this helps,
> Maarten
>
> --------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
>
> http://www.maartenbuis.nl
> --------------------------
>
>
>       
>
> *
> *   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/
>   
Hi Maarten,

Thanks for your input. I must say in advance that I am relatively new in 
Stata and my knowledge is very limited, so any help would be much 
appreciated.

I tried the estimation (cluster nested into strata) just to validate the 
results from estimating cluster nested into strata. The nesting is 
strictly fixed by design so I believe it is not a crossed random effects 
model. To be more specific, "clusters" are in my study "villages or 
communities" while "strata" are "administrative regions", each of which 
includes various communities (something similar to a Länder in Germany). 
I understand now that the H0 for this lrtest implies:

Var(strata) = Cov(strata, cluster) = 0

In my model, I need Cov(strata, cluster) = 0, while Var(strata) > 0.

If one of the conditions of the H0 fails (in this case Var(strata) ≠ 0), 
what would be the procedure to follow? In fact, the result I am looking 
for Var(strata) > 0 and Cov(strata, cluster) = 0 is what I got.

Thanks,

Evans
*
*   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/