Martin,
I'd recommend that you get a copy of Multilevel and Longitudinal Modeling Using Stata by Rabe-Hesketh and Skrondal. Chapter 5 provides an excellent demonstration of model building and evaluation. Without knowing more details of your specific dataset and hypotheses, I find it difficult to give advice.
Taking a simple example, let's say I have a growth-curve model. I begin with a random intercept model. In Stata's output, there's LR test vs linear regression. If that's significant, there's support for that random effect.
I then fit a random coefficient/slope model. Next, I conduct a likelihood-ratio test to compare this model with the random intercept model. If that's significant, there's support for including that additional random effect.
And it goes on from there.
Scott Millis
--- On Thu, 8/6/09, Martin kavao <[email protected]> wrote:
> From: Martin kavao <[email protected]>
> Subject: RE: st: How to get the P values of the random effects after running an xtmixed command
> To: [email protected]
> Date: Thursday, August 6, 2009, 1:37 AM
> Thanks Scott
> I thought about checking the confidence interval, but I
> thought they are
> always >0.
>
> About using the ICC to evaluate the model: does it mean if
> a model with an
> additional parameter (random effect) fits better, then the
> random components
> are all significant?
>
> Subject: Re: st: How to get the P values of the random
> effects after running
> an xtmixed command
>
> There's not a consensus regarding the nature, form, and
> effectiveness of
> single parameter tests for variance components. With
> that caution in mind,
> you can examine the confidence interval to see if it
> contains 0--to get a
> rough idea regarding the rejection of the null. A
> better method is to
> evaluate/compare models using the deviance statistic along
> with changes in
> ICC as you systematically build and evaluate your models.
>
> Scott Millis
>
>
>
> --- On Wed, 8/5/09, Martin kavao <[email protected]>
> wrote:
>
> > From: Martin kavao <[email protected]>
> > Subject: st: How to get the P values of the random
> effects after running
> an xtmixed command
> > To: [email protected]
> > Date: Wednesday, August 5, 2009, 11:14 AM
> > I am doing a multilevel analysis
> > using xtmixed command and I need to report
> > the significance of the random component in the final
> > models i.e. the p
> > values. Anyone has an idea how to compute it after
> running
> > the xtmixed
> > command
> >
> > thanks
> >
> > *
> > * 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/
> >
>
> *
> * 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/
>
>
> *
> * 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/
>
*
* 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/
