Re: st: Predictions based on reoprob and gllamm

[Buzz Burhans <wsb2@cornell.edu>]

Dear Sophia,

I have used -gllamm- with adaptive quadrature to model a repeated 
measurement in time of a continuous outcome , using a model that has both a 
level 2 random intercept and slope.  I would like to be able to make a 
statement about the relative proportion of the total variance associated 
with the 2nd level random effects relative to the proportion that is 
explained by a fixed effect treatment variable.

My reading (Snijders & Bosker, 1999; Hox, 2002) suggests this is 
problematic for a number of reasons. However, I think, in my case, that it 
would be very useful to be able to make such a statement, even if only 
approximate, about the relative magnitude of the total variance 
attributable to the level 2 random effects as it contrasts with the 
relative magnitude of the variance explained by a treatment variable. ( In 
my case, I would like to point out that despite the treatment being 
significant, the magnitude of the effect is small versus the inherent 
variance ascribable to the level 2 units.)

Can you suggest a reasonable basis for making this statement when gllamm is 
used for a model such as this? I would be more comfortable deriving an 
approximate R2 if the model was a random intercept only, I am unsure of how 
to do this given a random coefficient also.  I will confess to following 
conceptually, but not specifically, the Snijders & Bosker discussion, and 
don't think I am proficient enough to implement it.

I would be most appreciative of any insight you can share on this.

Thanks very much.

Buzz Burhans

Buzz Burhans
wsb2@cornell.edu


*
*   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/