st: FW: gllapred, mu following gllamm with multiple equations/responses?

["Rossen, Lauren M. (CDC/OSELS/NCHS)" <vyz5@cdc.gov>]

Hello,

I am using gllamm in Stata/SE 12.1 to model a zero-inflated outcome (e.g., dietary intake of a specific food).  The first 'part' of the outcome is the probability of consuming the food, and the second 'part' is the amount of food (given that some is consumed) - similar to a hurdle model in economics.  Since probability and amount are theoretically correlated, the two parts share a person-specific random effect (each person has 2 observations).  It seems I might be able to use the following type of model (adapted from Dr. Kolenikov's page on gllamm examples: http://www.unc.edu/~skolenik/stata/gllamm-demo.html):

eq het : s1  
eq load: s1 s2  
gllamm outcome x1_1 x2_1 x1_2 x3_2 s*, ///  
link(logit id) lv(v) fam(bin gauss) fv(v) ///  
s(het) i(personid) eq(load) nocons  

since this kind of approach would capture the separate probability and amount components of the outcome.  However, ultimately I would like to produce a predicted value for each unit that corresponds to the predicted probability*amount (incorporating the random effects), such as I might obtain with gllapred, mu following two separate gllamm's.  The gllapred, mu command seems to work following the above model, but I'm not sure how to interpret it.  For example, the lower bound is sometimes negative (and intake cannot be less than zero).  Can anyone describe what gllapred produces following this kind of multiple equation/response gllamm model?  Or alternatively, if there might be a better way to tackle this problem?  

Thanks so much,
Lauren Rossen, PhD, MS
Postdoctoral Research Fellow
Centers for Disease Control and Prevention

*
*   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/