st: Please help: Two-part random model with clumping at zero for correlated healthcare costs (MIXCORR in SAS)

[Anees Abdul Pari <anees.pari@dph.ox.ac.uk>]

Dear all,


Hello there. I have a repeated measures dataset where the primary outcome is a continuous variable (total costs) with a substantial proportion of zero values and continuous non-zero (positive) values that are right-skewed. Ideally, I would like to use a two-part model where the outcome variable has a binomial distribution with logit link for the zero versus non-zero part and a gamma distribution with logit link for the non-zero part. As the main outcome is clustered on individual, I would like to use a GLLAMM model to account for the correlation between measures (random intercept for individuals- as a basic model) whilst adjusting for excessive zeros.
Is there a way where I can use gllamm to run this two-part random effect model analysis or is it possible to use TPM command with some kind of adjustment for random effects? I am struggling to come up with a proper code.
I have seen similar queries in the past and all were answered or authors turned to SAS to carry out their analysis

http://www.stata.com/statalist/archive/2013-01/msg00255.html

http://www.stata.com/statalist/archive/2005-01/msg00438.html


Many thanks

Anees
PS.
I would like to use the following command, but after accounting for excessive zeros in the main outcome “costs"
xi: gllamm costs i.x1 i.x2 x3 x4 x5 x6 ,i(id) family( gamma ) link( log) adapt trace
is there any way I can include two families and links in the above command (one for the non-zero  part and one for the zero part

*
*   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/