st: gllamm: Two-phase study with non-random missing?

["Stein Atle Lie" <Stein.Lie@smis.uib.no>]

Dear list members

Any bright ideas on this?

In a troublesome population study over two phases we have non-random 
missing information.
In the first phase there are two sources/informants screening the 
population (y1_1 and y1_2: 0=negative, 1=positive, and -1=missing).
For the second phase diagnosis is evaluated for a sub-sample using a 
more precise tool (y2: 0=No, 1=Diagnosis, -1=missing). 
(This is not a random sample, due to several factors.)
wt1 is the number of cases with the patterns of y1_1, y1_2 and y2.

      y1_1    y1_2    y2    wt1   patt
      -1       0      -1   1929    1
      -1       1      -1    615    2
       0      -1      -1    303    3
       0      -1       0      4    4
       0       0      -1   4769    5
       0       0       0    352    6
       0       0       1      6    7
       0       1      -1    370    8
       0       1       0    180    9
       0       1       1     17   10
       1      -1      -1     78   11
       1      -1       0      9   12
       1      -1       1      2   13
       1       0      -1    266   14
       1       0       0    203   15
       1       0       1     25   16
       1       1      -1    214   17
       1       1       0    122   18
       1       1       1     87   19

This should (?) be able to set up in gllamm, or...?
I generate the long-data-file with all responses (for y1_1, y1_2 and y2) 
different from -1 in one variable (e.g. y)
Thus patt=1 will have one line and e.g. patt=6 will have 3 lines.
Furthermore, I make 3 dummy variables for where the information in response 
is obtained (y1_1, y1_2, or y2), and if the information in the given pattern 
is valid for each of the variables (non-missing: Vy1_1, Vy1_2 and Vy2) or not 
(missing: nVy1_1, nVy1_2 and nVy2). 
The problem is not very different from example 14.3, page 422, in Generalized 
Latent Variable Modeling by Skrondal and Rabe-Hesketh. 
See also: http://www.gllamm.org/books/readme.html
(However, they only have missing information for one measurement)

My rather extensive problem is thus:
Problem1: To set up the Exposure model, Measurement model, and Disease model. 
It becomes a long (!) row of equations (there must be potential to reduce the problem?!)
Problem2: To find good documentation for the constraints-syntax.
Problem3: The total syntax for this problem in gllamm.
Problem4: Last the idea is to run gllapred to estimate population probabilities e.g for P(y2=1).

Regards
S.A. Lie
Stein.Lie@halos.uib.no

*
*   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/