Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: crossed random coefficients using GLLAMM

Subject   st: crossed random coefficients using GLLAMM
Date   Fri, 7 Oct 2005 15:45:54 +0200

Dear all
I have data on dichotomous vote choices (i) nested within voters (j) nested
within referendums (k). Referendums comprise 1-6 propositions (thatís why I
potentially have multiple observations per voter). I have set up a GLLAMM-model
with random intercepts at the level of voters, propositions and referendums,
i.e. a cross-classified model where the cross-classification
(voters/propositions) is still nested within the higher level of referendums.
Unfortunately, Stata gives various error messages (conformability error,
likelihood cannot be computed etc.), and I have no idea whether they relate to
model specification and/or to computational difficulties. Therefore, I would
very much appreciate any comments on model specification:

/* generate intercepts for level j and k */
gen byte cons1=1
gen byte cons2=1
eq c1: cons1
eq c2: cons2

/* generate random effects for propositions 1,2,..,max per referendum to let
them randomly vary at level k */
eq p1: prop1
eq p2: prop2
eq p3: prop3
eq p4: prop4
eq p5: prop5
eq p6: prop6

/* constrain them to have equal variances =one random intercept at proposition
level!), i.e.  to hold the diagonal elements of the Cholesky decomposition
matrix equal; off-diagonal elements are constraint to 0 by nocorrel option in
GLLAMM (see below) */
constraint define 1 prop1=prop2
constraint define 2 prop2=prop3
constraint define 3 prop3=prop4
constraint define 4 prop4=prop5
constraint define 5 prop5=prop6

/* and this is the (empty) model.. */
gllamm choice, i(voter ref) link(probit) fam(bin) nrf(1,7) eqs(c1 c2 pro1 pro2
pro3 pro4 pro5 pro6) nocorrel constraints(1 2 3 4 5) nip(3) trace

In particular, I have some doubts whether to include c1, the random intercept at
the referendum level? Does anyone have some suggestions on model specification?
Thanks a lot for taking the trouble to help me!

Best, Peter

This message was sent using IMP, the Internet Messaging Program.
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index