Dear Statalisters,
I am trying to estimate a multilevel structural equation model using gllamm. Basically I am interested in the effect of a latent on an observable variable. This latent variable in turn is formed out of three other latent variables and I have a number of fallible measures for each of these three. I have coded this model and have pasted the relevant parts at the bottom. It runs ok for some subsamples; for others, however, the model does not converge (with the messages “discontinuous region encountered” or “backed up” or “not concave”). Does anyone have any advice? Any help would be greatly appreciated.
In the archives someone else appeared to have the same problem but there was a single observation per client, which is not my case.
Thanks a lot already
Marco
Do File:
*********************************************************
*********************************************************
. *** Define Latent variables
.
. eq fac1: depvar
. eq fac2: latent1 latent2
. eq fac3: latent3 latent4
. eq fac4: latent5 latent6
. eq cons1: cons
.
. *** Regress latent variables on covariates
.
. eq f2: cov1 cov2 cov3 cov4
. eq f3: cov1 cov2 cov3 cov4
. eq f4: cov1 cov2 cov3 cov4
.
. *** Matrix of covariances among latent variables
.
. matrix B = (0,1,1,1,0\0,0,1,1,0\0,0,0,1,0\0,0,0,0,0\0,0,0,0,0)
.
. *** Set constraints for B matrix
.
. constraint def 1 [b1_2]_cons = 1
. constraint def 2 [b2_4]_cons = 1
.
. gllamm outcome covariate , ///
> i(kids_id mother_id) eqs(fac1 fac2 fac3 fac4 cons1) geqs(f2 f3 f4) link(logit) ///
> family(binom) nip(2,2,2,2,2) nocorrel nrf(4,1) frload(1) bmatrix(B) constr(1/2) trace
General model information
------------------------------------------------------------------------------
dependent variable: outcome
family: binom
link: logit
denominator: 1
equation for fixed effects covariate _cons
Random effects information for 3 level model
------------------------------------------------------------------------------
***level 2 (kids_id) equation(s):
(4 random effect(s))
lambdas for random effect 1
kid1_1l: depvar
standard deviation for random effect 1
kid1_1 : depvar
lambdas for random effect 2
kid1_2l: latent2
standard deviation for random effect 2
kid1_2 : latent1
lambdas for random effect 3
kid1_3l: latent4
standard deviation for random effect 3
kid1_3 : latent3
lambdas for random effect 4
kid1_4l: latent6
standard deviation for random effect 4
kid1_4 : latent5
***level 3 (mother_id) equation(s):
(1 random effect(s))
standard deviation for random effect 5
mot2_1 : cons
B-matrix:
b1_2: _cons
b1_3: _cons
b1_4: _cons
b2_3: _cons
b2_4: _cons
b3_4: _cons
Regressions of random effects on covariates:
equation for random effect 2
f2: cov1 cov2 cov3 cov4
equation for random effect 3
f3: cov1 cov2 cov3 cov4
equation for random effect 4
f4: cov1 cov2 cov3 cov4
number of level 1 units = 15506
number of level 2 units = 1943
number of level 3 units = 1266
Constraints:
( 1) [b1_2]_cons = 1
( 2) [b2_4]_cons = 1
estimating 27 parameters
*********************************************************
*********************************************************
--
Jetzt kostenlos herunterladen: Internet Explorer 8 und Mozilla Firefox 3.5 -
sicherer, schneller und einfacher! http://portal.gmx.net/de/go/atbrowser
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
