Dear statalist members,
In attempting to replicate the latent class model for rankings from the
Gllamm manual, I keep running into trouble when issuing the last portion of
the command (see below). I get that initial values are not feasible, and the
r(1400) Stata error message (numerical overflow). Using weights, the
effective observations are about 20,000.
For a reference to this model see section 9.4 in Gllamm manual (October 28,
2001).
I use the latest version of both Stata (8.2 Special Edition) and Gllamm.
Below is the output from the latent class example with rankings, when
running Gllamm (this is the last step, after reshaping the data).
I would appreciate any comment on this.
Best regards,
Carlos
. gllamm item alt1 alt2 alt3, expand(set chosen o) i(patt) link(mlogit) /*
> */ family(binom) nrf(3) eqs(alt1 alt2 alt3) nocons weight(wt) nip(2) /*
> */ ip(f) trace
General model information
----------------------------------------------------------------------------
--
dependent variable: item
nominal responses: mlogit
denominator: 1
equation for fixed effects alt1 alt2 alt3
Random effects information for 2 level model
----------------------------------------------------------------------------
--
***level 2 (patt) equation(s):
(3 random effect(s))
class 1
location for random effect 1
z2_1_1: alt1
location for random effect 2
z2_2_1: alt2
location for random effect 3
z2_3_1: alt3
log odds for level 2
p2_1: _cons
number of level 1 units = 20358
number of level 2 units = 2262
Initial values for fixed effects
(using gllamm for inital values)
----------------------------------------------------------------------------
--
Iteration 0:
Coefficient vector:
item: item: item:
alt1 alt2 alt3
r1 0 0 0
initial values not feasible
(error occurred in ML computation)
(use trace option and check correctness of initial model)
r(1400);
end of do-file
r(1400);
*
* 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/
