Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Fitting a two-equation zero-inflated model in GLLAMM


From   Serguei Kaniovski <Serguei.Kaniovski@wifo.ac.at>
To   statalist@hsphsun2.harvard.edu
Subject   st: Fitting a two-equation zero-inflated model in GLLAMM
Date   Fri, 8 May 2009 08:37:58 +0200

Dear Stata / GLLAMM Experts!

I am trying to fit a latent variable GLS using GLLAMM. My dependent
variable "dist" is bound to the interval [0, sqrt(2)], about 11 percent of
observations are zeroes.

The zeroes denote concordant unanimous votes between a pair of parties in a
parliament. If all members of party A and all members of party B voted yes,
then dist=0. (sqrt(2) appears due to a special distance function I use).

I view zeroes is not qualitatively different from non-zeroes, but the
inflated-zero property has to be accounted for. I thus plan to use a
two-equation model, with the same set of explanatory variables but possibly
a different set for each model later.

The independent variables are "left_right" and "pro_eu", while "vote_id"
identifies the voting occasion.

I fit a level-1 GLLAMM, i.e. with random effects for each voting. Ignoring
the fact that dist is bound from above by sqrt(2), I do:

gen v = cond(dist==0,1,2)

foreach x of varlist left_right pro_eu {
gen v1_`x' = `x'*(v == 1)
gen v2_`x' = `x'*(v == 2)
}

gllamm dist v1_* v2_*, i(vote_id) link(logit reciprocal) family(binary
gamma) lv(v) fv(v)

Unfortunately the above specification does not converge. Is the
specification correct given what I am trying to do?

That you for your help,
Serguei

*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index