Statalist


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

st: Latent Class Analysis using gllamm


From   Ijon Tichy <[email protected]>
To   [email protected]
Subject   st: Latent Class Analysis using gllamm
Date   Mon, 9 Mar 2009 00:18:32 +0100

Dear All,

I want to do a confirmatory latent class analysis using gllamm. The
latent dependent variable is the social class of respondent, and there
are three categorical manifest variables:

occupation – the occupation of respondent’s father, with 534 categories;
education – the level of education of respondent’s father, with 5 categories;
income – the level of income in respondent’s household when a
respondent was 15, broken down into 6 categories.

I want to compare two models: first assuming one latent class (the
complete independence model) and second assuming seven latent classes
(my substantive theory suggests such a number of latent classes).
After reading McCutcheon 1987; Skrondal and Rabe-Hesketh 2004;
Rabe-Hesketh and Skrondal 2008; and a number of previous threads on
the latent class analysis using gllamm on the Statalist (particularly
this one: http://www.stata.com/statalist/archive/2004-10/msg00881.html
and this one: http://www.stata.com/statalist/archive/2009-02/msg00437.html),
I have come up with two specifications.

Common part for two models:

tabulate occupation, generate(occupdum)
tabulate education, generate(educdum)
tabulate income, generate(incomedum)

eq class: occupdum1 … occupdum534 educdum1 … educdum5 incomedum1 … incomedum6

eq latent: occupation education income

For the complete independence model:

gllamm class, i(respondent) eq(class) geq(latent) nrf (545)
link(mlogit) family(binom) ip(f) nip(1) adapt

For the seven-class model:

gllamm class, i(respondent) eq(class) geq(latent) nrf (545)
link(mlogit) family(binom) ip(f) nip(7) adapt

My question is whether the above specifications are correct in the
sense that the first would estimate the complete independence model
and the second would estimate the seven-class model, which I could
later use for further comparisons? Have I got the logic of specifying
models for latent dependent variables with gllamm right? Or is there
something that I have omitted or – conversely – added in the model
specifications wrongly?

I would very much appreciate your help.

Best regards,
Ijon

*
*   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 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index