Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
re: RE: st: GLLAMM: Predict Class Membership
From
"Ariel Linden, DrPH" <[email protected]>
To
<[email protected]>
Subject
re: RE: st: GLLAMM: Predict Class Membership
Date
Thu, 20 Sep 2012 11:21:14 -0400
Martin,
Have you considered using a finite mixture modeling approach? This creates a
latent variable and provides the posterior probabilities you require.
See the user written program -fmm- by Partha Deb (ssc install fmm), and its
post estimation commands that provide the posterior probabilities...
Ariel
Date: Wed, 19 Sep 2012 13:09:27 +0200
From: Maarten Buis <[email protected]>
Subject: Re: st: GLLAMM: Predict Class Membership
On Wed, Sep 19, 2012 at 12:34 PM, <[email protected]> wrote:
> is there any way to predict individual latent class membership
> as a dummy variable - and not in terms of probability - using
> GLLAMM?
You can always assign each individual to class on which it has the
highest probability. However, that does not fit well with the logic
behind these models: these models accept that we do not know which
person belongs to which class and are all about modeling the
probabilities instead. What might make more sense is to make multiple
datasets and in each of these datasets assign each individual member
at random to a class based on the predicted probabilities.
Hope this helps,
Maarten
- ---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany
http://www.maartenbuis.nl
- ---------------------------------
*
* 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/
------------------------------
Date: Wed, 19 Sep 2012 08:05:10 -0400
From: Cameron McIntosh <[email protected]>
Subject: RE: st: GLLAMM: Predict Class Membership
You will often see people assigning cases to classes based on posterior
probabilities (usually to run separate logistic regressions of classes on
covariates), but this removes uncertainty in class membership and thus
biases standard errors of coefficients downward. So optimally you would use
a concomitant-variable LCA, or improved multi-step approaches:
Vermunt, J.K. (2010). Latent Class Modeling with Covariates: Two Improved
Three-Step Approaches. Political Analysis, 18(4), 450-469.
Clark, S. & Muthén, B. (submitted). Relating latent class analysis results
to variables not included in the analysis.
http://www.statmodel.com/download/relatinglca.pdf
Cam
*
* 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/