Skip to content
Prev 4511 / 15075 Next

Latent class analysis issue

On 22/03/2008, at 4:18 AM, Byron Ellis wrote:

            
Agreed this would be better in r-help, but we've got this far, so may  
as well continue.

The lca function isn't particularly good. It is a bit specific about  
data, it requires a matrix, which is probably what is happening here.  
There are also some other failings, such as not allowing for the  
models to be fitted for a selection of random starting values, so is  
likely to get the wrong answer when fitting more than 2 or 3 classes.  
It also requires an observation for every subject rather than  
specifying frequencies. Yes, I should send my comments to the  
maintainer.

A better package is poLCA, but also doesn't allow frequencies to be  
specified. I would set the nrep parameter to at least 10, but 20 is  
better. Has a nice user manual.

I have a package in submission randomLCA which is mainly for LC  
incorporating random effects but fits LC as well. For the moment this  
can be downloaded at http://www.kjbeath.com.au/randomlca/ and  
installed as a Local Source Package. Comments welcome.

Ken