This is a bit late, but:
> About two years ago there was a thread about this which suggested that at
> that time nobody had these coefficients ready to go.
> (a) has anyone in the meanwhile programmed them?
http://www.qimr.edu.au/davidD/R/polyr.R
> (d) I appreciate this last item is not strictly an R question, but my
> intention is to use these as input into the sem package for structural
> equation models. If anyone thinks that is misguided I would be intersted to
> here.
>
> Michael Dewey
> m.dewey at iop.kcl.ac.uk
As you might know, there are various ways you can fit what in the
genetic literature is called a multifactorial threshold model (MFT ie
tetra/polychoric correlation model). The simplest is to treat them as
Pearson correlations and use conventional SEM methods, which often works
well (esp for large dimensional problems); the AWLS method of Browne as
implemented for example in LISREL (needs large N); or full ML fitting to the
multidimensional contingency tables, which is available in programs like
Mx, or could be fitted pretty easily using mvtnorm as you suggested (Mx
uses Genz's algorithms).
You should also check the adequacy of fit of the MFT to your data, and
look at the related loglinear models.
| David Duffy (MBBS PhD) ,-_|\
| email: davidD at qimr.edu.au ph: INT+61+7+3362-0217 fax: -0101 / *
| Epidemiology Unit, Queensland Institute of Medical Research \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia GPG 4D0B994A v
