Re: st: exploratory factor analysis with dichotomous and continuous data

["JVerkuilen (Gmail)" <jvverkuilen@gmail.com>]

On Thu, Nov 22, 2012 at 3:27 PM, Frauke Rudolf <FRAUKE.RUDOLF@ki.au.dk> wrote:
> The idea is to do an exploratory data-analysis (followed by a CFA) on 11 clinical variables, some dichotomous and some continuous, to reduce the amount of variables in the score. The only variable having this "structural zero problem" is the one described here. Do I really have to change the plan, or is there another way? Do have to write the article for my PhD and appreciate every help I can get here, since I am quite new to the subject (as you probably already guessed...)>

It turns out a colleague and I have done some research work on IRT and
by extension factor analytic models of data when there is functional
dependence between items.

      Liu, Y. & Verkuilen, J. (in press). Item response modeling of
presence-severity data. Applied Psychological Measurement.

However, the problem that we treated isn't directly analogous so I
can't say that you should be able to adapt what we did in a
straightforward way. I suspect Ying and I would be willing to take a
look at it (with appropriate credit of course) as dealing with the
effect of functional dependence in psychometric models is a research
area I'm active in, and we've had a devil of a time getting datasets,
but of course you may not be able to share it.

The best answer I have at the moment is exactly what Nick already
said: You have to exclude it.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/