st: RE: Multiple correspondence analysis & PCA with ordinal variables

["Nick Cox" <n.j.cox@durham.ac.uk>]

Stata has no notion of ordinal variables for this purpose. 

My intuition is this: It shouldn't matter any way what you do or do not
tell Stata about how each variable should behave. If it behaves like an
ordinal variable, that will come out in the results. If not, not. Either
way, you should learn something. 

In any case, why not just try the PCA as a kind of qualitative check? 

More generally, I am in favour of forcing (nicer word: guiding) analysis
when something absolutely must be true, e.g. that proportions belong in
[0,1]. But no low, medium, high grading I ever came across justified
total conviction in its infallibility. 

Nick 
n.j.cox@durham.ac.uk 

Owen Corrigan

I am trying to extract a single index from a set of 7/8 ordinal
variables measured over three categories (low, med, high). I
understand that MCA (multiple correspondence analysis) is an analogue
of PCA (principal components analysis) for categorical/ordinal data,
and so have been using that in Stata 10. My concern is that I am not
properly 'telling' Stata that the data is ordinal, and thus Stata may
not be taking into account that category 3 > category 2 > category 1.
I have read through the help file and options, but see no way to
'force ordinality' on the program; so I am worried that the analysis
is merely treating each of the categories of the variables as
independent, and not related.

Is there a way to force ordinality? Should I be using a different
method?
All assistance greatly appreciated.

*
*   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/