Michael I. Lichter wrote:
>So what are you advocating instead? ML, which isn't very robust to
weirdness (technical term) in the data? Or something other than >FA/PCA?
No, I'm following the research of people like Rod McDonald (retired to
his beloved Sydney), Albert Maydeu-Olivares (Barcelona) and Bud McCallum
(UNC-Chapel Hill), who support the use of the Unweighted Least Squares
criterion function, or multi-stage estimation methods that uses ULS
followed by a few iterations of ML or GLS. ULS is mathematically
equivalent to MINRES. Stata doesn't have ULS/MINRES (alas).
One of the downsides to ULS is the fact that it doesn't generate
standard errors and GoF statistics, but Albert's been rectifying that
situation in a sequence of articles in various journals (Psychometrika,
JASA, etc.): http://www.ub.edu/gdne/psicometriaen.html.
*
* 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/
