Jay:
If I understand you correctly, you are making two kinds of suggestion
here.
1. StataCorp should investigate why you and your students sometimes get
puzzling results with Stata's factor analysis routines.
This is not a soluble problem without details. Next time you get such
results send to StataCorp copies of a log with results, the dataset, and
an explanation of what puzzles you. StataCorp want very very much to fix
any and all bugs, but a report like yours strikes me as hopelessly
vague.
Also, if the issue is that other software gives different results for
what is apparently the same problem, then clearly that in turn needs
documentation. (In territory like factor analysis, some differences on
doctrinal grounds are often not surprising.)
2. StataCorp should consider work on EFA [whatever that is; possibly
exploratory factor analysis] by Michael Browne etc. etc.
This would have a much higher chance of getting high on StataCorp's
to do list if you spelled out exactly what you would like, precisely
where it is explained and why its omission is a notable gap for at least
several Stata users. Again, contacting StataCorp direct is the best way
to proceed.
StataCorp may already have contacted you privately with a message to
similar effect for all I know. I make this comment in public because you
made your complaint in public, and I want to underline, from my direct
and indirect experience, that certain kinds of suggestions have
immensely higher chance of getting acted upon than others.
Nick
[email protected]
Verkuilen, Jay
Kelly,
I don't have anything to add specifically to your question but have
noticed some odd behavior in the factor analysis program, in particular
the factor analysis of a correlation matrix, though occasionally the
regular factor analysis program does strange stuff too. So you may be
bumping in to another issue.
This is annoying because I want to have students in my class use these
programs and inconsistent behavior of programs REALLY BUGS students.
If Statacorp people are listening, there are some suggestions I'd like
to offer that would make on top of "fix the bugs". Honestly, they're
really "take a look at the recommended practices discussed by the
foremost expert on EFA, Michael Browne" but I can list them out. Stata's
rotation commands are already quite good but the fact that standard
errors for loadings and other simple but important quantities aren't
listed. Also, I can't tell for sure (but believe) that the EFA program
analyzes correlation matrices, not covariances. Providing the option for
covariances would be useful.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
