"Rich Ulrich" <[EMAIL PROTECTED]> wrote in message
[EMAIL PROTECTED]">news:[EMAIL PROTECTED]...
> Ah... I notice, orthogonal factors can do a serious job of
> "reflecting the reality" of factors that are moderately correlated.
> If two axes are 70-degrees instead of 90-degrees, there's
> a massive projection onto each of the nearest of a pair
> of 90-degree perpendiculars, since each is only 10-degrees
> away.
...
>What is the power for disentangling the confounding among
>variables, for correlated factors? This is just my impression,
>but I think you might need a sample of a few thousands,
>to justify very much. <I will try to check a few references,
>shortly.>
I have a preference for methods that test hypotheses, so that I would be
happy performing a factor analysis to confirm (or disconfirm) the idea that
certain items went in Group A, certain items went in Group B, but I find it
less preferable to use a factor analysis in order to construct item groups
on a questionnaire/survey/measure from scratch. - Along the lines of gee,
how many factors should I specify, guess I'll use varimax rotation because
it's the default in the software...etc.
Laterally, what about doing a Rasch analysis instead? I've seen that type
of analysis performed (on the FIM - Functional Independence Measure?) to
perform essentially a factor analysis to group items - it managed to
successfully separate "cognitive" items from "physical" items on the test
(which measures, using pretty much rank categories, the ability of a person
to perform certain daily activities such as walking).
cheers
Michelle
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
