----- Original Message -----
From: "Arthur J. Kendall" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Wednesday, July 02, 2003 1:34 PM
Subject: Re: Factor Analysis
###
>>
> Hope this helps.
Well, not really! At least, some more definite advice could be given.
Though it's really more a matter of degrees of freedom (roughly, cases minus
variables) a sample of forty would never be enough. As a simple parallel,
consider how the standard error of (say) a regression slope depends on
sample size. FA models that include errors for the coefficients (see
Joreskog's work) make this clear; older methods like those churned out by
SPSS don't.
You can't get out more tham you put in!
>
> Art
> [log in to unmask]
> Social Research Consultants
> University Park, MD USA
> (301) 864-5570
>
>
> Neville Calleja wrote:
> > Dear colleagues
> >
> > I just had PhD student in my office to consult me about problems she is
having
> > using factor analysis. She is doing a thesis on the behaviour of
sufferers
> > of genetic diseases.
> >
> > While this is not exactly my field, she has filled me in with literature
on
> > the subject. Apparently, (since I have to admit that, being a recent
graduand,
> > factor analysis is not my forte) factor analysis is not recommended for
samples
> > of less than 50, according to the articles I have seen. Her sample only
involves
> > 40 patients, which is basically the maximum it can reach in a small
country
> > like Malta.
> >
> > Would any of you have different opinions about this? If this statement
is to
> > be adhered to, are there any ways to go around it - maybe non-parametric
methods?
> >
> >
> > Moreover, in another group of patients with a different type of disease
she
> > has another 80 individuals. She was recommended to join the two and
just put
> > in a nominal variable to identify which group they belong to. However,
the
> > factor analyses conducted on the two groups separately are resulting in
very
> > different profiles. I don't think you can just join them into one
analysis.
> > What are your opinions about this?
> >
> > The ultimate target is to generate multiple regression models from the
data
> > collected about these individuals.
> >
> > Thanks once more
> >
> > Yours truly
> >
> > Dr Neville Calleja
> >
> > ====================
> > Dr Neville Calleja
> > 328A St Paul Rd
> > Xemxija
> > St Paul's Bay SPB10
> > MALTA
> > Tel: +356-21-582905
> > Fax: +356-23-331027
> > email: [log in to unmask]
> > Mob: +356-99800088
> >
|