Date:         Mon, 22 Oct 2001 12:17:01 -0700
Reply-To:     John Uebersax <jsuebersax@YAHOO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         John Uebersax <jsuebersax@YAHOO.COM>
Organization: http://groups.google.com/
Subject:      Re: assumptions of FACTOR
Content-Type: text/plain; charset=ISO-8859-1

schendera@NIKOCITY.DE (Christian F.G. Schendera) wrote in message news:<00a801c1598f$8f39a680$b997603e@notebook>...

> I am summarizing the basic assumptions of FACTOR,
> (1) Multivariate normal distribution of residuals (not necessary for
> ML-FA).
> (2) Interval scale.
> (3) Numbers of observation is reasonably larger than number of variables.
> (4) Variables should correlate (groupwise).

Hi Christian,

Is (1) above backwards?  I think that the common factor model itself
has no distributional assumptions.  It can be estimated with no
distributional assumptions using iterated principal factors (PRINIT)
or several other methods.

I think that Maximum Likelihood estimation of the model does make
multivariate-normal distributional assumptions (whether the
assumptions concern the data, residuals, or both, I do not know).

Usual significance tests for the number of factors also make
distributional assumptions.   (Signficance tests based on
bootstrapping, though, do note make such assumptions.)

For an excellent discussion of Factor Analysis and it's assumptions,
see:

http://www2.chass.ncsu.edu/garson/pa765/factor.htm
--------------------------------------------------------------------------------
John Uebersax, PhD             (805) 384-7688
Thousand Oaks, California      (805) 383-1726 (fax)
email: jsuebersax@yahoo.com

Agreement Stats:   http://ourworld.compuserve.com/homepages/jsuebersax/agree.htm
Latent Structure:  http://ourworld.compuserve.com/homepages/jsuebersax
Existential Psych: http://members.aol.com/spiritualpsych
--------------------------------------------------------------------------------