Date:         Tue, 1 May 2001 16:32:38 -0400
Reply-To:     "Edgar F. Johns" <efjmoj@MEDIAONE.NET>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "Edgar F. Johns" <efjmoj@MEDIAONE.NET>
Subject:      Re: Factor Analysis
Comments: To: "Parise, Carol A." <PariseC@SUTTERHEALTH.ORG>
In-Reply-To:  <0D17115C1823D311800100805FE68B1B02002AC6@gnsc8mx.sutterhealth.org>

Carol Parise asks "Does this mean that when you are trying to develop
subscales from a
questionnaire that FA would be more appropriate than PCA even though the
results are very similar?"

That's the way I like to think about it.

In my own experience, the two solutions yield similar results, too.
Notwithstanding that, I do believe that it's the purpose that determines the
"preferred" solution to use. So, if I don't want to make any statements
about the structure of the underlying data (e.g., when I'm doing a
regression analysis and I have too many predictor variables, I suspect
multicollinearity in my predictor variables), then I would use PCA. On the
other hand, if I want to try to define and interpret "factors" (i.e.,
constructs) underlying my variables, then I'll go with some method of FA
(principal axes, maximum likelihood) AND probably use an oblique method of
rotation (PROMAX).

Hope this helps.
Edgar
_____
Edgar F. Johns <EdgarJohns@obik.com>
Obik, LLC
2906 River Meadow Circle
Canton, MI 48188
Tel. 734.495.1292, Fax 734.495.1981
http://www.obik.com
 [snip]