Date: Thu, 14 May 1998 14:46:39 0500
ReplyTo: Max Martin <mmartin@EDGEWOODSA.K12.TX.US>
Sender: "SPSSX(r) Discussion" <SPSSXL@UGA.CC.UGA.EDU>
From: Max Martin <mmartin@EDGEWOODSA.K12.TX.US>
Subject: Re: SMCs
ContentType: text/plain; charset="iso88591"
If I remember my grad training, Principal Axis Factoring uses the SMCs of
the variables as initial estimates of communalities along the principal
diagonal of the R matrix before factor extraction. These estimates are then
changed at each iteration until the final communalities are calculated upon
termination of the factor routine. Hence, there would be no reason to expect
the SMCs to be identical with the final communalities.
Hope this helps
Max
%&%&%&%&%&%&%&%&%&%&%&%&%&%&%&%&
Dr. Max R. Martin
Senior Evaluator
Department of Research, Evaluation,
Info Systems & Technology
Edgewood ISD, San Antonio, TX
2104319120 FAX: 2104335134
mmartin@edgewoodsa.k12.tx.us
%&%&%&%&%&%&%&%&%&%&%&%&%&%&%&%&
Experience is a wonderful teacher
it allows us to recognize a mistake
after we've made it again!
%&%&%&%&%&%&%&%&%&%&%&%&%&%&%&%&
Original Message
From: John A. Schinka, Ph.D. <jschinka@COM1.MED.USF.EDU>
Newsgroups: bit.listserv.spssxl
To: SPSSXL@UGA.CC.UGA.EDU <SPSSXL@UGA.CC.UGA.EDU>
Date: Thursday, May 14, 1998 8:13 AM
Subject: SMCs
> I am examining the characteristics of a personality scale
>consisting of 33 items. The Reliability routine does not produce item
>SMCs (squared multiple correlations) because the determinant of the
>covariance matrix is essentially zero (this information provided via a
>warning). Yet, the Factor routine, employing principal axis factoring +
>promax, produces values for the communalities for each item. I believe
>these communalities are the item SMCs. The calculated determinant for
>this analysis is essentially zero. Can someone explain the
>inconsistency here? Am I correct in assuming that the communalities are
>the item SMCs? It seems unlikely, because the values are all <.70 and
>I'm assuming the determinant problem is caused by multicollinearity.
>Clarification?
>
> Is there an easy way to determine the SMC for each item in a
>matrix? As I'm examining about 100 scales of length from 2070 items,
>using multiple linear regression procedures are out of the question from
>a practical standpoint. Appreciate your comments!
