home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!stanford.edu!bu.edu!olivea!netsys!ukma!darwin.sura.net!paladin.american.edu!auvm!MIZZOU1.BITNET!PSYWOOD
- From: PSYWOOD@MIZZOU1.BITNET (Phil Wood)
- Newsgroups: bit.listserv.sas-l
- Subject: Proc Calis Independence Models under GLS
- Message-ID: <SAS-L%92110516370647@UGA.CC.UGA.EDU>
- Date: 5 Nov 92 21:31:38 GMT
- Sender: "SAS(r) Discussion" <SAS-L@UGA.BITNET>
- Reply-To: Phil Wood <PSYWOOD@MIZZOU1.BITNET>
- Lines: 23
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
-
- This semester I have the pleasure of teaching a course in
- structural equations. In a recent lecture I explained to my class
- the default null model used by Calis for computing Bentler-Bonnett
- Normed and Non-Normed Fit measures. I explained that this is the
- Chi Square fit which would result in you would input a model with
- each variable loading on one factor, factor loadings fixed to one,
- and factor variances free. (Or, alternatively, each variable
- composed of only an error term, with error variances free- makes no
- difference).
- Now, as it happens, a good student decided to validate what I
- claimed in class. He found that, for ML estimation, he was able
- to get the null model chi square reported in calis. However, for
- GLS estimation, he can't reproduce the chi square. Am I missing
- something here?
- What happens is that Calis generates default start values and then
- goes through one interation (whereas ML correctly estimates
- the std's on the first pass and then stops).
- Am I missing something here? I would have thought this solvable
- by inspection for both the ML and GLS cases-
- Any advice/info appreciated!
- Phil Wood
- University of Missouri-Columbia
- SAS under Unix version 6.0.7
-