home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!darwin.sura.net!paladin.american.edu!auvm!UTAHCCA.BITNET!LABENISH
- X-Envelope-to: stat-l@MCGILL1.BITNET
- X-VMS-To: IN%"stat-l@mcgill1.bitnet"
- Message-ID: <6D4F861E70008B4E@CC.UTAH.EDU>
- Newsgroups: bit.listserv.stat-l
- Date: Thu, 10 Sep 1992 08:41:00 MST
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: LABENISH@UTAHCCA.BITNET
- Subject: LISREL 7 help
- Lines: 76
-
- I need some help running a structural equation model using
- LISREL 7. I have a sample size of about 170, 2 indicators for
- each of two exogenous latent variables and 2 indicators for one
- of my 3 endogenous latent variables; I have one indicator for
- the other 2 latent endogenous variables.
-
- Prior to embarking on the structural model, I completed a
- confirmatory factor analysis on the measurement piece of the
- model just to make sure that the measures were loading fine on
- the hypothesized constructs. This was confirmed.
-
- The commands, freed and fixed elements, etc. for my basic
- structural eq. model are:
-
- / DA NI=8 MA=CM
- / MO NY=4 NX=4 NE=3 NK=2 BE=FU PS=DI
-
- / FR BE(2,1) BE(3,2)
- / FR LY(1,1)
- / FR LX(1,1) LX(4,2)
-
- / VA 1 LY(2,1) LY(3,2) LY(4,3)
- / VA 1 LX(2,1) LX(3,2)
-
- / FI GA(2,2) GA(3,1) GA(3,2)
-
-
- The error statement I get when I ran this model is that my PHI
- and PSI matrices are not positivie definite.
-
- With regard to my output (for what it may or may not be worth),
- LX -- looked reasonable
- LY -- one value close to 1.0
- BE -- VERY small values (virtually 0.0)
- GA -- small negative values
- PS -- VERY small values (virtually 0.0)
- TE -- Values ranging from 1.6 to 27.0
- TD -- Two VERY small values (.04) and two VERY large
- values (130.5)
-
- * Each of the values of the squared multiple correlation for
- the y variables was virtually 0. Thus, the coefficient of
- determination was also 0.
-
- * Regarding the sq. mult. corr. for each of the x variables,
- 2 of the 4 values were virtually 0 and the other two were
- about .85; the coeff. of deter. was .89
-
- * The values of the squared multiple correlations for the
- structural equations were -.16; 1.01; -.22
-
- A few weeks ago a woman from Drake was encountering some problems
- with one of her error matrices not being positive definite. I
- re-ran my model in accordance to the recommendations she received
-
- * Using EQ statements in error matrices [since the model
- is underidentified (i.e., gamma matrix)]
- * Identifying parameter estimates greater than 1.0 and
- using ST values for those estimates
- * Requesting Ordinary Least Squares and Generalized Least
- Squares solutions
- * Playing around with setting negative values in my
- error matrices at a value greater than 0.
-
- At best, I got rid of the non-positive definite PH matrix, but PS
- continued to be problematic.
-
- I would VERY MUCH appreciate any 'pearls of wisdom' aboutit's
- just difficult to express that over electronic mail)
-
- orbenish@cc.utah.edu
- la
- labenish@utahcca.bitnet
-
- (801) 626-6406 -- Work
- (801) 531-1823 -- Home
-