Date:         Wed, 10 Jul 1996 18:05:52 GMT
Reply-To:     Walley <walley@SOI.HYPERCHAT.COM>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From:         Walley <walley@SOI.HYPERCHAT.COM>
Organization: State of Insanity
Subject:      ULS standard errors in LISREL 7

QUESTION:
 
I have a structural model that fits the data well, but only if ULS is used as
the estimation procedure (I have not used DWLS or WLS due to small sample size
 of n = 191).  How accurate are the standard errors for parameter estimates
generated by LISREL 7?  My observed variables are highly skewed, precluding
the use of ML estimates; in addition, due to strong autocorrelated errors, ML
fails to iterate.
 
The ULS estimates are theoretically sound and replicate across two independent
data sets.  Does anyone know of examples of using ULS estimates that are
published?  Are the standard errors biased toward overestimation and thus
yielding conservative t-statistics?
 
Thanks.
 
William Panak
Dept. of Psychology
Iowa State University
 
wpanak@iastate.edu