Tom posted on Thursday, December 29, 2005 - 1:47 pm
In comparing some models that I have run, when I include type = missing, the sign of some factor loadings from latent variables change.
Any idea what may be going on?
bmuthen posted on Thursday, December 29, 2005 - 6:17 pm
Type = Missing is the preferred option with missing data because it leads to using all available data. In contrast, if you don't say Type = Missing you get listwise deletion (note the sample size change).
Tom posted on Thursday, December 29, 2005 - 8:27 pm
I appreciate the contrast in sample size. Using type = missing, the sample size is 891 and without type = missing, the sample size is 774.
It is striking that the difference in t-scores would change rather dramatically with those cases deleted.
Would there be a difference if the indicators are categorical (vs. continuous)?
It sounds like you have very selective missingness if the results from listwise deletion and TYPE=MISSING; are so different. This can happen for catgegorical or continuous indicators.
Joseph posted on Tuesday, February 21, 2006 - 10:34 am
Dear Linda, I am runing SEM using Mplus. If TYPE=MISSING in the ANALYSIS command and STANDARDIZED in the OUTPUT command were used at the same time, the results only provided "Estimates" (no S.E., Est./S.E., Std, StdYX). Any solution? Thanks.
The default in Mplus is to use all available information. For maximum likelihood estimators, this is often referred to as FIML. See the Little and Rubin book referenced in the user's guide. For weighted least squares, it is pairwise present. Mplus also can do multiple imputation.