
Message/Author 

Anonymous posted on Monday, February 14, 2005  12:01 pm



I am testing a 5 factor measurement model in multiple groups with missing data. In preparation for multiple group measurement invariance analyses, I estimated the factor means (for each group separately) and set the intercepts of the indicators to 0. When I compare the parameter estimates to a model where the factor means were 0 and the intercepts of the indicators were freely estimated, the factor loadings and standard errors change, sometimes quite substantially. This is true whether I set the metric by setting the first loading to 1 or setting the variance of the latent variables to 1. Is this due to differences in the ML estimation procedure because the models have different degrees of freedom or is there another explanation? 


I would have to see the outputs you are comparing at support@statmodel.com to comment on this. 

JAD2 posted on Tuesday, January 24, 2006  12:34 pm



I HAVE USED THE EXAMPLE USED IN THE USER'S GUIDE FOR CFA ANALYSIS WITH CATEGORICAL INDICATORS, BUT IN THE EXAMPLE IN THE WEB YOU HAVE USED TYPE=MEANSTRUCTURE , BUT NOT IN THE EXAMPLE IN THE USER'S GUIDE, MY QUESTION IS: WHAT IS THE IMPORTANCE OF USING THIS OPTION IN CFA ? 


By specifying TYPE=MEANSTRUCTURE, means are included in the model. See Examples 5.9 and 5.10 in the Mplus User's Guide. 

Back to top 

