
Message/Author 


I am fitting a True IntraIndividual Change (TIC) SEM (Steyer, Eid and Schwenkmezger 1997) in both AMOS and Mplus and am getting rather different results, which makes me think I may be doing something wrong. The basic idea of the model is to specify change in the same construct over time as a latent variable; this can then be modeled as both exogenous and endogenous to other events. I have six attitude items at each of three waves of measurement. I have specified the model in Mplus as: ANALYSIS: TYPE IS GENERAL; ESTIMATOR IS WLSMV; ITERATIONS = 1000; CONVERGENCE = 0.00005; att91 BY aopfama (1) aopfamb (2) afamc (3) afamd (4) afame (5) aopfamf (6); TIC93 BY aopfama (1) aopfamb (2) afamc (3) afamd (4) afame (5) aopfamf (6) copfama (1) copfamb (2) cfamc (3) cfamd (4) cfame (5) copfamf (6); TIC95 BY aopfama (1) aopfamb (2) afamc (3) afamd (4) afame (5) aopfamf (6) copfama (1) copfamb (2) cfamc (3) cfamd (4) cfame (5) copfamf (6) eopfama (1) eopfamb (2) efamc (3) efamd (4) efame (5) eopfamf (6); the model converges okay and the estimates seem sensible but they are, in places, quite different from those I obtain when fitting the same model in Amos. In particular the correlation between att91 and TIC93 is considerably smaller in Mplus. Could you suggest any reasons for this? Thank you, Patrick Sturgis 


I don't see your full input but I see that you are using the WLSMV estimator which means that you are treating your observed variables as categorical not continuous. AMOS does not have an estimation method for categorical outcomes as far as I know. If you treat the variables as continuous and use ML and you have the same model, you should get the same results. 

Anonymous posted on Monday, August 08, 2005  1:29 pm



I started an SEM model in AMOS 4.01, but moved to MPlus 3.12 to take advantage of some additional features. Before proceeding to the more complicated analyses, I wanted to make sure my models were set up identically and am running into a puzzling situation. As far as I can tell, the models are the same(i.e., same df, same parameters estimated), but my fit indices and some parameter estimates do not match. Many parameter estimates are identical; others are close, but not exactly the same. I originally thought that this might be due to estimation with missing data, but when I use the subsample with complete data I find a similar problem. The model is TYPE = MEANSTRUCTURE MISSSING H1. Any ideas on why I might be getting discrepant results? Thanks! 

bmuthen posted on Monday, August 08, 2005  2:30 pm



Please send the outputs from the two programs to support@statmodel.com. If you can, please also send the data. 

Back to top 

