Message/Author 


Hi, I have previously estimated a structural equation model with latent variables using Amos (maximum likelihood estimation). I am now running the identical model (I think) with the identical dataset in Mplus (maximum likelihood estimation). The unstandardized and standardized estimates obtained in Mplus are identical to those obtained in Amos, but the standard errors (and therefore the p values) are quite different. I'm struggling to understand why the standard errors I'm obtaining from the two softwares are different. Are there computational differences between the softwares that could lead to this? Or are there settings that might be set differently between Amos and Mplus? Or does this indicate that I must have made a mistake specifying the models in one of the software packages? Thanks! 


I suspect that the estimators are not the same. You may have ML in Amos and MLR in Mplus. The parameter estimates will be the same but the standard errors will be different. 

Gwen posted on Friday, March 01, 2013  10:32 am



Thanks. I was thinking that might be the difference, too, but my Mplus syntax says: "ESTIMATOR = ML;" so it should be using maximum likelihood estimation, right? Are there any other possibilities? 


It may be that AMOS is not ML. I would need to see the two outputs and your license number at support@statmodel.com to know for sure. 


Your Mplus run uses bootstrap standard errors because you say Bootstrap = 1000; I don't see that in your Amos run, which would suggest that Amos uses regular ML SEs. If you delete your bootstrap line you should get similar SEs, although note that Mplus computes them using an observed information matrix as recommended in the literature when there is missing data, and I am not sure Amos uses that but perhaps instead the expected information version. 

Back to top 