gibbon lab posted on Thursday, October 27, 2011 - 10:30 am
I am running two SEMs in Mplus 5.21, where the only difference is the endpoint. Both endpoints came from the same variable. One is the original continuous variable and the other one is a categorical version of the continuous variable. But these two SEMs give me different observations (used in the model). It seems like the SEM with continuous endpoint only excludes the subjects who miss all VARIABLES and the SEM with a declared categorical endpoint excludes subjects who miss all X VARIABLES (exogenous variables). Is this true in Mplus? Any option to make them use the same observations? Thanks.
Situation: asn, bsn, csn and dsn are latent variables, bs0071 is a binary observed variable. asn, bs0071 are exogenous. bsn, csn and dsn are endogenous. If the correlation between asn and bs0071 is fixed to zero, the number of observations doesn't match the nobs in a model with free estimation of the correlation. It is larger for the model with the fixed correlation. I was expecting nobs to be the same in both models, since cases with missing values on x-variables cannot be included in neither model.