Rob White posted on Monday, September 24, 2007 - 10:46 am
I generated X multiply imputed datasets from outside Mplus. That is, the missingness has already been taken care of outside of Mplus. I want to estimate my structural equation models on these X datasets and use Mplus to make the necessary corrections to the coefficient estimates and standard errors. How do I do this? Thanks.
Rob White posted on Tuesday, September 25, 2007 - 12:51 pm
Thanks very much. One question: why do my MPlus standard errors differ from the standard errors I get from using the ice commands in Stata?
Mplus Stata est. s.e. est. s.e. -1.05 .025 -.1046 .0264 0.837 .089 .8370 .0945
Stata documentation indicates that it uses the same Rubin rules for correcting the errors as referenced in Schafer (1997). I've confirmed that the Stata calculation is correct with hand calculations. Is there some subtle difference I'm not understanding? Since I'm using Mplus to estimate only the model, rather than generate the MI data and estimate the model, the datasets are identical across both programs.
Rob White posted on Tuesday, September 25, 2007 - 12:53 pm
Please note that there is a decimal error in the Stata estimate. -.1046 should actually read -1.046
Rob White posted on Tuesday, September 25, 2007 - 1:09 pm
Also, are the corrected estimates and errors available in the output coming from SAVEDATA, RESULTS ARE ? I cannot find them in there. Thanks.