Hi, I started reading about missing data handling techniques in Mplus and I'm not sure to what extent are FIML and MI equivalent for my needs.
Here's a little more information on the model. It contains continuous latent variables and has indirect effects and latent interactions. I have two measurement points. My T1 (n=570) data contains missing data for 3 to 10% of observations and my T2 data (n=380), 10 to 30% missing data (many respondents just completed T1).
So here is my question: if I simply used the default FIML for handling missing data, will all the model be fitted for 570 respondents? (as if I had made MI)? And would that be the right technique?
N_2018 posted on Friday, April 06, 2018 - 10:45 am
Hello, I have several missing cases for the dependent variable in my analysis (I am conducting a linear regression with 1 DV and 4 IVs). The DV is a proportion score (ranging from 0 to 1), reflecting performance on a behavioral (computer) task. The data are missing due to technical glitches with the task. Is FIML an appropriate way to handle missing data for discrete variables? If not, are there any sources that discuss alternatives? Many thanks in advance.
I am working with two samples: N=402 (almost not missing data) and N=460 (out of 44 variables, 12 are missing data in half of the sample; they were added later during data collection, and this was not planned). I ran a multiple imputation generating 10 datasets. However, I am unable to test for measurement invariance of my model across patients and non-patients, since I cannot conduct a chi square difference test with imputed data. Can I use FIML in this case to be able to conduct measurement invariance analyses?