Longitudinal missing by design PreviousNext
Mplus Discussion > Missing Data Modeling >
Message/Author
 Lucy posted on Thursday, March 21, 2013 - 4:41 pm
Hi,
I have a data set on which I want to conduct an LCA on 5 indicators, at each of 6 time points. My problem is that at 3 of the time points, some of the indicators were ommitted. Whilst this missing is likely to be MCAR, as it is missing by design, I need to impute values for these indicators so that I can complete the LCA on the same set of indicators at each time point, so that the I can constrain the meaning of the classes to be constant over time.

Based on my reading of the pdf on Multiple Imputation in MPLUS, Version 2, 2010, and error messages, I have realised that I cannot impute values for the variables with completely missing data (due to missing by design), due to identification issues (there must be more observations for the imputed variable than variables being imputed).

Is there a way that I can impute the missing by design variables in MPLUS?

Regards,
Lucy
 Linda K. Muthen posted on Friday, March 22, 2013 - 9:12 am
I would simply code the values as missing and use FIML.
 Lucy posted on Friday, March 22, 2013 - 4:24 pm
Hi Linda,
Thankyou. Unfortunately this solution won't be adequate for me. Firstly, the missing variables are dichotomous, and I don't feel comfortable with the assumed normality of FIML. Secondly, even with FIML the LCA won't run for the variables with completely missing data (due to variable being omitted from survey).

Is there a way to impute the variables missing omitted from the survey, perhaps similar to the longitudinal imputation offered in PAN
http://www.google.com.au/url?sa=t&rct=j&q=&esrc=s&frm=1&source=web&cd=10&sqi=2&ved=0CIEBEBYwCQ&url=http%3A%2F%2Fcran.r-project.org%2Fweb%2Fpackages%2Fpan%2Fpan.pdf&ei=UudMUb2FDIuPkwWo54DIBw&usg=AFQjCNE8fFazzBmEV33uyOusxI2NPGlUyw&sig2=1Xs4xkdK-cOGOrrHvsKMSw

in MPLUS? I wish to use MPLUS so that I can incorportate a latent class model into the imputation model also.

Regards,
Lucy
 Linda K. Muthen posted on Friday, March 22, 2013 - 5:08 pm
FIML can be used with categorical variables for which normality is not assumed.

You cannot impute values for a variable that has missing values for everyone in the sample.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: