I am considering to use a missing data design(3-form design of Graham (1996)) in my PhD-research. The design comes down to the use of 3 types of questionnaires in which there is always an overlap of ¾ of the items and ¼ of the items is present is all 3 types of questionnaires.
I was wondering to what extent Mplus is able to handle this missing data. I want to apply the missing data design on my IV's, which are dimensions of culture, measured on a 7-point likert scale. I was wondering whether Mplus uses listwize deletion for missings on incomplete independent variables ore uses full information maximum likelihood.
I read somewhere that I should make my IV's DV's. Is this only true if this IV is an observed variable?
Model: belongness BY BE1 BE2 BE3; egalit BY EG2 EG3 EG4 EG5; persaggr BY ON38 ON39 ON40; persaggr ON egalit; persaggr ON belongness;
If I have used the missing data design on my IV's BE1 BE2 BE3 EG2 EG3 EG4 EG5, does Mplus use full information maximum likelihood and not listwize deletion on the independent variables? I think Mplus uses FIML in that case but I am not sure…
If egalit and belongness are observed variables, can I make them independent by adding the following to the syntax? persaggr BY ON38 ON39; persaggr ON egalit; persaggr ON belongness; egalit; belongness;
Some additional questions. - How much missings will Mplus be able to handle, or in other words is there a limit to the number of variables I can put in my design of missingness? - Is there a requirement of a large realized sample? And how large should that sample be? - Can I apply this design on my dependent variables (which are different kinds of unethical work behaviours, where the distributions of the items are often very positively skewed)? - Are there any pitfalls I should be aware of by using this design and my analyses in Mplus? For example if I want to use WLSM?