Categorical nominal variables with missing data and FIML

[andrea]

Hello, 

I would like to know if I can use FIML to deal with missing data for nominal variables (e.g. race), binary and ordinal variables. 

If I do a SEM model using the nominal variable and the binary variable with dummy coding and the ordinal variable treated as numeric,  would be FIML a good option for the missing data? 

Or should I do Multiple imputation instead? 

Thank you very much, 

Andrea 

[Yves Rosseel]

On 07/08/2016 10:47 AM, andrea wrote:
> Hello,
> I would like to know if I can use FIML to deal with missing data for
> nominal variables (e.g. race), binary and ordinal variables.

Not in lavaan 0.5.

> If I do a SEM model using the nominal variable and the binary variable
> with dummy coding and the ordinal variable treated as numeric, would be
> FIML a good option for the missing data?

No.

> Or should I do Multiple imputation instead?

Yes.

But in principle, if the variables are exogenous, and you treat them as
'fixed', then listwise deletion would be the only option if you have
missing values on those variables.

Yves.