EFried posted on Friday, August 24, 2012 - 3:57 pm
I have two z-variables that are highly correlated in my dataset, which I guess leads to the same error message as mentioned above:
*** FATAL ERROR PROBLEMS OCCURRED DURING THE DATA IMPUTATION. THE PSI MATRIX IS NOT POSITIVE DEFINITE. THE PROBLEM OCCURRED IN CHAIN 1.
As far as I understand the UG and technote, all variables mentioned in the names list that are not being imputed are still automatically used as z-variables for imputation purposes.
Do I have to drop the problematic variable z4 from my dataset and impute a different dataset without that variable, or is there a way of telling MPLUS to only use variables z1 z2 z3 for imputation (or by telling MPLUS to exclude z4)?
If there is no USEVARIABLES list, all of the variables on the NAMES list are used for imputation including the variables for which data are being imputed. You can exclude z4 by using the USEVARIABLES option.
EFried posted on Friday, August 24, 2012 - 4:29 pm
I'm running into a similar problem as mentioned above.
My dataset contains highly correlated indicators each with 5 categories (data is not normally distributed). When I attempt multiple imputation treating these indicators as categorical, I get the following error message:
"***FATAL ERROR. PROBLEMS OCCURRED DURING THE DATA IMPUTATION. THE PSI MATRIX IS NOT POSITIVE DEFINITE"
When I attempt multiple imputation treating these indicators as continuous it runs without problems. However, these indicators are not continuous and treating them as such results in poor model fit in my measurement model.
I'm wondering if you know of a way to resolve the NPD issue so I can run imputation with categorical data or if there is an appropriate way to treat the imputed datasets with continuous values as categorical for my analyses.