Hello Statausers,

I have a two-level logit model where I have used group centered variables along with group means. I have been advised to use precision weighted estimates instead of group means in my multilevel model. For this, I need to follow these directions
HTML Code:
"Calculate a separate null two level model with individual wealth as the outcome and then get precision weighted estimates of cluster wealth from this model . This estimate - the level 2 residual can then be used in your original model."
This was easy for binary variable "x". I used the following code:
Code:
svy:melogit x||psu: , 
    predict gm1, reffects
Note that though my model uses the ‘subpop’ command I I have not used it above because I want these estimates from the whole population. I hope that is right?

But I am not sure how to do this for categorical variables. I have multiple such variables such as education (4 categories) and household wealth (5 categories). Even for continuous variables, "mixed" does not allow using survey weights. So how do I go about that?

I will appreciate your advice.

Thank you
Deepali