Dear all,

I am using xtmelogit to run a multilevel logistic regression with PISA data. My data is hierarchical (individuals nested into schools, schools nested into countries).

My dependent variable is the expectation of college graduation among the fifteen-years old students who were interviewed for PISA. I am regressing my dichotomous dependent variable (expecting college graduation or not) to a number of controls and two key independent variables: gender of the respondent and father's education.

There is the case of which one of the two parents is more important for the phenomenon that I am trying to explain. In other words, if the model with father's education has a better goodness of fit than the model with mother's education or vice versa. Here there are the two models:

Model with father's education (fisced4)
Code:
xtmelogit expect_ISCED5A female ib4.fisced4 || country3: || schoolid:, variance
Model with mother's education (misced4)
Code:
xtmelogit expect_ISCED5A female ib4.misced4 || country3: || schoolid:, variance
Since they are not nested models, I suspect that I should use BIC o AIC in order to compare the goodness of fit of the two models, but I am not sure.

Could I ask you for some guidance regarding to the best way of making this model comparison and, if this is the case, how to get the AIC reported after xtmelogit?

Many thanks for your attention and your help

Kind regards

Luis Ortiz