I run a multilevel model
xtmelogit vd vi if eurozone==1 || countryid:, ml var
and the results come out with the independent variable as statistically significant.

If I repeat the same multilevel model but adding different slopes for the same variable that was significant before
xtmelogit vd vi if eurozone==1 || countryid: vi, ml var
the v1 loses its significance because the standard deviation is very large.

Does this make sense? Should I keep the first model or the second one?

I would like to represent the margins of an interaction in the model
xtmelogit vd vi##vi2 if eurozone==1 || countryid: vi, ml var
How can i do it?

Stata only allows me to represent the fixed part when I make margins.

If not a get an error.
margins, dydx(vi2) at(v1=(5(10)20)) predict (mu fixed)
Thank you.