[Arianna Vera Vivoli]

Hello everyone,

I am joining this conversation because I have a very similar problem to Juan Gomez.

I am estimating a multilevel model and until now I was doing it with the xtmelogit command. With xtmelogit the model converges, both in its simplest version and in the version with many covariates. When I switch to using melogit however, the model no longer converges, even in its basic version with only the outcome variable and the nesting variable. I tried to follow what suggested by Clyde in #12 but the model does not converge, although log-likelihood still increases and Stata reports an error "cannot compute improvement -- discontinuos region encountered". How can I work out where the problem is?

Many thanks in advance,

[Clyde Schechter]

The command formerly known as -xtmelogit- still exists in current Stata, under the name of -meqrlogit-. The very raison d'etre of -meqrlogit- is that it can converge on some models that -melogit- cannot. It estimates the same model, but it has a different computational approach that is robust to some of the pitfals of -melogit-'s algorithm. (The converse is also true, though I think this is less frequent: -melogit- will sometimes converge on a model that -meqrlogit- cannot.)

So I suggest you just go back to using -meqrlogit- (or -xtmelogit- if you are using an older version that predates the change of name).