Dear Statalisters,

I am analysing longitudinal data in long format from a complex survey design.
Those in charge the data distribution suggest to handle them in this way:

svyset psu [pweight=weightres4], strata ( strata ) singleunit(centered) (1)

Weights are time-varying so, they vary for individuals and waves.

So far, I ran a model of this type
svy: logit y x

but I wish I could use a random effect model to control for individual unobserved heterogeneity.
xtlogit is not supported by svy.

So I was wondering whether melogit could be used and in that case
how to set model such that it considers the complex design in (1).

In the end I think that the hierarchy of the levels should be:

_n (t) <-- ID <--strata<-- psu
.
and the model this one:

melogit y x [pweight=weightres4] ||pidnew: || strata: || psu:, or level(95)

But I would like to have an opinion on whether this specification is correct,
as computing this model and the margins from this model is quite long
and I am not sure whether this specification is correct.

Thank you and best,
Lydia