Ways to incorporate first order autoregression in logistic multi level models? (Difficulties doing this with melogit)

[Claire Leppold]

Dear Stata forum,

I am running a multi level model of repeated measures panel data, with level one as time (year) and level two as individuals (id), with a binary outcome in Stata 14. I have been able to do this using the melogit command and have the models running with no problem, but I have been advised that I should include a first order autoregressive term to account for the fact that the dependent variable of any individual at a given time point is strongly predicted by their dependent variable at the previous time-point.

However, I can find no way to incorporate an autoregressive term with the melogit command.

If it really is impossible to incorporate a first order autoregression when using the melogit command, could you please advise me on the other ways to include a first order autoregression in other logistic multi level modelling commands in stata? I would greatly appreciate any feedback on this.

Thanks,
Claire

[Claire Leppold]

Bumping up because of time elapsed. Any advice on this please?

[Joseph Coveney]

If it really is impossible to incorporate a first order autoregression when using the melogit command, could you please advise me on the other ways to include a first order autoregression in other logistic multi level modelling commands in stata?

help xtgee

[Stephen Jenkins]

Claire: you could also search the internet and Statalist for "dynamic random effects probit". Adding a lagged dependent variable has non-trivial consequences: you need to also account for a potential "initial conditions" problem, else you'll get biased estimates. The previous search contains references to literature on this as well.