xtlogt, fe with too many factor levels

[Luiza Lins]

I am running xtlogit on a panel. I have some continuous independent variables x1, x2, x3, and some categorical independent variables y1, y2, y3. I'm particularly interested in estimating with y3 fixed-effects, like below:

xtset panelid year

xtlogit dep_var x1 x2 x3 i.y1 i.y2 i.y3, fe

The problem is that y3 can take ~67.000 values, so the estimation takes too long and produces a table with thousands of coefficients. For context, the data is a national survey and y3 is the interviewer's id. Is this the best approach?

[Carlo Lazzaro]

Luiza:
this is clearly unmanageable.
Knowing nothing about your data, I'd check whether it is feasible to gather -y3- levels in 3/4 macro-levels and then recode -y3- accordingly.