Constant term omitted without specifying "nocons", help

[Emil Krabbe]

I ran a logit regression and the constant term was missing. How can this be?

xtlogit vigact3 years years2 marpart logass RS RL i.time, fe

this produces an output with no constant.. and I want a constant!

[William Lisowski]

The fixed effects model specifies that each panel has its own "fixed effect" that is the same for each observation in the panel - that is, the fixed effect is a panel-specific constant.

Thus there can be no overall "constant" term for the model, because it would be collinear with the collection of panel fixed effects. You could pick an arbitrary constant - for example, 42 - and then subtract 42 from each of the fixed effects and the results would be unchanged.