mlogit - marginal effects of constant

[adhoul]

Does it make sense to have marginal effects for a constant in a multinomial logistic regression (mlogit) ?

To me it makes sense in the sense that you see how would the probability change if the constant changed, but does this make sense ? Or is there some other interpretation?

[Maarten Buis]

The constant is constant (hence the name), so it cannot change and a marginal effect won't make sense.

[adhoul]

Thanks Maarten. Does this mean I should not include a constant in an mlogit model, or simply not report the marginal effects for a constant ? (because they are calculated anyway)

[Maarten Buis]

The latter.

Exactly what code did you use that resulted in returning (incorrect) marginal effects for the constant? In the (silly) example below margins did not return any marginal effects for the constant:

sysuse nlsw88, clear
mlogit race wage
margins, dydx(*)