margins after logit

[Dario Maimone Ansaldo Patti]

Dear All,

I think this point was addessed in the forum but I was not able to find anything. Suppose I have the following:

webuse margex, clear
generate female=(sex==1)
generate agefem=age*female

logit outcome c.age##i.female, nolog
margins, dydx(age) at(female=(0 1)) post

logit outcome age female agefem
margins, dydx(age) at(female=(0 1)) post

Why the estimation of marginal effect changes slightly if i use factor notation instead of constructing the interaction term manually?

Thanks for your help.

Best

Dario

[Fei Wang]

In the first setting, Stata calculates the margin based on coefficients of age itself and age*female. In the second case, Stata only looks at the coefficient of age because it treats agefem just as a different variable even though we know it relates to age.

[Richard Williams]

For a more detailed explanation, see

https://www3.nd.edu/~rwilliam/stats/Margins01.pdf

[Dario Maimone Ansaldo Patti]

Richard Williams and Fei Wang Thanks for your explanation. I read with a gread interest Richard's presentation. Very useful!