mfx, predict(xbsel)

[KIM SUJIN]

dear all
may I use mfx, predict(xbsel) command after heckman to get marginal effect?
thank you

[Maarten Buis]

mfx is pretty old and superseded by margins. After heckman you can predict many different types of outcomes: e.g. the outcome including the non-selected case, the outcome conditional on being selected, the probability of being selected, and more. Consequently, you can also have many different marginal effects, one for each type of outcome. The type of outcome you used in your command is highly unusual, so probably not what you want. However that obviously depends on what it is exactly that you want with that model, which you did not tell us.

[KIM SUJIN]

I hope to get selection model and outcome model marginal effect separately!!





I used
mfx compute, predict(ycond)
and
mfx compute, predict(psel)


but mfx compute, predict(ycond) doesnt show standard error!
thank you

[Maarten Buis]

I may not have been clear: Use margins instead of mfx

[KIM SUJIN]

what is the difference~?

[Maarten Buis]

There are lots of subtle differences, so I could write a long post on that. However, it would cost me a lot of time to write and you a lot of time to sift through all that information to find the information you are looking for. So instead, why don't you tell me why you think mfx is better, so I can give you an answer that is relevant to you.

[KIM SUJIN]

i tried it

. margins,dydx(*) predict(ycond)
predict option ycond not appropriate with margins

but i got error~~

[Maarten Buis]

It works for me:

webuse womenwk, clear
heckman wage educ age, select(married children educ age)
margins, dydx(*) predict(ycond)

If I cannot reproduce the problem, it becomes very hard for me to diagnose the problem. Maybe you should directly discuss this with Stata's techsupport.

[KIM SUJIN]

Dear Joao,

I wonder why I cant get constant term when i use 'xtpoisson' command!

thank you