The marginal effect after performing fmlogit (fractional multinomial logit)

[Jimmy Yuan]

hi

I am using Stata 14 for performing fmlogit. After running the code of fmlogit and margins, I gained the result of converse sign for same variable and also different significant power.
I was really confused and didn't know how to interpret the result. Did I do the wrong steps?
Here are the code I used and the result

Code:
fmlogit per_in_cc per_in_take per_in_ap ,etavar(int_ap_pca q75 q51_1 q53_1 q54_1 q56_1 q59_1 q60_1 q61_1 q64_2 q71_1)
margins, dydx(*) predict(outcome(per_in_ap))

Result:
variable = int_ap_pca
coefficient of per_in_ap / per_in_cc = - 0.0232
dy/dx = 0.0346847***

How do I interpret the result.?

Thank you,

Jimmy

[Hong Il Yoo]

In (fractional) multinomial logit models, marginal effects and coefficients can take different signs. For a fuller explanation, you can read any textbook that covers these models.

[Bilal Caliskan]

Hi,
Please could you some explain us how can we get margins after using factional multinomial logit model (with fmlogit commad)? In explanation part written by Maarten Buis, dfmlogit command is suggested but it doesn't work new versions of stata. So, what we (new version of Stata users) to get average marginal effects? Thanks in advance.

[Richard Williams]

Bilal, see the response I gave you in this thread:

https://www.statalist.org/forums/for...ogit-variables

If possible, try to keep a conversation in one place, since otherwise there can be unnecessary duplication of effort. Someone wouldn't want to spend several minutes trying to answer you here only to find that someone else had provided the same answer (or a better answer) in a thread that was going on simultaneously.