Marginal effects after ologit with original and log-transformed variable

[Christian Zscherpe]

Dear Statlists,

I ran an ordered logit regression using Stata17 on Windows 10, where the model contains both the original variable (milk_quantity) and a log-transformation of the corresponding variable (ln_milk_quantity).

This is my model:

 ologit count_adopt i.gender i.under35##i.Successor c.milk_quantity c.ln_milk_quantity, vce(robust)

Now I want to calculate the marginal effects of the variable milk_quantity. For this, milk_ quantity and ln_milk quantity must change together.
In the book by Long & Fresse 2014: Regression Models for Categorical Dependent Variables Using Stata, there is a suggestion on page 377, on how to compute the marginal effects in this case:
"The only solution is to compute the appropriate predictions, specifying both the values of age and age-squared at two values of age and then subtracting the predictions."

I tried to calculate this on the given example but failed.

Is there anyone who knows how to code this in Stata?
Any help is appreciated.


Thanks Christian

[FernandoRios]

you could use f_able (install from ssc)

webuse fullauto
ssc install f_able
fgen logmpg=log(mpg)
ologit rep77 foreign length mpg logmpg
** you indicate which variable was created here logmpg, and add the option "auto"
f_able logmpg, auto
** obtain margins as usual
margins, dydx(mpg)

[Andrew Musau]

Or compute the derivative and use the -expression()- option of margins.

[Christian Zscherpe]

Thank you, Fernando! That works great.