how to compute marginal effects for a variable with its quadratic term

[Huy Dang]

Hi stata experts,

I'm having a problem with the computation of the composite marginal effects of a variable with its quadratic term after mlogit

Let's say the dept. variable is categorical with four options (0,1,2,3)

I have Age and Age2 in the model and need to compute only 1 composite marginal effect of Age instead of two for Age and Age2 for each outcome (0,1,2,3) like the below figure.
How do I accomplish that post estimation. The command mfx2 yields useful marginal effects of both Age and Age2 for each outcome but I want to compute the composite marginal effect of Age only (just 1). How do I do that?

The below command works just fine to estimate the model
mlogit loantype age age2 gender edu farmexp logincome logfarmsize..., base (0) vce(robust)

The result after mfx2 shows

[Maarten Buis]

This is best solved by creating the square term with factor variables, and than use margins to create the marginal effects. So there is no need for any user-written/community contributed programs.

sysuse auto, clear
replace price = price / 1000
label var price "price in 1000s dollars"
replace rep78 = 3 if rep78 < 3

ologit rep78 c.price##c.price

margins, dydx(*)

Now the result could easily be misleading: by adding a square term the effect of that variable will differ from person to person, to the extend that the effect could easily change sign. To summarize that with a single number is problematic. Instead, I would present the results in a graph:

margins, at(price=(3/16))
marginsplot