Interactions terms and predictive margins

[Marine Jouvin]

Hi,

I'm running a tobit regression and want to report marginal effects. The thing is that I have interactions (dummy#continuous variable), and that average marginal effect doesn't work in such case.

I was wondering which command I should use to obtain the marginal effect at the mean of the continuous variable when my dummy = 1.

So far, I've been testing the following ones:

margins i.treated, dydx(c.educ)

For which Stata reports this result:



and :

margins i.treated, at((mean) c.educ)

With the following result:



I'm confused about which option is the right one (if any)...

[Andrew Musau]

You do not show your tobit command. With left-censoring limit at 0, you'd want

margins, dydx(*) predict(ystar(0,.)) atmeans

[Andrew Musau]

If I am misreading #1 and your intention is to report "marginal effects" for the interaction terms, then note that there is no such thing as a marginal effect for an interaction term. You cannot change the interaction term holding its constituent variables constant. So either report the tobit coefficients as they are with the interaction coefficients or report average marginal effects using #2 that exclude the interaction terms.

[Marine Jouvin]

Thanks Andrew!

You do not show your Tobit command. With left-censoring limit at 0, you'd want

I use various of them! But do you mean that if my limit is not 0, the margins command is slightly different? For instance, I'm using the following Tobit regression:

                
tobit date_creation $chararctistics, vce(cluster section) ul(23080)

[George Ford]

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

[Andrew Musau]

But do you mean that if my limit is not 0, the margins command is slightly different? For instance, I'm using the following Tobit regression:

tobit date_creation $chararctistics, vce(cluster section) ul(23080)

With no left-censoring and a right-censoring limit at 23080, you have

margins, dydx(*) predict(ystar(.,23080)) atmeans

The syntax is described in:

help tobit postestimation

ystar(a,b) calculates E(y*), where y* = a if xb + u < a, y* = b if xb + u > b, and y* = xb+u otherwise, meaning that y* is censored. a and b are specified as they
are for pr().

a missing (a > .) means minus infinity; pr(.,30) calculates Pr(-infinity < xb + u < 30);
pr(lb,30) calculates Pr(-infinity < xb + u < 30) in observations for which lb > .
and calculates Pr(lb < xb + u < 30) elsewhere.

b missing (b > .) means plus infinity; pr(20,.) calculates Pr(+infinity > xb + u > 20);
pr(20,ub) calculates Pr(+infinity > xb + u > 20) in observations for which ub > .
and calculates Pr(20 < xb + u < ub) elsewhere.