Creating a marginsplot for my quadratic interaction term

[Jean Hadji]

Dear all,

I'm working on my masters thesis switched from SPSS to Stata in order to create plots for my interactions. I read that I can create plots using the margins command. However, I can't finy any how-to's for my problem. I have a linear direct effect and want to plot an interaction based on my quadratic moderator (MODSQ).

reg DV IV MOD MODSQ IVxMOD IVxMODSQ CONTROL1 CONTROL2 CONTROL3

Can anyone help me creating a plot showing the marginal effect of IV on DV for different values of my moderator (MODSQ)?

Thank you so much.

Best,
Jean

[Carlo Lazzaro]

Jean:
welcome to this forum.
Just one step behind: if you have created square term and/or interactions by hand, there's nothing that -margins-and -marginsplot- can do for you, as Stata will interpret those independent variable as different predictors.
The easy fix is to use -fvvarlist- for creating categorical variables, square terms and interactions, run -regress- and then rely on the wonderful capabilities of -margins- and -marginsplot-.

[Jean Hadji]

Hi Carlo,

Thank you for your quick reply.

I changed the syntax to the following. I guess that is what you meant?

reg DV c.IV c.MOD c.MOD#c.MOD c.IV#c.MOD c.IV#c.MOD#c.MOD

[Carlo Lazzaro]

Jean:
try the following more compact code and check if you get the same results from your regression:

reg DV c.IV c.MOD##c.MOD c.IV##c.MOD

[Jean Hadji]

I get different results. I don't know why.

I used the following syntax to create a graph showing the marginal effect of IV on DV at different levels for the moderator (MOD). Is that correct?

margins, dydx(IV) at(MOD=(0(0.5)7))
marginsplot, yline(0)

[Carlo Lazzaro]

Jean:
check if what follows give you what you want:

reg DV c.IV##(c.IV##c.MOD)

-margins- and -marginsplot- codes seem OK, although I cannot see what the resulting graphs looks like.