STATA COMMAND/SYNTAX FOR CALCULATING NET EFFECTS OF INTERACTIONS (dummy variables)

[Chei Bukari]

Hi every one!
I am currently estimating interactions of households with credit*insurance, credit*insurance*savings where all the variables ( credit, savings and insurance) are dummies. Now I am interested in the net-effects of these interactions.
Please can any one assist me on the stata syntax/command for calculating net-effects of these interactions?

[Maarten Buis]

There is no such thing as a net effect when you add interactions. The whole point of an interaction effect is to allow the effects to differ, so logically there are now multiple effects. If you don't like that, then you need to remove the interaction effect and live with the consequence that your model will fit less well.

[Dave Airey]

Are you using factor variables? How about using a command like:

margins credit#insurance
marginsplot

Or looking at the -estat esize- regress postestimation command. What do you mean by net effect?

[Chei Bukari]

Please just for an illustration:

My dependent variable (Y) –household consumption expenditure.

–Independent variables (X) Binary credit, where ‘0’ if household has no credit and ‘1’ if household has credit.

Binary insurance, where ‘0’ if household has no insurance and ‘1’ equal household has insurance.

Interaction term cre_insu = credit * insurance.


After regression, the equation is Y = 20 –4.1*insurance + 2.1*credit + 1.3*cre_insu


The effect of credit on the expenditure is +2.1 but given the interaction term (and given that all coefficients are significant), the net effect is 2.1 + 1.3*insurance. If insurance is 0, then the effect is +2.1(which is credit coefficient), but if insurance is 1 then the effect is 2.1 +1.3 = 3.4 In this case, the effect of credit is more positive in households with insurance.

But my headache is calculating this manually as illustrated above. So I was thinking stata could have a command/syntax that will add net effects results to my regression output.

Otherwise I would have to do it manually. I guess you understand me now?

[Clyde Schechter]

Dave Airey's response in #3 is precisely what you are looking for. But you need to rerun your regression using factor-variable notation. See -help fvvarlist- for details on that. Your code should look something like this:

regress y i.credit##i.insurance
margins insurance, dydx(credit)

Note: Not tested, beware of typos.