[Carlo Lazzaro]

Giorgio:
what if you delete interaction and add a-e predictors one at a time (that is, other things being equal, a; a-b; a-b-c; a-b-c-d; a-b-c-d-e).

[Giorgio Carlini]

Carlo, the aim is to investigate the impact of covid and gdp and their interaction on the dep. var. Some of the var (a-e) are not statistically significant, while other are.
I was thinking to add a quadratic term for both the covid and gdp. the utest suggest that this is an option.
How to add both quadratic terms? I'm not convinced if I need to interact the sqaured terms of covid with the linear and the quadratic of gdp and also the sqaured term of gdp with the linear and quadratic term of covid? In other words, 4 interaction terms are correct?

Y=beta1*X+beta2*X(squared)+ beta3*Z+beta4*z(squared) +beta5*(X*Z)+beta6(Xsquared*Z)+beta7(X*Zsquared)+b eta8(Xsquared*Zsquared)

If this is the case, is this correct:
reg firms c.covid##c.covid##c.gdp c.covid##c.gdp##c.gdp , robust or this reg firms c.covid##c.covid##c.gdp##c.gdp , robust

[Carlo Lazzaro]

Giorgio:
you may have in mind something along the following lines:

reg firms c.covid##c.covid c.gdp##c.gdp, robust

[Giorgio Carlini]

Carlo, to your specification I want to add the intercation between linear and quadratic of the two var.
that is +beta5*(covid*gdp)+beta6(covidsquared*gdp)+beta7(c ovid*gdpsquared)+b eta8(covidsquared*gdpsquared)

[Carlo Lazzaro]

Giorgio:
you may want to try you previous code, then:
[CODE]
reg firms c.covid##c.covid##c.gdp c.gdp##c.gdp##covid, robust /CODE]

I suspect that collinearity will bite hard, though.

[Giorgio Carlini]

Hello Carlo,
yes this is the case, high collinearity.
One option is to add only one squared term at once only and see what happen. Do you agree?

[Carlo Lazzaro]

Giorgio:
yes, I would give it a shot.