Too many interaction terms?

[Lauw]

I'm using a panel data regression to see effects of return on assets on CEO remuneration. Furthermore i've added variables such as board size and board independency. I've also added interaction terms board size*ROA and board independency*ROA, but once i start adding those, ROA falls from significant positive to significant negative. Intuitively, negative ROA-remuneration relation would make little sense. Fixed or random effects change nothing. Is there any way past this besides delete interaction terms?

[startz]

Think carefully about what your estimates imply for the partial derivative of remuneration with respect to ROA.

[Lauw]

3 ROA terms?

[Lauw]

Any way i can get past this then?

[startz]

Did you look at the estimated partial derivatives?

[Lauw]

How would i do this?

[Lauw]

Anybody? I'm pretty stuck.