IV regression with continuous interaction term

[paula0]

Hi guys,

I am still far away from being a good econometrician and this is why I'm totally lost with the following problem.
I did an IV regression with logem4 as instrument for my independent variable avexpr:

ivreg logpgp95 mdata1250 (avexpr=logem4), first

Now I want to include an interaction term of both independent variables (both continuous). In OLS I would do a regression like

reg logpgp95 mdata1250 avexpr c.mdata1250#c.avexpr

but how does it work in IV regressions?
ivreg logpgp95 mdata1250 (avexpr=logem4) c.mdata1250#c.avexpr OR
ivreg logpgp95 mdata1250 (avexpr=logem4) c.mdata1250#c.logem4 OR
ivreg logpgp95 mdata1250 (avexpr=logem4) c.mdata1250#c.(avexpr=logem4) ??

Thank you very much in advance,

Annett Paul

[paula0]

Does anyone have an idea? All the ivreg-commands above do not work

[Jeff Wooldridge]

I think this is what you want. To instrument for both avexpr and c.avexpr#c.mdata1250 use

ivreg logpgp95 mdata1250 (avexpr c.avexpr#c.mdata1250 = logem4 c.logem4#c.mdata1250)

Even better, use ivregress 2sls in place of ivreg.

Incidentally, to get meaningful coefficients on the level terms you might want to center the variables about their means or medians before constructing the interaction. Otherwise you will get partial effects at avgexpr = 0 and mdata1250 = 0. I have no idea if that makes sense, but usually it doesn't.