Dears Statalists,

In my panel data analysis, I am using interaction term. The model looks something like Y=b0+b1*X+b2*Z+ b3*XZ+e.

Now the interaction term is strongly correlated with X or Z. In order to get around this problem, I read in Azman et al. (2010) and Burill (2007), that I can follow the following proceducer....

reg XZ X Z

then predict errors e_XZ and use these errors instead of XZ.

I did that and its great to see that e_XZ is obviously not correlated with X or Z at all. However, I was expecting e_XZ to be strongly correlated with XZ and thus would justify it usage. However, its not so strongly correlated.

Does anyone have any suggestions please? Should I use it? Is it a problem that e_XZ and XZ are not so strongly correlated and I am using it?

Thank you very much in advance

Arshad


links to the referred papers:
http://minitab.com.au/uploadedFiles/...regression.pdf
https://www.sciencedirect.com/scienc...64999310000635