how to find interaction terms in regression modeling

[winecoding]

Given a set of predictor variables, are there any ways to test interaction terms? If I test one by one, it will be impossible to do it?

 

[hlsmith]

Typically there should be some rationale behind the exploration of interaction terms, otherwise you are just mining for significance and it may present itself randomly without logical or real cause.

Exploring all terms can be achieved different ways with different software. In SAS you can seperate terms with| and add @2 at the end of the model statement for 2-way interactions and @n, n any integer, for higher order interactions.

 

[winecoding]

Thanks for the reply. I agree with your point. But customer would like to let software exploring all terms. How to do that in R? Thanks.

 

[kiton]

Well, an interaction is simply a multiplication of two (or more) variables. So, if your model is y=a+bx1+bx2+u, then the model with interaction would be y=a+b1x1+b2x2+b3x1x2+u.

Keep in mind that adding interactions (could) increases the multicollinearity. A good reference book, from my point of view is:
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. Sage.