0

There has been a lot of discussion on interaction terms on this site but I'm looking at a specific situation:

I have 3 groups and a continuous independent variable X;Group1, Group2 are represented by indicators G1 and G2 and Group 3 is the baseline group. The regression 

Y = m + b1X + b2(X*G1) + b3(X*G2)

shows the interaction terms alone to be significant and that the main effect X is not. The trouble is I think that this has something to do with the way the underlying data were grouped into the 3 groups (different customer segments; X can be something like Unemployment rate). While there is strong reason to expect that X influences groups 1 and 2, for business reasons it is not possible to change the way the customer segmentation is done. But it is very important to capture the effect of X on G1 and G2 ( we need the slopes b2 and b3 for further analysis). In this situation, is it okay to drop the main effect X from the model? Edit : according to some of the discussion below, Do all interactions terms need their individual terms in regression model?

...they say it's ok to drop the main effect sometimes...just want to clarify if it's okay in the situation I have described. Thanks for any advice!

Community
  • 1
user2450223
  • 317
  • 3
  • 13
  • That discussion doesn't answer the situation in my example. I looked at another thread too where sometimes they say it is ok to drop the main effect....so want to know if the type of situation above will qualify as an exception.... – user2450223 Apr 08 '17 at 22:02
  • I really think you'll want to read that thread again as cited by @T.E.G. And note that not only X but G needs its main effect addressed if you are going to test the X*G interaction. – rolando2 Apr 08 '17 at 22:24

0 Answers0