Interaction term analysis

[Abdullah Ijaz]

Hi All,

My question is related to this reg results. This is panel data -fixed effects reg results. I want to understand the "Use of Interaction term" in this regression.

As you can see when terrorism var is interacted with inequality var, we get significant results. However, individually terrorism is also significant and the coefficient is even bigger. Hence, does it make sense to use interaction term in this reg ??

Please suggest.

[Daniel Schaefer]

The short answer is yes, it does make sense.

The longer answer is that I've seen at least one scholar argue that interactions where one term is not significantly associated with the outcome are "better" because they are easier to interpret: it is clear that the effect of the significant term is changing with respect to the context provided by the nonsignificant term. I don't think that it matters that one effect size is bigger than the other. Take care when interpreting these as the units of the underlying measure matter.

However, there is nothing statistically invalid about this model, and this kind of thing is done all of the time. The authors are just modeling the fact that the size of the effect of terrorism on the outcome changes with respect to levels of inequality, and the size of the effect of inequality on the outcome changes with respect to levels of terrorism. Again, this is perfectly valid, just a little bit more difficult to put into language that is clear to a non-statistical audience.

[Abdullah Ijaz]

Daniel Schaefer That's really helpful. Much appreciated!