I have really been curious about this, but I have been searching around for studies which utilize Spatial Durbin models and I have yet to find any one model that has incorporated the use of interaction effects. For instance, imagine it a Durbin model was executed and you see results like this...
Coefficients:
Estimate Pr(>|z|)
(Intercept) -2.0324e-02 0.8452031
DummyX 3.5832e-01 0.0290771
X1 2.5424e-02 0.7522377
X1:DummyX -4.3316e-01 0.0001841
LagDummyX 3.1791e-02 0.0567951
LagX1 1.4026e-02 0.0015517
LagX1:DummyX -9.6428e-03 0.3418146
I don't see any kinds of models like this. I have seen the use of both continuous and discrete variables but never a SDM utilizing an interaction between the two. Why is that the case? I think there may be difficulty with interpretation, granted, but these studies do not seem to even mention the disregard of using said interactions, which seem theoretically viable for me. I have run a few models that appear to have different I don't really get it. Anyone with spatial regression experience? Any help would be very appreciated.
