choice-metrics.com

Hello,

I used the following orthogonal design to generate a choice experiment (CE) with three attributes, and NO interactions. Yet for my curiosity, I estimated an RPL with interactions between b3 and b4 and received good results that make sense. My question:

Can I use and interpret these interaction results even though I did not initially include them in the choice design?

;alts = alt1, alt2,alt3
;rows=12
;block=2
;orth=ood
;model:
U(alt1) = b1 + b2 * A[1.50,2.50,3.50,4.50] + b3 * B[0,1,2,3] +b4*c[0,1] /
U(alt2) = b2 * A + b3 * B +b4*C$

Best,
P

Yes absolutely, if you can estimate the model then you can interpret it.

The fact that you did not include the interaction effect at the design stage does not mean you cannot estimate the interaction effect. Including it in the design stage ensures that you will be able to estimate it, but in most cases you will still be able to include many (but not all) interaction effects. Interaction effects cannot be estimated if they are correlated with other effects, and in that case the model would simply not estimate.

Michiel

Thank you Michiel!