choice-metrics.com

Hi everybody,

we are planning to enter one of our quantitative attributes in non-linear form in the utility specification of the design's alternatives. I only found the "dummy" solution in the Ngene manual but we would be interested in employing a quadratic specification of the variable. Is there the possibility to do that in designs such as orthogonal and d-efficient (which we are planning to employ for the pilot and final choice experiment surveys, respectively)? Does Ngene support this option? And if so which is the command to be employed in both designs for a quadratic specification of the attribute term?

Thanks in advance for your support,

Best regards

It depends on what exactly you mean with the quadratic terms. Do you mean a quadratic attribute, or a quadratic parameter?

Quadratic attribute levels can easily be taken into account by converting for example the levels 1,2,3 into levels 1,4,9, the efficiency of the design is then automatically computed correctly.
However, quadratic parameters are more complicated, as they have impact on the efficiency as they require more elaborate derivatives. This is currently not (yet) supported by Ngene.

Michiel

Thank you very much Michiel! Yes, you are right, I meant a quadratic form of the attribute.

Best wishes

A colleague also mentioned that you can also use an interactions to do the same thing, so use b * X * X. Both ways should work.

Thanks a lot Michiel!

Best regards,

Michela