choice-metrics.com

Hi,
We are working on a choice experiment for which we don't have any previous pilot done yet. Thus (to begin with) we would like to specify a labelled orthogonal design. In the experiment we have a scenario variable (which is the same across all alternatives) and an attribute which should have the same level for 2 of the 3 presented alternatives. This seems to work with the ;cond or ;require or using the following specification, but only for efficient designs.
U(alt1)=b1*x1[1,2,3,4]...
U(alt2)=b2*x1[x1]...
U(alt3)=b3
Is it possible to implement these restrictions in an orthogonal design? And if it is not possible, how would you suggest to proceed given that we do not have any priors?
Thank you very much in advance.

It is not possible to restrict orthogonal designs since they are very specific desigms where each level combination needs to appear an equal number of times. Any constraint would break orthogonality.
Luckily, strict orthogonality is not needed; you can approach orthogonal desigms by creating an efficient design with zero priors. So you can just omit the prior values in your syntax and Ngene will assume that they are zero.

Michiel