"Faking" an availability design?

[mcgard]

Hi guys,

I understand that availability designs aren't possible with the current version of Ngene, but I'm wondering whether it's somehow possible to work around this e.g. by generating a BIBD externally and somehow bringing this in to Ngene - or some other method?

Cheers,
Miles

[Michiel Bliemer]

A good question. I do not think it is possible to easily fake it in the current version of Ngene, which is why we have implemented in our research version and hope to distribute in the future.
Your idea of a BIBD design is one of the ways of creating an availability design (which is one of the methods we implemented in our research version) and yes you could combine with a design generated in Ngene. However, the efficient design needs to be aware of the BIBD design, otherwise it cannot optimise on the data that appear in the design. At this moment it is not possible to make Ngene aware of such a BIBD design (or any other 'master' design, which we discussed in the paper Rose, Bliemer, and Louviere, presented at the Choice Modelling Conference in Sydney a few years ago). So simply using a BIBD design and select choice tasks from an efficient design will not work well.

The best way to 'fake' it I think is to create multiple efficient designs for each combination of alternatives, i.e. a design for [alt1, alt2], another design for [alt1, alt3], and for [[alt1, alt4], [alt2, alt3], and [alt3, alt4]. Next, you could create several designs by picking choice tasks from each of these designs. This ensures that all combinations of alternatives are available, and also that the data is efficient. The downside of this approach is that you likely have too many choice tasks, so you may wish to create a heterogeneous design (i.e., multiple versions of the survey in which respondents see one version of the design). I would not recommend leaving some choice tasks out, since this may lead to some parameters to become unreliable to estimate, so you will likely want to use all choice tasks from all efficient designs in the survey.

Michiel

[mcgard]

Thanks, Michiel - that seems like a workable solution. Cheers!