question regarding number of alternatives

[premc]

Hi,
I am doing a choice design with 24 rows and 5 alternatives (modes: 1,2,3,4 and 5). My problem is that each choice set should have only 4 alternatives and mode 1 should always be present and the rest 3 have to be chosen from the 4 remaining modes. Is it possible to do this using Ngene ?

I am currently doing that by creating a dummy variable in the choice models which takes levels 2-5 which corresponds to modes 2-5. When I run the code, I have a dummy value for each choice set and then I delete the alternative corresponding to that dummy value, for ex if a choice set has a dummy value 4 then I will delete mode-4 from the alternatives. Is this a right way to do ??

Thank you,

-prem

[Andrew Collins]

Hi Prem,

This is a tricky one. The approach you suggest will likely undermine the efficiency of the design from which you remove alternatives, as the removal of the alternatives will affect the choice probabilities, which in turn will effect the fisher matrix and so the efficiency of the design.

The most feasible solution will probably be to create separate designs for each of the 4 (4C3) combinations of modes 2-5. Six rows per combination is probably not going to be enough, depending on how many parameters are in your model, so you might need to increase the number of rows of each design and in total. If you intend to block the 24 rows so that each respondent views less choice tasks, then you will need to manually block, drawing choice tasks from each of the 4 designs.

We have come across a few situations similar to yours recently, and we plan to add functionality that would handle this situation in the next major version of Ngene.

Andrew

[premc]

Hi Andrew,
thanks a lot for the reply.

I will surely try what you have suggested but what bothers me is the number of choice sets in each combination. I am already blocking my sp questions in 4 blocks as we already have a lot of other question in our survey.

I was also trying something else with the models, I have included a dummy variable in all the utilities except the compulsory one, which takes values either "0" or "1" . "0" implying that the mode is not present and now using a condition that the sum of dummy values in all the utilities is equal to three I can by constrain the number of alternatives in each choice set to 4 (including the compulsory mode). But this approach still will not solve the problem with the probabilities of the utilities, so I am trying different coefficients for the dummy variable which will make the utility of the alternative that will be deleted very low (or very low probability). Just hoping I will be able to find the coefficients.

Please let me know if this makes sense to you.

thank you,

regards,
Prem

[Michiel Bliemer]

Your problem is definitely an interesting one, but your approach will indeed lead to wrongly computed probabilities and therefore is unlikely to find an efficient design.

Perhaps you can be more pragmatic, and just use 1+4 alternatives instead 1+3 alternatives, find the most efficient design for these 1+4 alternatives. Then afterwards, investigate each choice task manually and delete the alternative with the lowest probability. In my opinion, this should maintain the efficiency as much as possible.