choice-metrics.com

Dear Concern,

I am having a stability issues in my experimental design. I have 5 attributes each with 3 levels (i am using linear coding -> Level 1, 2, 3), and each one pair of the levels occurs equally across one pair of attribute. I am not sure if it is possible to a D- efficient design? I am preferring to make 2 versions/blocking in my design.

Any suggestions would be highly appreciated.

Could you please clarify what you mean with "each one pair of the levels occurs equally across one pair of attribute", this sounds like orthogonality? How is this related to "stability issues"? Note that using linear coding (1, 2, 3) is not appropriate for efficient designs, you need to use the actual levels that you would use in model estimation, including any coding scheme (e.g. dummy coding for categorical variables).

Once I better understand the issue I hope to be able of more assistance.

Michiel

Dear Sir,

In one of my choice task even after providing the prohibitions in Choice based conjoint analysis, the dominant alternative is still persisting and hence it affects the efficiency of my design.
I am using the actual design levels (Lvl 1,Lvl 2, Lvl 3), which i would use in my model estimation. The choice experiments does not have a categorical variable.

Is there any ways to overcome such problem?

So I understand that the issue is that you have a dominant alternative?

In Ngene, you can indicate that you have generic alternatives by adding an asterisk after the alternative names, for example:
;alts = alt1*, alt2*

This ensures that dominant alternatives are avoided (but only if you provide informative priors or noninformative priors with the sign of the coefficient, e.g. -0.000001 or 0.00001 for a positive or negative prior) and it avoids identical and repeated choice tasks.

Michiel

Thank you for your reply Sir.