choice-metrics.com

Hi,

I am new to Ngene. I am doing an unlabelled design with 2 attributes with 4 levels each. When I ran the following code 4 of the 8 choice sets are dominant. So how can I make sure that my design is statistically efficient? I am not sure if I can run a d-efficient design if I don't have any data on priors. Any suggestions would be greatly appreciated.

Design
;alts = alt1, alt2
;rows = 8
;orth =seq
;model:
U(alt1) = b1 + b2 * A[1,2,6,12] + b3 * B[5,15,25,40] /
U(alt2) = b2 * A + b3 * B
$

Kind regards,
Tara

Orthogonal designs cannot avoid dominant alternatives, the only way to do that is to use a different type of design, such as an efficient design. If you do not know the priors, you can use values that are near-zero, namely -0.00001 for a negative prior and 0.00001 for a positive prior. I do not know whether A and B have a positive or negative prior, but in the end your code would be something like:

Design
;alts = alt1*, alt2* ? note that I have added * to let Ngene check for dominance
;rows = 8
;eff = (mnl,d) ? I have ;orth with ;eff
;model:
? Assuming both negative priors, but please select appropriate sign for your case
U(alt1) = b2[-0.00001] * A[1,2,6,12] ? I have removed the constant, which should not be there because it is an unlabelled design. If you like, you can add the constant in model estimation to account for left-to-right reading bias.
+ b3[-0.00001] * B[5,15,25,40]
/
U(alt2) = b2 * A + b3 * B
$

Michiel

Thanks Michiel, that's really helpful.

Tara