choice-metrics.com

Hi Michiel,

Hope everything goes well for you.

I got 0% of D optimality when I run a ood design. the codes I run is below. Is there anything wrong?
Thanks so much,
Wei

Code: Select all

Design
;alts = alt1*, alt2*, nobuy
;rows = 24
;block = 3
;orth = ood
;model:

U(alt1) = b0 + b1 * Price[1.20, 3.26, 5.32, 7.38, 9.44, 11.50] +
               b2.dummy * CFI[0,1] +
               b3.dummy * Org[0,1] +
               b4.dummy * RFA[0,1] +
               b5 * CFI.dummy[1]*Org.dummy[1] +
               b6 * CFI.dummy[1]*RFA.dummy[1] +
               b7 * RFA.dummy[1]*Org.dummy[1] /
U(alt2) = b0 + b1 * Price +
               b2 * CFI +
               b3 * Org +
               b4 * RFA +
               b5 * CFI*Org +
               b6 * CFI*RFA +
               b7 * RFA*Org
     $


I am not sure why it generates a D-optimality of 0%. OOD designs are tricky and sometimes the D-optimality measure as proposed by Street and Burgess gives 0% for no obvious reason. The D-error looks fine so there should be no issue using this design to estimate the coefficients. Note that OOD designs have limitations, for example they cannot remove dominant alternatives (so the * in the ;alts command will not work because OOD designs require an orthogonal design and do not allow removing choice tasks with a dominant alternative).

Instead, you could opt for a D-efficient design using ;eff = (mnl,d) with zero priors or small positive/negative priors in order toremove dominant alternatives.

Michiel

Michiel,

Thanks so much for your help. One more question, we mentioned that "The D-error looks fine", and how could you check the D-error in ood design? I did not find the place to check D-error on the result page displayed in Ngene.

Wei

Tick the box "Design properties, MNL" on the left in the design output window.

Thanks so much. I really appreciate your help.

Wei