choice-metrics.com

Dear all,

I hope it is not too trivial question (I was searching in the forum and couldn't find the answer)
In order to obtain priors for the Bayesian design (i.e. the pilot stage) I am running D efficient design.
I know the signs of the parameters, I was writhing the following code assigning [0.0] for positive prior and [-0.0] for negative ones.
Does Ngene treat them as zeros? if so a better approach will be to assign [0.000001] and [-0.000001] respectively ?
Thank you for your time,
Anat

Code: Select all

Design
;alts = buffer_zone,coast_B, mountain_B
;rows=36
;block=4
;eff=(mnl,d)
;reject:
buffer_zone.facilities=3 and buffer_zone.cost=0,
buffer_zone.facilities=2 and buffer_zone.cost=0,
;model:
U(buffer_zone) =c1+b2[0.0]*facilities[1,2,3]+b4[-0.00]*cost[0,1,2]/
U(coast_B)=c2+b1[0.0]*greening[1,2]+b2*facilities[1,2,3]+b3[0.0]*cleanair[1,2]+b4*cost[0,1,2]/
U(mountain_B)= b1*greening[1,2]+b2*facilities[1,2,3]+b3*cleanair[1,2]+b4*cost[0,1,2] $


0.00 and -0.00 are both treated as zero, so indeed you need to use 0.00000001 and -0.00000001.

By the way, using -0.000001 and 0.0000001 to just indicate the sign is only useful if you ask Ngene to remove dominant alternatives in an unlabelled experiment. In your case you are not doing this, so then you may as well use 0. If you want to let Ngene check for dominance, then you may want to include a * after alternatives coast_B and mountain_B which seem to be generic.

Thank you very much Michiel.
In fact it's supposed to be a labelled design.
In coast_B "greening" (trees planting) will take place at the coastal area, whereas in mountain_B it will be on the mountain.
Does it mean that I need different coefficients for greening? in both case the levels are identical (low extent, medium)

In a labelled experiment you can estimate different parameters for greening, facilities, cleanair, and cost for alternatives coast and mountain, but you can also estimate a generic parameter as you currently have specified. This depends on what you expect the behaviour to be. If you think the parameters could be different, then perhaps best to use different parameters and test for differences in model estimation to see whether they should be generic or not.

Perfect, that answers my question. Thanks!