Design with Opt Out and Status Quo Alternatives

[julian]

Dear Forum Members,

I am designing a Discrete Choice Experiment containing two alternatives plus one fixed alternative(all level being fixed to 1 (in one case 0)) and an opt out alternative. In total hence there are four alternatives
To generate the design, I am wondering what would be a good strategy. One option would be to only include alternatives 1 and 2 in NGene and manually add the other two alternatives later to the design. However, the below syntax was used to incorporate all alternatives. Alternative 4, the opt-out, is expected to derive very very little utility relative to the other alternatives, regardless of the levels (In particular, we ask people about their favorite beach sites after having selected only those people who earlier indicated to plan to go to a beach) and serves only as a residual, in case someone is really unhappy with the presented alternatives. As Ngene, per definition tries to balance utilities, I am thinking to not include this alternative in the design. NGene anyways would never manage to give it a high share. For alternative 3, I am also unhappy with the coding. First, I find it rather bad, that NGENE interprets the same attribute as a different variable with the same parameter, but am not sure what this means in the algorithm.
Another point that I am unclear about is, what NGENE thinks of not defined alternatives in a Utility function. In the output, it displays “blank” instead of 0, yet as far as I understand the manual, it actually treats them as zero. So is it correct to code the attributes, when present with 1 or higher, and when absent, with zero?

Thanks so much for a reply!

Regards,

Julian


Design

;alts = alt1*, alt2*,alt3, alt4

;rows =24
;bseed=74857
;alg = swap
;bdraws= halton(500)
;eff = (mnl,wtp(ZB),mean)
;wtp = ZB(*/b5)
;block=3
;model:
U(alt1) = b7[(n,0.1,0.4)]+b1[(n,0.25,0.04)]*Wassertransparenz[1,2,3] + b2[(n,0.6,0.04)] *Küstenschutz[1,2,3]+b3[(n,0.35,0.04)] *Vogelarten[1,2,3]+b4[(n,0.35,0.04)] *Strandbreite[1,2,3]+b5[(n,-0.5,0.04)] *Naturtaxe[1,2,3,4,5,6] /
U(alt2) = b7+b1*Wassertransparenz + b2*Küstenschutz+b3*Vogelarten+b4*Strandbreite+b5*Naturtaxe /
U(alt3) = b6[(n,0.05,0.4)]+ b1*WSQ[1] + b2 *KSQ[1]+b3 *VSQ[1]+b4 *SSQ[1]+b5 *NSQ[0]

$

[Andrew Collins]

Hi Julian

First of all, unless you explicitly specify it to do so (by specifying utility balance in the eff property), Ngene will not seek to balance utility. Instead, it will try and maximise the information retrieved from the parameters, if you use the d or WTP errors.

I would definitely suggest leaving in the alternatives, if they will be there in the final design. They may be chosen, depending on their utility relative to the others. If you find the choice probability to be low for an alternative, this will be a consequence of the relative utilities across the alternatives, based on the priors specified. The alternative that is not in the utility function will not be displayed in the design matrix, as there are no associated attributes, but the utility is normalised to 0, and that can be verified by examining the utilities in the design properties.

I'm not 100% certain on your comment regarding alternative 3, but it seems like you want to fix the attribute levels, but keep the associated parameters generic across the first 3 alternatives. These can be the same attribute in the final design as for the first two alternatives, but if you use the same attribute name, Ngene will retrieve all of the levels from the earlier alternatives, for convenience. So by using a different attribute name, you can just specify one level (the status quo), and keep the parameters generic. In the final design that you present, you could just treat these alternative 3 specific attributes as the same as their equivalent attributes in the first two alternatives (by having them in the same row of a choice task, with the same attribute name, for example).

Andrew