choice-metrics.com

Hi,
I want to design a pilot for understanding preferences for reifiable packages (of shampoo for instance) vs. the regular package (opt-out).
I have screened the literature for the relevant att. and levels, but I'm afraid that for the second and the third atts (below) the level range is too narrow to identify the preferences.
I'm interested in both forecasting and understanding preferences.. according to the following design (the priors are a a guess), the probabilities are very close though the choice distribution is fairly equal.
My question is, is there an a-prior way to know if the level range is too narrow other than actually piloting and seeing if the coefficients are significant)?

Code: Select all

Design
;alts = refilla*, refillb*, regularshampoo*
;rows = 12
;block = 3
;eff = (mnl,d)
;model:
U(refilla)     =  recyclability[0.00001]*recyclability[50,75,95] ? percentage of packages that can be recycled
                    +saveplastic[0.001]*saveplastic[75,85,95] ?percentage of saved plastic 
                    +savevolume[0.000001]*savevolume[75,85,95]? the volume of volume waste that is being saved
                    +availability.dummy[0.0001]*availability[2,1]? availability in chains 
                    + price[-0.01]* price[14,18,22,24]/
U(refillb)  =     recyclability*recyclability
                    +saveplastic*saveplastic
                    +savevolume*availability
                    + price* price /

U(regularshampoo)  =  b0[0.2]+ price[-0.01]* pricers[24,26,28,30] $         


There is not really knowing whether the range is too narrow until you have done some pretesting and/or pilot testing. You can probably add more levels if you want to.

Note that with near zero priors your probabilities will be close to equal, so that is simply an artefact of using uninformative priors. I would probably make the price prior closer to zero as it now stands out as the only one that has a more informative prior and one should preferably not mix informative with uninformative priors.
Further, your regular shampoo does not have the same attributes as the other alternatives, so Ngene will not be able to perform dominance checks with respect to regular shampoo. You may want to indicate its properties with respect to recyclability, saveplastic, and availability so that it is clear what people are choosing between. If for example recyclability is 0 then just add recyclability * recyclability_sq[0] or something.

Michiel

Perfectly clear, thanks!!!