choice-metrics.com

Dear NGEGE team, thank you for validation of my forum account.

I am working on unlabeled DCE and efficient design is new for me. Having read Rim’s topic, I am concerned about potential impact of priors choice on estimation (that can compromise our DCE).

We have unlabeled choice set of three offers of LED bulbs for kitchen and no-buy option. There are 5 attributes chosen after focus group:
- Color temperature (warm, cold, neutral)
- Electricity consumption (4 levels)
- Price (4 levels)
- Two environmental attributes (2 levels both)
We will soon proceed to first small pilot test (15-30 persons).

My questions are:
1) There are no priors. We are almost sure about the sign of last 4 attributes, but no information about their contribution to utility function. It may be equal or not. The first one (color) will depend on consumer preferences. Focus group shows that color and intensity of light depend on where the bulb will installed and on the consumer preferences. A priori, our respondents should prefer cold/neutral color for kitchen, but not all of them. What is the best? Set all priors to 0? Set to 0 the first one and add to other negative/positive effect with equal contribution?
2) We would like also evaluate two-way effect when both environmental attributes have best values. We would like to know if this two-way effect is negative, positive or no significant. So it is very important to us do not affect this analysis by efficient design. If we set the value of prior for interaction to zero, is it ok?

For the moment, I have the following design:

Code: Select all

Design
;alts = alt1*, alt2*, alt3*, alt4
;rows = 36
;eff = (mnl,d)
;model:
U(alt1) = b1.dummy[0|0] * Color[0,1,2] + b2.dummy[0] * Material[0,1] + b3.dummy[0|0] * Envir[0,1,2] + b4[0] * Consumption[4,8,12] + b5[0] * Price[2,5,10,15] +  I[0] * Material.dummy[1] * Envir.dummy[2] /
U(alt2) = b1 * Color + b2 * Material + b3 * Envir + b4 * Consumption + b5 * Price +  I * Material.dummy[1] * Envir.dummy[2] /
U(alt3) = b1 * Color + b2 * Material + b3 * Envir + b4 * Consumption + b5 * Price +  I * Material.dummy[1] * Envir.dummy[2] /
U(alt4) = c[0]
$

Thank you in advance.
Elena

1. Setting priors equal to zero is always safe. Using zero priors in an unlabelled experiment is somewhat similar to using an orthogonal design where also no information on priors is used.

2. You can simply set the interaction priors to zero. Including these interactions in the utility function will ensure that the efficient design can always these estimate interaction effects. Please include all interactions that you think you may want to investigate.

Note that using zero priors does not allow Ngene to remove dominant alternatives, therefore you can remove the * behind the alternative names. It sounds like there is no clear dominance issue anyway in your design, so this should not be a problem. However, if you do feel that certain alternatives in the design become dominant you can include a very small positive or negative prior (e.g. -0.00001 or 0.00001) to indicate the influence of the attribute on the utility (while possibly leaving some of them zero if there is no clear ordering in the attribute levels).

Michiel

Thank you very much, Michel.