choice-metrics.com

Dear Ngene experts:
I am relatively new to Ngene. In the design of DCE, I encountered some detailed questions. I want to find the answers in the forum. If I can get your help, I would appreciate it very much.

1 I found a sentence in the forum. I am not sure. I try to find the answer here. It is “In an efficient design you should not use design coding or orthogonal coding but instead use the actual attribute levels” Does this mean that design coding cannot be used in all efficient design? What about efficient design? What about efficient design that includes Dummy coding or effect coding?

2 In the case that we only know the direction of the priors (We used + or - 0.0000001 to indicate the direction of priors), we use the syntax shown below to carry pilot.

Design
;alts=alt1*,alt2*,alt3*
;rows=6
;eff=(mnl,6)
;model:
U(alt1)=b1.effects[0.0000001]*A[0,1]+b2.effects[-0.0000001|-0.0000001]*B[0,1,2]+ b3.effects[0.0000001|0.0000001]*C[0,1,2]+b4.effects[0.0000001|0.0000001]*D[0,1,2]+ b5.effects[0.0000001|0.0000001]*E[0,1,2]/
U(alt2)=b1*A+b2*B+b3*C+b4*D+b5*E/
U(alt3)=b0[0]
$

3 And is there any problem with the above syntax? Is “rows=6” and “U(alt3) =b0[0]” feasible? As shown above, I use effect coding, but after collecting the data, I will calculate WTP. And my question is whether the choice of dummy coding or effect coding will have an impact on the final calculation of WTP?

4 Another option is to set the priors to 0, and my question is whether “*” can be used when the priors are zero?

I am sorry I have so many questions. Thank you very much for your help.

Hi Xiaojin, I am also doing some DCE design at the moment for my project. I have some questions about DCE, and do you have Wechat, we can have some discussion about DCE maybe.

Some answers:

1. Design coding is not used in model estimation. Efficient designs require you to write down the utility functions as if you were estimating a model. You can use dummy or effects coding.

2. If you only know the sign, you can use priors close to zero as you mention to indicate only the sign of the prior. This allows Ngene to remove dominant unlabelled alternatives. You should only use a * for unlabelled alternatives, so please use ;alts = alt1*,alt2*,alt3, since it is not possible to check for dominance for alt3 (since it only contains a constant).

3. Your ;eff command, is wrong, please use ;eff = (mnl,d)

4. Dominant alternatives can only be detected if priors have a sign and if dummy/effects coded priors have an ordering, so using * when you have zero priors is not useful. Note that you have not yet provided an ordering in your effects coded attributes. For example, b3.effects[0.00001|0.00002]*C[0,1,2] means the following: level 0 has a utility of 0.0001, level 1 has a utility of 0.0002 and level 2 has a utility of -0.0003. This means that the ordering is: 2<0<1.

Feel free to ask further questions.

Michiel