Pilot study syntax. Am I doing it right?

[hafizah8987]

Hi all,

I'm having a doubt about my design and I need help to confirm that I'm on the right track for my pilot study. Honestly, I have zero knowledge in the choice experiment but really willing to explore more. I just started using Ngene a month ago and develop my syntax based on the manual (very helpful !!!) and the posts of this forum.

Here is my syntax for pilot study:

Design
;alts = alt1*, alt2*, alt3
;rows = 16
;eff = (mnl,d)
;model:
U(alt1) = b2.dummy[0|0|0] * A[0,1,2,3] + b3.dummy[0|0] * B[0,1,2] + b4.dummy[0|0] * C[0,1,2] + b5.dummy[0|0] * D[0,1,2] + b6[-0.0001] * E[1.30,1.50,1.70,1.90] /
U(alt2) = b2* A+ b3* B+ b4* C+ b5* D+ b6 * E
$

Few questions which might be sound very naive and nonsense :

1) I had tried running a syntax without (-) negative prior on the price variable (E). It works fine. But, according to literature, price relationship should be negative in nature and that's why I inserted the negative prior (-0.0001). However, Ngene unable to provide me with a design. Am I doing something wrong?

2) When all the prior assume as zero, S estimate should be zero as well?

3) I plan to insert a compensatory effect into my study. I read about the 'attribute non-attendance' literature by David Hensher and Rose. And I found that the effect or model will be inserted/tested later when I get the final data and not at this choice design stage. Am I right?

Please help me!!!!

Hafizah

[Michiel Bliemer]

You are well on your way, your syntax looks fine.

Answers to your questions:

1. putting an asterisk (*) after alt1 and alt2 makes Ngene check for dominant alternatives. Ngene can only check for dominant alternatives if it understands the trade-offs between attributes, i.e. if they have a positive or negative sign. While it is OK to have some priors set to zero, only having one prior with a negative sign does not allow Ngene to properly check for dominant alternatives and will think that everything is dominant. Is there a natural ordering in your dummy coded variables? For example, if level 0 is better than level 1, and level 1 is better than level 2, then you can set very small priors such as b3.dummy[0.0002|0.0001]*B[0,1,2], noting that level 2 is the reference level and hence set to zero. If there is no natural ordering, then it is best to remove the * or to set the price parameter to zero since Ngene will not be able to check for dominance.

2. If priors are zero, then the S-estimate will be infinite. S-estimates will only provide a meaningful value if your priors are set to realistic values (e.g. coming from a pilot study).

3. The analysis of attribute non-attendance is an analysis after you have collected the data, so indeed this does not play yet at the choice design stage.

Further, since alt3 seems to be your opt-out, you will need to either add a constant for this alternative, or add a constant to alt1 and alt2 (the same constant).

Michiel