HELP ON CE DESIGN FOR WTP

[Kofi Larvoe]

I am quite new to NGENE recently designed a choice experiment. Please, I need a check on my design if it is right because it is important that everything is correct before I start the data collection.

Brief background: The experiment is about consumers´ willingness to pay for sustainable wines i.e. wines produced with a new technology that allow farmers to reduce pesticides application at the farm.

I used three attributes:
1. Pesticide(PEST) reduction with 4 levels: 10%, 20%, 30% and 40% -reductions from current applications
2. Certification(CERT) with 4 levels: No certicfcation (0), Third-party cert(1)., Participatory gurantee cert(2). and Eu/National cert(3).
3. Price(PRIC) with 4 levels: 5%, 20%, 35% and 50% -increments from current prevailing market prices

Design
;alts = alt1*, alt2*, NONE
;rows = 16
;con
;eff = (mnl,d)
;model:
U(alt1) = b2[0] * PRIC[5, 20, 35, 50] + b3.effects[0|0|0] * PEST[10, 20, 30, 40] + b4.effects[0|0|0]* CERT[1, 2, 3, 0]/
U(alt2) = b2 * PRIC + b3 * PEST + b4* CERT/
U(NONE) = b6[0]
$

As you can see, I used efficient design with zero PRIORS and two blocks.
I settled on the evaluation 449 because it has the highest D-optimality of 89.29%

Some concerns:
1. Is the D-optimality more important than the D- and A-errors? Becasue the D-optimality reduces to zerro as the iteration increases and the errors reduces.
2. Will the use of effects instead of dummy affect my outcome?
3. Could you please assist me with the use of stars(*) on alt1 and alt2. I was advised to add them but I cannot explain its significance
4. Can the design still be improved in any way possible? Thus, increase the D-optimality or reduce the errors
5. How can I ensure that higher price increment levels are paired with higher pesticide reduction levels only and vice versa

[Michiel Bliemer]

1. You can ignore D-optimality, which is only useful for optimal orthogonal designs, so just focus on reducing D-error.
2. The behavioural model with dummy and effects coding is exactly the same; the parameters will of course be slightly different because the reference level is different.
3. The asterisks (*) indicate that alt1 and alt2 are generic and allows Ngene to avoid dominant alternatives if non-zero priors are specified. I have used small positive/negative priors in the script below to indicate the preference order of the attribute levels. No dominant alternatives now appear.
4. I think it looks fine. Consider using ;block = 2 as perhaps 16 choice tasks per respondent is quite a lot, blocking it in 2 would mean only 8 per respondent.
5. You could add constraints, see script below.

Note that you will not be able to determine WTP in monetary values unless you also know the current market prices of the wines and the respondents also know this. Otherwise, using a percentage increase for price will be problematic. It is best to use an absolute price level to allow WTP calculations.

Code: Select all

Design
;alts = alt1*, alt2*, NONE
;rows = 16
;con
;eff = (mnl,d)
;cond:
if(alt1.PRIC = 5,  alt1.PEST = [10,20]),
if(alt1.PRIC = 20, alt1.PEST = [10,20,30]),
if(alt1.PRIC = 35, alt1.PEST = [20,30,40]),
if(alt1.PRIC = 50, alt1.PEST = [30,40]),
if(alt2.PRIC = 5,  alt2.PEST = [10,20]),
if(alt2.PRIC = 20, alt2.PEST = [10,20,30]),
if(alt2.PRIC = 35, alt2.PEST = [20,30,40]),
if(alt2.PRIC = 50, alt2.PEST = [30,40])
;model:
U(alt1) = b2[-0.0001]                   * PRIC[5, 20, 35, 50]
        + b3.effects[0.001|0.002|0.003] * PEST[20, 30, 40, 10]
        + b4.effects[0.001|0.002|0.003] * CERT[1, 2, 3, 0]
        /
U(alt2) = b2 * PRIC + b3 * PEST + b4* CERT
        /
U(NONE) = b6[0]
$

Michiel

[Kofi Larvoe]

I am so thankful for the explanation Michiel.
Yes, I will use two blocks as you indicated.
I understand your concern regarding the WTP estimation. Just to clarify that I am using a two stage sampling method where consumers first state how much they pay for the product in the first atage. The percetage levels are then applied in the second stage.

Thank you once again.