Questions on efficient design for choice experiment

[bpaudel]

Hi,
I am designing a choice experiment for my research project. I want to use efficient design to generate choice sets for my choice experiment. I am thinking to generate: 4 blocks with a total of 24 choice sets and include 3 alternatives (Option A, Option B and Neither) in a choice set.

Regarding the attributes and levels, I have 5 attributes (Price, nutritional content, use of pesticide, technology and developer) with their levels mentioned below:

Price ($/48 fl. oz.):$1.99, $3.49, $4.99, $6.49 ($1.5 incremental)
Nutritional content (Oleic acid): High, Normal
Use of pesticide: Reduced use, No reduction
Technology: Gene-editing, Genetically modified, Conventional, No Information
Developer: University, A multinational firm, Domestic start up, No Information

The prior values in the Ngene design are taken from literature. There are some restrictions in the design too. The Ngene design that I have developed is mentioned below:

Dummies in design:

High oleic acid=1, Normal oleic acid= 0
Reduced pesticide use=1, No reduction in pesticide use= 0
Gene-editing=3, Genetically Modified=2, Conventional=1, No information=0
University=3, A multinational firm=2, Domestic start-up=1, No information=0

Ngene design:

Design
;alts= optA*, optB*, Neither
;rows=24
;eff=(mnl,d, mean)
;block=4
;alg= mfederov
;reject:
optA.Tech =3 and optA.OA =0,
optB.Tech =3 and optB.OA =0,
optA.Tech =3 and optA.Pesticide =0,
optB.Tech =3 and optB.Pesticide =0,
optA.Tech =2 and optA.OA =0,
optB.Tech =2 and optB.OA =0,
optA.Tech =2 and optA.Pesticide =0,
optB.Tech =2 and optB.Pesticide =0,
optA.Tech =1 and optA.OA =1,
optB.Tech =1 and optB.OA =1,
optA.Tech =1 and optA.Pesticide =1,
optB.Tech =1 and optB.Pesticide =1
;model:
U(optA) = b1 [(n,-0.501, 0.009)] * price[1.99, 3.49, 4.99, 6.49]
+ b2.dummy[(n, 0.535, 0.041)]*OA[1,0]
+ b3.dummy[(n, 0.537, 0.038)]*Pesticide[1, 0]
+ b4.dummy[(n, -0.987, 0.040)|( n, -1.158, 0.039)| (n, 0.55, 0.019)] * Tech[3, 2, 1, 0]
+ b5.dummy[(n, 0.36, 0.041)|(n, -0.408, 0.046)|(n, 0.022, 0.038)] * Developer[3, 2, 1, 0] /
U(optB) = b1 * price
+ b2 * OA
+ b3 * Pesticide
+ b4 * Tech
+ b5 * Developer /
U(Neither) = b0 [(n, -1.695, 0.053)]
$

Questions/Concerns:
• D-error of design is quite high.
• The generated choice sets have one price level missing. Most of the choice sets have price level of $1.99 and $6.99 (highest and the lowest one).
• I have designed model with some technology restrictions (Technology Gene-editing and genetically modified cannot occur with nutritional content level normal oleic acid, and
pesticide use level no reduction. Similarly, technology conventional cannot occur with nutritional content level high oleic acid, and pesticide use level reduced use. Here, my
concern is, when I restrict technology with these restrictions, choice options showing technology-gene editing, genetically modified and conventional (which have restrictions) are less while choice options showing technology with no information (which have no restriction) are more.
Basically, one of the objectives of my is study is to examine consumer willingness to pay for the mentioned technologies. Does less occurrence of technology -gene editing, genetically modified and conventional in choice sets affect the results for willingness to pay estimates of these attributes?

I kindly request you to provide insights on above mentioned concerns.
Can you please suggest me, how can I improve the design?

With regards,
bpaudel

[Michiel Bliemer]

Your D-error looks fine to me. D-errors with many dummy coded variables tend to be higher than D-error with continuous variables only.

The modified Federov algorithm does not guarantee attribute level balance. If you would like to have more attribute level balance, you need to impose attribute level constraints as in the syntax below, where I for example require that each price level needs to appear between 4 and 8 times in the design. Note that these are again additional constraints and hence will reduce efficiency (but I would not worry about that too much).

Code: Select all

Design
;alts= optA*, optB*, Neither
;rows=24
;eff=(mnl,d, mean)
;block=4
;alg= mfederov
;reject:
optA.Tech =3 and optA.OA =0,
optB.Tech =3 and optB.OA =0,
optA.Tech =3 and optA.Pesticide =0,
optB.Tech =3 and optB.Pesticide =0,
optA.Tech =2 and optA.OA =0,
optB.Tech =2 and optB.OA =0,
optA.Tech =2 and optA.Pesticide =0,
optB.Tech =2 and optB.Pesticide =0,
optA.Tech =1 and optA.OA =1,
optB.Tech =1 and optB.OA =1,
optA.Tech =1 and optA.Pesticide =1,
optB.Tech =1 and optB.Pesticide =1
;model:
U(optA) = b1 [(n,-0.501, 0.009)] * price[1.99, 3.49, 4.99, 6.49](4-8,4-8,4-8,4-8)
+ b2.dummy[(n, 0.535, 0.041)]*OA[1,0](10-14,10-14)
+ b3.dummy[(n, 0.537, 0.038)]*Pesticide[1, 0](10-14,10-14)
+ b4.dummy[(n, -0.987, 0.040)|( n, -1.158, 0.039)| (n, 0.55, 0.019)] * Tech[3, 2, 1, 0](4-8,4-8,4-8,4-8)
+ b5.dummy[(n, 0.36, 0.041)|(n, -0.408, 0.046)|(n, 0.022, 0.038)] * Developer[3, 2, 1, 0](4-8,4-8,4-8,4-8) /
U(optB) = b1 * price
+ b2 * OA
+ b3 * Pesticide
+ b4 * Tech
+ b5 * Developer /
U(Neither) = b0 [(n, -1.695, 0.053)]
$

Please note that this forum is only for general questions about stated choice experimental design, please ask Ngene related questions in the Ngene forum.

Michiel

[bpaudel]

This is really helpful for me. Thank you so much for your valuable suggestions.

With regards,
bpaudel