Page 1 of 1

NGENE INTERACTION TERM

PostPosted: Tue Feb 19, 2019 9:41 am
by Angelos_Lagoudakis
Dear all,

I would like to ask you an Ngene related question. When in our design we have an interaction term, i.e.: I have 3 attributes:
Price (1,2,3,4)
Organic (0,1)
Cage Free (0,1)
and of course we have alternatives which have both the cage free and organic present. Hence, the interaction term (organic*cagefree), should the ngene give us more choice questions (rows)?. We ran the design with and without the interaction term and the ngene provided the exact same number of choice questions and efficiency. Our design is an " optimal orthogonal in the differences (OOD) design".

My code without the interaction term is is the following:

Design
;alts=alt1,alt2,alt3
;rows=8
;block=2
;orth=ood
;model:
U(alt1)=b1 * PRICE [0,1,2,3] + b2 * ORG [0,1] + b3 * CAGE [0,1]/
U(alt2)=b1 * PRICE +b2 * ORG + b3 * CAGE$

Thank you in advance for your response and I am sorry if there is a similar post in the forum that I didn't see.

Best,

Angelos Lagoudakis
Michigan State University

Re: NGENE INTERACTION TERM

PostPosted: Tue Feb 19, 2019 10:52 am
by Michiel Bliemer
Note that OOD designs work with unlabelled experiments without interactions. You seem to have a no-choice alternative (alt3) and it ignore interactions I believe.

Regarding the number of rows, Ngene generates a design with the number of rows you request (if the orthogonal design of that size exists). Note that adding the interaction org*cage can be estimated without increasing the number of rows.

You should be fine using this design even if you add an interaction term and a no-choice alternative in model estimation.

You could choose to include all possible profiles. There exist 4*2*2 = 16 possible profiles (full factorial) for each alternative. If you set ;rows = 16 and ;block = 4 then you will be sure that there are no issues in estimating any interaction effects.

Michiel

Re: NGENE INTERACTION TERM

PostPosted: Tue Feb 19, 2019 11:15 am
by Michiel Bliemer
You may want to consider running this syntax:

Design
;alts=alt1*,alt2*,alt3
;rows=8
;block=2
;eff=(mnl,d)
;model:
U(alt1)=b1[-0.0001] * PRICE [0,1,2,3] + b2[0.0001] * ORG [0,1] + b3[0.0001] * CAGEFREE [0,1] + b4 * ORG * CAGEFREE/
U(alt2)=b1 * PRICE +b2 * ORG + b3 * CAGEFREE + b4 * ORG * CAGEFREE/
U(alt3)=b0
$

This syntax makes sure that there are no silly choice tasks with dominant alternatives in the survey (i.e. alternatives with cheap organic cagefree eggs) and also takes the constant for the no-choice alternative into account as well as the interaction effect.

Michiel