choice-metrics.com

Dear Ngene Team,

thank you for accepting my enrollment request. I am designing my first DCE, but I have problems with coding. Briefly, my labelled CE includes two alternatives, five attributes and their levels as follow:

price = (775, 875, 975, 1075)
mkt = (274, 312, 350, 388)
damage = (0=no damage, 1= low damage, 2=high damage)
fung =(0=not required, 1= required)
ev= (0=low, 1=medium, 2=high)

I have designed the model as:

Design
;alts = A, B
;rows = 60
;block = 6
;eff = (mnl, d)
;cond:
if (A.damage=2, A.fung=1 and A.ev=2), if (B.damage=1, B.fung=1 and B.ev=1), if (B.damage=0, B.fung=0 and B.ev=0)
;model:
U(A) = a0 [0.0001]
+ b1[-0.0001]* price [775, 875, 975, 1075]
+ b2[0.0001] * mkt [274, 312, 350, 388]
+ a1[-0.0001]* damage [0,1, 2]
+ a2[-0.0001]* fung [0, 1]
+ a3[-0.0001]* ev [1, 2, 3]/

U(B) = c0 [0.0001]
+ b1*price
+ b2*mkt
+ c1*damage
+ c2*fung
+ c3*ev
$

I am aware that there are several problems:
-) the 'cond' is not correct. As shown in the design, I need to constraint different attributes of the same alternatives, but this formula could work only on attribute leves. How can I manage it? Should I insert it as a 'require'?
-) I cannot run a pilot survey before, so I choose as prior '0.0001' and put a sign in accordance to the kind of alternative, that might be correct;
-) I have concerns also about the last three attributes (damage, fung, ev). It should be better if I specfiy them for the two alternatives? The 'price' and 'mkt' attributes must be generic.

Sorry for the long message, and thank you in advance.
Best Regards,
LF

Please note that Ngene related questions should be asked in the Ngene forum, not in this forum which is for general questions about stated choice experiments.

I do not understand what you mean with "I need to constraint different attributes of the same alternatives, but this formula could work only on attribute levels", can you please elaborate?

Note that you cannot use constants in all utility functions, please remove a0 or c0.

Your priors are fine if you only the know sign. The sign is only used for dominance checks. Since your experiment is labelled, there is no dominance issue to you can simply use zero priors since the sign will not have any impact. If you would have assumed an unlabelled experiment then you would use ;alts = A*, B* (which removes dominant alternatives) and the sign would be of importance.

If you are unsure about whether coefficients are generic or alternative specific, then it is best to use alternative specific coefficients (as you have done) since this allows testing for both situations.

Michiel

Dear Michiel,

thank you for your kind and helpful response, sorry for having posted in the wrong section, I am going to reply to your first question and then I am going to move in the Ngene forum.

To the best of my knowledge, constraint are usually posted as:
; cond:
if (alt1.att.1 =x, alt2.att1 = [y, z])

Actually, I need that:
; cond:
if (if (alt1.att3=2, alt1.att4=1 and alt1.att5=2),
if (alt2.att3=1, alt2.att4=1 and alt2.att5=1),
if (alt2.att3=0, alt2.att4=0 and alt2.att4=0)

that is, if damage in alternative 1 is equal to 2, then the fungicide of alt1 must be equal to 1 and ev of alt 1 must be equal to 2, and so on.

I am not sure how to manage these constraints, I think that the syntax is not correct.
Thank you for your help,
L
__________________________________________________________________
PS. (Update) it seems to work righ now, there was pobably some other mistakes!

Most people use ;cond constraints exactly the way you specify it, namely that an attribute in one alternative affects the level of another attribute in the same attribute, so your syntax is fine.

Michiel