An efficient design with zero priors

This forum is for posts that specifically focus on Ngene.

Moderators: Andrew Collins, Michiel Bliemer, johnr

An efficient design with zero priors

Postby PBallco » Fri May 24, 2024 4:32 am

Hi,
I am trying to estimate an efficient design with zero priors (except price considering the economic theory) and I am not sure which of these two (if any) is the most appropriate one.

This is my first syntax, where I use continuous variables without defining a baseline:

Design
;alts = alt1*, alt2*, optout
;rows=18
;block=2
;eff=(mnl,d)
;model:
U(alt1) =b1 * price[1.1,3.3,5.5] + b2 * com[20, 35, 50] + b3 * ori[1, 2, 3] + b4 * agu[10, 15, 20]/
U(alt2) =b1 * price + b2 * com + b3 * ori + b4 * agu$

After checking a previous post with a similar question, I also created this syntax using dummy variables:

Design
;alts = alt1*, alt2*, optout
;rows=18
;block=2
;eff=(mnl,d)
;model:
U(alt1) =b1 [-0.001] * price[1.10, 3.30, 5.50]
+ b2.dummy[0|0] * com[1, 2, 0] ?
+ b3.dummy[0|0] * ori[1, 2, 0]
+ b4.dummy[0|0] * agu[1, 2, 0]
/
U(alt2) =b1 * price
+ b2 * com
+ b3 * ori
+ b4 * agu
/
U(optout) = b5
$

Any help will be highly appreciated!
PBallco
 
Posts: 7
Joined: Tue Apr 09, 2024 7:59 pm

Re: An efficient design with zero priors

Postby Michiel Bliemer » Fri May 24, 2024 9:57 am

Since you have an optout alternative, which has a different label, you will need an alternative-specific constant in the utility function of either alt1 and alt2, or in the utility function of the optout. So you did this correctly in the second script.

Further, when using zero priors, I generally recommend using dummy coding because otherwise you will likely see comparisons of the high and low levels across alt1 and alt2 as well as middle versus middle levels. This is an artefact of how efficiency works with numerical attributes in combination with zero priors. So I would use script 2. Note that you need to set the prior for price to 0 since you did not specify any other attribute level preference orders. Unless com, ori, and agu also have specific preference orders in their levels then you can keep prior -0.001 and specify also non-zero priors for the other parameters.

Michiel
Michiel Bliemer
 
Posts: 1773
Joined: Tue Mar 31, 2009 4:13 pm

Re: An efficient design with zero priors

Postby PBallco » Fri May 24, 2024 3:58 pm

Thank you so much, Michiel!
PBallco
 
Posts: 7
Joined: Tue Apr 09, 2024 7:59 pm


Return to Choice experiments - Ngene

Who is online

Users browsing this forum: No registered users and 28 guests