Orthogonal fractional factorial designs with 2-way interacti

This forum is for posts covering broader stated choice experimental design issues.

Moderators: Andrew Collins, Michiel Bliemer, johnr

Orthogonal fractional factorial designs with 2-way interacti

Postby daniele.asioli » Thu Jun 29, 2017 5:12 pm

Hello, I am a new user of Ngene. and I apologise in advance if my question has already been previously discussed. I am working on a choice experiment (unlabelled with two product alternatives and a no-buy option) and I would be interested in including two-ways interactions affects across the experimental attributes. I would like to generate an orthogonal fractional factorial design with two-way interactions using the function "foldover". However, I am wondering whether this design could also work well if the interaction effects are not taken in consideration in the econometric analysis.

I thank you in advance for your kind attention.

Daniele
daniele.asioli
 
Posts: 1
Joined: Wed Jun 28, 2017 6:08 am

Re: Orthogonal fractional factorial designs with 2-way inter

Postby Michiel Bliemer » Thu Jun 29, 2017 10:24 pm

Dear Daniele,

Yes such a design would also work fine if you do not estimate any interaction effects.
The fact that it is orthogonal in the main effects and orthogonal between each main effect and interaction effect simply ensures that all main effect combinations and all combinations between main effects and interaction effects are nicely covered.

Note that orthogonal designs have no benefits over non-orthogonal designs in estimating discrete choice models since an orthogonal design for such (nonlinear) models does not translate into independent parameter estimates. Only in linear (regression) models this is the case. So strictly adhering to orthogonality is not necessary. The main benefit of orthogonal designs is (as said) that it nicely covers the space assuming that each attribute and each interaction is equally important in the data collection.

Using the foldover increases the number of choice tasks, giving you 'richer' data, which allows estimating more parameters, but will clearly also work if you estimate less parameters.

Michiel
Michiel Bliemer
 
Posts: 1705
Joined: Tue Mar 31, 2009 4:13 pm


Return to Choice experiments - general

Who is online

Users browsing this forum: Google [Bot] and 8 guests

cron