[R] Re : Design of experiments for Choice-Based Conjoint Analysis (CBC)

John Williams john.williams at otago.ac.nz
Mon Jun 21 23:47:09 CEST 2010


On 21/06/10 20:32, Aline Deschamps wrote:
> Thank you for your mail and the attached documents.
You're most welcome :-)

> I've read the article and I found it very interesting, but I have some
> questions :
>
> - The 5 steps in this article are for creating a choice set for one
> respondent, are they?
Yes, that's correct.

> - In this article, it is written : "When two or more alternatives having
> the same combination of attributes levels appear in one question, it is
> necessary to repeat step 4". Does that mean we have to test each
> question and go back to the beginning of the step 4 if we are in this
> type of case, e. g. re add a new random number to each design?
Yes, this is common to all conjoint studies where a fractional factorial 
is generated by computer.  So, to put it in CBC terminology, you examine 
each choice set for duplicate alternatives.

> - In the function "optFederov", they choose the model " ~. " , but if I
> want to study the first order interactions between the different
> attributes, should I choose the model " ~.^2 " ?
Don't know, sorry.  I am just learning about this myself, and do not 
claim to be expert.  I have just printed out the AlgDesign manual and 
will be reading it when I get sick of marking exams (10 down, 260 to 
go!).  So, while I may be able to answer this question eventually, it 
might be next week ...

> - Finally, I need to make sure that the following condition : "every
> respondent has to see, for all attributes, each of its levels
> approximately the same number of times", is verified in my design. Is it
> the case with that method? I'm not really sure...
Nor am I.  I think D-Optimality measures that, but I can't swear to it.

I won't be able to work on this for the next few days, but am certainly 
willing to share any solutions I've found to our common problems in the 
future.  Would you let me know if you find the answers?  Perhaps we 
should be having this discussion on-list?  (I've assumed "yes", so have 
CC-ed R-Help.)

Cheers,

John



More information about the R-help mailing list