Conjoint analysis or discrete choice analysis ? And which software?

Question

Suppose I would like to model the light-bulb preference of respondents. I would like to ask their preferences between existing bulbs on the market. And when you choose a bulb in a shop you usually see a package with many information gathered togheter such as:

• technology (halo, fluo, led)
• price
• energy label EURO norm (from E to A++)
• shape (pear, candle, spiral, steam)
• watts.
• brand (philips,osram , etc)

As there are too many combination respondents should only see max three combinations and check [i said check NOT rate...] the one they prefer (of course many iterations are needed per respondent to understand how he choose his bulb) At the end, respondent should also give socio-demo infos such as.

• Gender
• Age
• Income

I would like to see the global effect of attribute. Like what's in general the most important criteria to choose a bulb. I would like to get a kind of econometrical analysis to know if there is difference of choice between the socio demo infos.

I've heard about conjoint analysis or discrete choice analysis methods. In my case which one would be more appropiate?

Which software(s) would also be more appropriate for the overall process (from the design to the analyis)?

Answer 1

You can use a few packages in R, such as described in this answer. You would probably need to used a hierarchical Bayes model for this, so the package bayesm would be required.

Another way would be to use off-the-shelf SaaS software, such as Conjoint.ly, which can handle the required task.

Answer 2

If you want a software package for conducting and analyzing conjoint and discrete choice modeling, Sawtooth is the probably the preeminent provider in the field and hosts regular workshops, conventions and conferences to promote the method and their software. They are expensive though.

If you have Qualtrics, you can also do it in Qualtrics. Qualtrics has more limited functionality that Sawtooth but can do a lot of the basic stuff.

If neither of those are options or if you want to analyze conjoint experiment conducted in Qualtrics or Sawooth yourself, the following tools are available:

Experimental Design:

• cbcTools (R)
• ExpertChoice (R)
• support.CEs (R)
• idefix (R)
• {choiceDes} (R)

Online Surveys:

• formr (R)

Modeling / Modeling Estimation:

• logitr (R)
• apollo (R)
• mlogit (R)
• gmnl (R)
• mixl (R)
• xlogit (Python)
• Stan

mlogit is probably the most typical package to use for analysis in R, but logitr is a newer package I quite like that I think the usage is more intuitive and it's faster and more efficient than mlogit.