Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

How to design a conjoint with not mutually exclusive attribute levels?

Hello everyone,

I want to find out, which features of a product are most important to potential customers. But the different levels/features of the attributes are not mutually exclusive (E.g. Attribute: Optional Features. Levels: Sunroof, GPS System, Extended Warrenty).

Sawtooth gives 2 options how to deal with this issue:
a) either create an attribute with all potential combinations of these features or b) formulate three separate attributes each with two levels (feature exists and feature does not exist). [Taken from: https://www.sawtoothsoftware.com/download/techpap/formatt.pdf]

I understand that with option a) one can measure the interactions between the features, but does it make sense to do a conjoint analysis for a problem like this? Besides the price I would only have non mutually exclusive features. Therefore I'd either have many attributes with only "exists" and "does not exist" level or I'd have many levels per attribute with different combination of the features.

Which option is better regarding the validity of the results? Which is easier reagarding theevaluation and interpretation of the results?

I'd really appreciate your help!
asked Oct 28, 2013 by BowTie

1 Answer

0 votes
Conjoint analysis indeed can work for a series of binary attributes (series of exist/doesn't exist attributes).  You can just estimate the utility of each "on" vs. "off" level and use them in an additive market simulation model (main-effects only model).  But, you'll typically overpredict the preference for products involving lots of "on" levels and underpredict the preference for products involving lots of "off" levels.

If you include interaction terms in your models (which our CBC software supports), you can greatly reduce the modeling problems.  You can even decide to collapse (say) three binary attributes into a single attribute with 8 levels, etc.  These are all options for modeling (back-end analytical steps) within our software.

But, there are other more advanced ways to deal with a series of binary features, such as was recently discussed by Kevin Lattery at our conference earlier this month.  His solutions would require hiring a high-level consultant with considerable modeling expertise and experience in this area.
answered Oct 28, 2013 by Bryan Orme Platinum Sawtooth Software, Inc. (173,090 points)
Hi Bryan,
thank you for your answer!
Since we don't have the ressources to hire a high-level consultant, it sounds like having three binary attributes in one attribute is the best way.
Modelling of not-mutually exclusive attributes