Have an idea?

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

Intersection of attribute levels in ACBC

Dear Sawtooth community,

For my ACBC I got an attribute with the following concept for the levels:

Level 1: feature A
Level 2: feature B
Level 3: feature A+B

Just wanted to make sure that this does not pose any problem in terms of statistical evaluation? All 3 levels are mutually exclusive right?

Thank you.

Best regards,
asked Dec 16, 2020 by danny Bronze (1,260 points)
retagged Dec 16, 2020 by danny

1 Answer

0 votes
In conjoint analysis (whether CBC or ACBC), you can create a 3-level attribute as you suggest.  However, make sure that you will not need to analyze (market simulate) what happens if lacking feature A & feature B.  The  attribute could thus take on 4 levels:

Level 1: No A, No B
Level 2: A
Level 3: B
Level 4: A&B
answered Dec 16, 2020 by Bryan Orme Platinum Sawtooth Software, Inc. (184,140 points)
Understand, thank you for the tip! My goal is to identify the relative importances of all attributes and the part-worth utilities of the levels. Hence I guess I would not need to use the market simular, and can go ahead with my suggested 3 levels?
It seems strange to me to take all the time to do a conjoint analysis and stop with utilities and importance scores.  The market simulator allows you to turn utilities and importances into predictions of market choice, which is what better guides marketing/strategy decisions.

Furthermore, how can you assess the importance of having features A&B or not without including the level where they both are not present?

Also, have you thought about creating two separate attributes, to assess the importance of these features separately?

Attribute A:

level 1: No A
Level 2: A

Attribute B:
Level 1: No B
Level 2: B
Thanks again for your quick reply Bryan!

We have to have those features in 1 attribute otherwise we end up having too many (we already have 7-8 attributes). In this case we have no other choice.

Let me add some details to help you better assess our design.

Please note that feature A is considered basic, a feature that every product of the investigated product type has. With the attribute we want to understand whether users would still prefer to have B as an alternative feature, or a combination of both. (Basically we want to test the hypothesis that feature A is truly considered sufficient as compared to the other feature B). Random example to show you how I mean it:
L1: Basic encryption method (A)
L2: Enhanced encryption method (B)
L3: Option to use both encryption methods
(Enhanced method does not build upon the basic one, they are different.)

So in practice, for our investigated product category, A is the "minimum" hence there is no "none" option. Now you might argue that since A is a must-have, it should not be included in the feature at all and we should just have two levels L1 "Feature B available" and L2 "Feature B not available". In our case it would be however interesting to compare the part-worth utilities in the attribute to directly see how big or small the utilitiy gaps are in between those 3 levels. Does that make sense to you?

Good point regarding market analysis. That's something we should consider and further investigate. But in the end this is an academic study and hence we are not trying to predict any market shares. For us it's all about the preferences in feature selection.