Have an idea?

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

handling discount in MBC

I have an MBC study in which we will be testing a bunch of benefits.  Each benefit attribute will have 3 price levels.

However, we also need to test a discount attribute for bundling benefits together. There is no predetermined bundles. Respondent may create them a la cart. Below are the discount levels.  Discounts for 2 and 4 benefits go in tandem.  

How do I test this?  Do I just show one of the 4 levels of discount on each screen?  Or is the problem more complex?  I need to make sure that I understand the impact of discounting.  Are people selecting more benefits to get a discount?

1.    no discount
2.    if 2 benefits – 5% off, if 4 benefits – 25% off
3.    if 2 benefits –15% off, if 4 benefits – 35% off
4.    if 2 benefits – 25% off, if 4 benefits– 45% off
asked Jan 15, 2015 by ula

1 Answer

0 votes
Wow, that's very interesting!

I'm sure there are multiple ways to do this.  Also, I don't know if the following suggestion is very good.  But, it occurs to me that one could do the following:

Assuming that each product could have from 0 to 4 benefits configured for it, then there are 2^4=16 possible ways to configure a product alternative.  So, respondent answers could be coded as a categorical dependent variable with 16 outcomes.

Now, the discount variable (with its four levels) could be set up as two separate alternative-specific attributes (one for the 2- or 3-benefit case and one for the 4-benefit case).   

The alternative-specific attribute for the 2-benefit discount levels would be a generic predictor (applicable) for any of the 16 categories of the dependent variable which imply 2 or 3 benefits.  The 4-benefit discount attribute would be an alternative-specific attribute predicting into the categories of the dependent variable that imply 4 benefits.
answered Jan 15, 2015 by Bryan Orme Platinum Sawtooth Software, Inc. (191,140 points)
Happy to find your answer to the question I have :) But please tell me if I understand you right. We can design experiment as if it has one discount variable with a "matrix of discounts" depending on size of the bundle, then during the analysis we split it to two attribute specific variables? We could also use two variables from the start, which allows us to simulate more, but will be more demanding during design generation...