Have an idea?

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

Interpreting results of study relying on conditional pricing


I am facing a challenge when interpreting the results of my CBC study. The study relies on conditional pricing (price is defined by two discrete attributes and also varies in a -30% - +30% interval). I am computing individual utilities using CBC/HB (price is simulated as linear attribute, as described here: https://sawtoothsoftware.com/forum/798/conditional-pricing-in-hb?show=798#q798), then the results are processed through SMRT to get the zero-centered utilities.

My question are as follows:

1/ Should I include interactions between attributes in my HB analysis to reflect the conditional pricing? Or is it not recommended / relevant?

2/ If yes, do I need to include the interaction when calculating the relative importance of each attribute (using utility range, as mentioned here: https://sawtoothsoftware.com/forum/5872/simulating-interaction-between-categorical-linear-variable)? And how?

Many thanks in advance for your help,
asked Jun 22, 2014 by anonymous

1 Answer

0 votes
My experience is that when using conditional pricing in CBC, interaction terms between the one attribute that drives conditional pricing and the price attribute are often needed.  You can test this with the 2-log likelihood test that is described in the manual.  This is an aggregate logit test.  The chi-square tests from Counting analysis can also be used to test significance of interaction effects.

However, if conditional price is based on 2 or more other attributes, then things certainly get trickier for figuring out which interaction terms to include in the final model.

Importance calculations are not terribly useful in general.  They are totally influenced based on how wide of ranges the researcher includes in the attribute list.  I would rather use sensitivity analsis within the market simulator to see how changes for attribute levels affect the share of preference for a specific product within a specific competitive market simulation scenario.

When our software computes importance scores, it ignores the interaction effects and only pays attention to main effects.
answered Jun 23, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (154,205 points)
Dear Bryan, thank you very much for your prompt answer. I must admit that it is quite complicated to understand exactly what is the "2 log likelihood" test you are referring to. Do you have any link / documentation that I could read to understand more precisely the concept (I went through the help of SSI Web and SMRT with no luck)?

Additionally, how should one take into account interactions when calculating share of choice for example? Is it something to add to the total sum of utility?

Thanks again,