Have an idea?

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

Attribute importance when using Conditional Pricing


I have a question regarding a CBC conjoint study, done on telecom market, where conditional pricing was used with one of the attributes.
The attribute conditioned with price has the highest importance, higher than price itself (clearly this is the effect of price). Is there a way to disentangle the importance of this attribute from the one of price? Is there another way to compute attribute importance in this case? Can we use the importance of the other attributes (which are not conditioned by price), but not the one of the conditioned attribute? Should we interpret the market simulator differently when using conditional pricing?
What is the recommended approach?

Any ideas and recommendations are more than welcomed :)

Thank you,
asked Jan 19, 2021 by Simona Moisescu

1 Answer

0 votes
You are correct that when conditional pricing is done, the utilities for the attributes are no longer independent of one another.  Thus, the importance scores will also be wrong.  For example, if you set a conditional price between brand and price, then the brand utilities each include an intercept related to the average difference in price levels associated with each brand.  Market simulations (predictions of choices) are not bothered by this (as the parameters were developed for maximum likelihood fit) and are correct for predicting choices.

The only way to disentangle the results of a conditional pricing design for the purposes of computing independent attribute importances is to re-estimate the model using something outside of Lighthouse Studio, such as our standalone CBC/HB system.  And, you'd need to do some extra coding work to prepare the CSV data file for HB MNL estimation.  Specifically, you'd need to collapse the Price attribute into a single column for linear price estimation, where that column contained the actual prices shown to respondents (from the conditional pricing table).  And, you'd want to rescale those prices in that single X column of the design matrix to be in the singles of units.  For example, if your prices shown to respondents were $1000, $2000, $3,500; then you'd want to scale the prices in the X column to be like 1.0, 2.0, 3.5.  Otherwise, convergence in HB doesn't work as well.  Once utilities are estimated in this way, the brand utilities are independent of the price utilities (the price slope).  And, importance scores could then be computed.
answered Jan 19, 2021 by Bryan Orme Platinum Sawtooth Software, Inc. (196,215 points)