Have an idea?

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

How reliable is "Purchase Likelihood" estimates in the market simulator using CBC-HB estimates?

Hello everyone,

I have some questions on the "Purchase Likelihood" estimates in the market simulator.
When I combine service options using  interaction HB estimates and simulate, it shows different estimates based on the selection criteria.

(a) Does the purchase likelihood take into consideration the "Yes" the respondents chose for dual response "none" option?

(b) How reliable are these "Purchase Likelihood" estimates? Can I use these estimates instead of "Share of preference" to determine the likelihood of of a particular Service combination of interest?
For example, the standard error for a particular Purchase Likelihood standard error is 8.2% for a particular price point.

I had tried to find the WTP for a specific attribute and level by taking into consideration competing products as suggested by  Keith Chrzan in one of my questions before.

However, this only provides an estimate of the price point of a specific attribute. To determine the likelihood of purchase at a particular price point for a combination of attributes, I am bit confused if I can take that same price point for the combination.

Kindly advise,

asked Apr 17, 2018 by sacharya (470 points)

1 Answer

0 votes
Best answer
Purchase Likelihood equation is:


, where U is the total utility of the product concept.  

It does not take into account the respondent's None answers.  And, unless you've done something special to scale the Us to be predictive of purchase intent after submitting to the equation above, it is only strongly correlated with stated purchase intent and not a prediction of the level of stated purchase intent.

To compute the likelihood of picking the product concept versus the None, you would use the following equation:


, where U is the utility of the product concept and NoneU is the utility of the None concept.  Our market simulator will do this automatically for you if you use the "Share of Preference" setting and specify one product in the simulation (with the None share turned "on").

To compute the "demand curve" for a specific product combination, use the "Share of Preference" or "Randomized First Choice" simulation model.  Put your test product in competition with all the relevant competition and the None alternative that would be expected to be in the base case competitive environment.  Run the simulation multiple times, changing the price of the test product each time and holding the price of the competitive products constant.
answered Apr 17, 2018 by Bryan Orme Platinum Sawtooth Software, Inc. (198,315 points)
selected Apr 17, 2018 by sacharya