Have an idea?

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

CBC conjoint analysis

We collected small sample of 22 records for CBC conjoint analysis (6 attributes by 4 levels design/15 tasks).  I used results from HB estimates and imported into SMRT simulator, then I used the purchase likelihood in the simulation model.

The simulator results are not making any sense. For example, I run a product configuration with average attribute levels and end up with 87% purchase intent. But when I run a product configuration with the best attribute levels, it end up with only 81% purchase intent.

Is there anything we can do to “smooth out” the results or to adjust the results for the simulator?

Thank you
asked Jul 25, 2014 by anonymous

1 Answer

0 votes
The "Purchase Likelihood" model is for use with ACA (when calibration concepts have been asked) or for use with CVA (traditional ratings-based conjoint) when single-concept presentation is used and the logit transform of the dependent variable is in place.

For CBC, purchase likelihood essentially means: "what's the likelihood of choosing this product when it is placed in a set of two products where the other product has average (zero) utility".  Recall that we employ effects-coding in CBC, so the "average" utility is zero.

The purchase likelihood simulation math is as follows:

100*[e^Ui/(1+e^Ui)]

where Ui is the total utility for the product in question.

Regarding what could be going on with your data, you have so few respondents that it would be a good idea just to read your raw utilities into Excel and perform the simulations using the math above for each of the 22 records.  Doing it "by hand" in an Excel spreadsheet will probably reveal to you what's going on with your data set and whether you've made any errors.
answered Jul 25, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (162,290 points)
...