Have an idea?

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

Respondent-specific MBC exercise

I’m kind of stuck on a project design and am wondering if MBC could be a solution.  Client wants to test a large number of feature concepts, with hope of understanding the best bundles of features.

Here is what I’m thinking...

- 40 “innovation” concepts

- Respondents rate all 40 using a rating scale

- Based on how the respondent rates the concepts, the top 20 concepts are selected to be shown in their respondent-specific MBC exercise.  Thus, there are:
   --  20 “selected” concepts
    -- 20 “reject” concepts

- MBC design uses four “price” levels, and accounts for all 40 features:
   -- Levels 1-3 = $4, $6, $8
   -- Level 4 = not shown

-Respondents complete ~8 menus; n = 1,000

- Post hoc, the following are simply coded as missing data:
   -- All 20 reject concepts
   -- Selected concepts when “not shown” in the menu

asked Sep 24, 2014 by anonymous

1 Answer

0 votes
My experience is that if you are going to be using the logit or HB-logit modeling within our MBC software, such a design gets very tricky due to all the availability terms.  If you haven't done this sort of model before, make sure to generate dummy data and play with it until you feel confident you've got the proper model specification and reasonable standard errors on the parameters.  Another option to keep the model simpler is just to assume that each respondent saw all items and just never picked the items that were screened out (that saves you the hassle and extra parameters to estimate of dealing with the availability terms!)

Your approach above also might work decently as just a counting analysis experiment, where you think of each product built as a BYO (build-your-own) optimization for each respondent for each task.  If you kept your analysis focused on counting results (combinatorial counting of configured outcomes), you might avoid a lot of pain and actually end up being able to present results (Counting only) to the client that are easy to explain.  The notion of changing prices between tasks could be just a good excuse to rationalize for the respondent why we are asking them to complete multiple MBC tasks (while obtaining more data from each respondent).  You might consider just ignoring the price variables during counting analysis of most commonly-configured bundles.

Obviously, you could just tally up the most commonly configured products.  But, you could consider market segmentation issues as well (most commonly configured products within different market segments).
answered Sep 29, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (191,140 points)