Have an idea?

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

Coding and estimation of the none utility in ACBC HB

Dear Forum,

could you please clarify to me, if the following statement is correct regarding the coding and estimation of the none utility in ACBC HB studies?

"In ACBC studies all data from the three sections (BYO, Screener, Tournament) is coded as a sequence of choice tasks in one design matrix, using effects coding. The none threshold is dummy coded, comparable to an attribute level, and estimated as a part-worth utility during the HB procedure. It therefore depicts a parameter of the vector Beta of the individualĀ“s part-worth utilities in the multivariate normal distribution (upper model) and MNL (lower model). The respondentĀ“s indication of must-haves and unacceptables leads to some automatic choices in the tournament and affects the shown choice tasks, but is not directly represented in the design matrix."

Thank you very much!

Best,
Christian
asked Jul 1, 2019 by Chris Berlin Bronze (570 points)
edited Jul 1, 2019 by Chris Berlin

1 Answer

+1 vote
In some ways, that is a correct description.  But, let me make some clarifications.

First, if you did not include the optional "Purchase Likelihood" section in your ACBC study, then the None parameter that is estimated is not really a true "None" utility as from CBC.  Rather, the None parameter is derived from the Screener section, and is the utility associated with the alternative labeled "not a possibility".  Recall that in the Screener section, the respondent is presented a series of product concepts and is asked if each of those is a possibility or not.  That section of the design matrix is coded as a series of binary choices (the product alternative vs. the "not a possibility" alternative).  The None utility is the utility of the "not a possibility" alternative.  As you can see, this really isn't the same as the None alternative from a standard CBC questionnaire.

When the respondent indicates a "must have" or an "unacceptable" rule in the Screeners section, then we look to any of the near-neighbor concepts that have not yet been evaluated by the respondent.  If any of those concepts would be rejected per the "must have" or "unacceptable" rule that the respondent has confirmed, then we make it look like the respondent indeed evaluated those additional concepts and marked them as "not a possibility".  Those inferred choices are added to the design matrix as if the respondent had explicitly seen them and answered "not a possibility".  Therefore, this information is indeed represented in the design matrix.
answered Jul 1, 2019 by Bryan Orme Platinum Sawtooth Software, Inc. (198,715 points)
Thank you Bryan, your quick response is much appreciated.
...