Have an idea?

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

ACBC utilities for attributes and levels dropped via constructed lists

Hi, I'm planning to manipulate ACBC choice data  and then analyze it using CBC/HB. I realize that the information of attributes and levels dropped via constructed lists is missing then. Is there a way to recreate your assumptions (per manual) via coding? Or can I just manually set the corresponding utilities to Zero ex-post estimation?
My goal is to test predictive power with individual screening task holdouts. Please advise on any other possible way of testing models against each other for ACBC as the given fit measures do not reflect the predictive power reliably from my experience.
asked Sep 2, 2016 by alex.wendland Bronze (2,410 points)

1 Answer

0 votes
Best answer
Hi, Alex.

When you export the .CHO file from SSI Web (now called Lighthouse Studio), the information about dropped levels within attributes should be written out to the file as long as you tell the software that you'll be treating dropped levels as "inferior".  Double-check that by opening the .CHO file and looking for the series of paired comparison tasks comparing each item to a threshold item that is invented in the design matrix for each attribute (involved in dropped levels).

But, the dropped attributes are of course just 0s in the design matrix and you'll want to do something post hoc to the part-worths for them (as you suggest) to make these attributes get less (or zero) importance relative to the attributes that were taken forward into the ACBC survey for the respondent.
answered Sep 2, 2016 by Bryan Orme Platinum Sawtooth Software, Inc. (179,015 points)
selected Sep 2, 2016 by alex.wendland