Have an idea?

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

Levels randomization for traditional full-profile CBC Design


I was trying for a while now to randomize the levels for my CBC questions. However, I have not managed to do it.

Could you please help me with that?

Thank you,
asked Dec 5, 2017 by Madalina
You want to randomize the order the attributes are displayed in your CBC, is that correct?  What version of SSI Web / Lighthouse Studio are you running?
Yes, Zachary, that's right. I am on the last version of Lighthouse Studio.

1 Answer

0 votes
Randomizing levels already occurs automatically with the CBC designer. Usually, people have difficulty UNrandomizing them. To UNrandomize them, go to the Design tab and look for "Concept Sorting". (If you're on v9, later versions have moved this under the "Additional Settings" button.) You can UNrandomize them by natural level order (would always show Brand A, then Brand B, then Brand C, for instance), or by randomized level order, where respondent 1 might always see Brand C, then Brand A, then Brand B, while respondent 2 always saw Brand B, then Brand C, then Brand A.

Occasionally, users will want to import a design and have the software randomize it for them. Unfortunately, there is not a setting in the software to randomize the concept order, so you will have to do this prior to importing the design.

Best regards,
answered Dec 6, 2017 by Aaron Hill Gold Sawtooth Software, Inc. (10,695 points)
Hi, Aaron! Thank you for that. If you look at the question like that, yes it is true. What I meant was how can I randomize the attributes, so their randomised levels will not show up in the same order for each choice set.

Kind regards,
Attribute randomization is found in the same area in the software. However, the randomized attribute order is kept constant for each respondent across all their choice tasks. Respondent 1 will see attributes 1, 3, 5, 4, 2 for all tasks, respondent 2 might see attributes 5, 3, 1, 2, 4 for all tasks, etc. There isn't a way to randomize the attribute order per choice set because we feel this makes the tasks too difficult for the respondent. Any statistical efficiency gained by controlling for order effects is more than offset by increased response error, higher dropout rates, etc.
Alright, I can see now why that is not possible. Thank you very much!

Kind regards,