Have an idea?

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

Whats the perfect Design Tab setup for a complicated ACBC?

Hello everybody!

I have an ACBC with 5 attributes and 3 levels each. My levels are quite complicated and long, most of them are whole sentences. Unfortunately thats the only way I can do it.

The problem is now, that it takes a lot of concentration and time to get through the survey. I think most respondents won't finish the whole questionaire.

Right now I have the following setup (according to Sawtoth recommendations):

-6 screening tasks
-3 unacceptables
-2 must-haves
-max. 14 concepts in choice tournament
-4 concepts per screening task
-2 attributes to vary from BYO selection

What setup would you recommend in order to keep it as short as possible?

asked Nov 12, 2012 by anonymous
edited Nov 12, 2012

1 Answer

+1 vote
So much depends on how much resolution (precision) you need at the individual level.  If you want to shorten the interview, you can sacrifice some precision at the individual level, but make up for it by increasing the sample size.

Based on our experience with ACBC, I think with just 5 attributes and 3 levels each, you could make your questionnaire much shorter than you currently have it and still obtain excellent individual-level information.  (See justification below, based on our rule-of-thumb).

The general Rule-of-Thumb in ACBC is for each level to appear at least 2x and preferably 3x per respondent.  With your current design specifications, if you run a sample test interview, and then you download the data and examine the data file, you will find the Count Array in the raw data (24 concepts are generated; so this counting array tells you how many times each level of each attribute appears in this one respondent’s design):

levelCnt1:  [[14,5,5],[14,5,5],[14,5,5],[15,4,5],[15,5,4]]

...which tells you that each non-byo level (I selected first level of each attribute) is appearing at minimum 4x.  That is too many, and not necessary based on our tests.

For example, I think you could do:

-5 screening tasks
-3 concepts per screening task
-vary 2 levels per concept from the BYO selections

And, when you run a test survey, you can open the data file and look at the new counts array (each respondent now gets 15 concepts in the near-neighbor plan for the Consideration Phase):


This will be only 15 product concepts in the Consideration Phase for the respondent to manage, rather than 24 as you currently have.

You can see that each non-BYO level appears at minimum 2x and usually 3x.  This would be plenty good in my experience.  If you wanted, you could also use 7 screener questions each showing 2 concepts per screener, for a total of 14 concepts in the near-neighbor plan.  That would also be good for your study, and would show less words on the screen, because your attributes are so long.)

Either of these suggestions will shorten the survey substantially over what you had planned.  Based on our experience, this will give very nice results.
answered Nov 13, 2012 by Bryan Orme Platinum Sawtooth Software, Inc. (160,485 points)
Thank you very much!