Have an idea?

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

counting analysis of a cbc alternative-specific design

Hello everyone,

I have a little question regarding the counting-analysis of a CBC with alternative-specific design.

When an attribute has more levels than the number of concepts in a task is it possible to divide these probabilities by their sum for making comparisons across the attributes?

Example (each choice task including 3 concepts and a "None" option):

attribute levels of a conditional attribute X:
X1: 0.1 => 0.1/1.2 => 0.09
X2: 0.4 => 0.5/1.1 => 0.45
X3: 0,3=> 0.3/1.1 => 0.27
X4: 0,3=> 0.3/1.1 => 0.27

Thanks a lot!
asked Jan 14, 2020 by anonymous

1 Answer

+1 vote
Lighthouse Studio is already making the counting analysis comparable level to level.  When we report the percentage for level X1, we're taking the number of time it was chosen by respondents and dividing it by the number of times it was available to be chosen.  This helps account for situations where a level shows up twice in a question, doesn't show up at all, or when you have a natural imbalance due to prohibitions.
answered Jan 15, 2020 by Brian McEwan Platinum Sawtooth Software, Inc. (55,670 points)
Thanks for your answer Brian

But how should I then interpret the following case (3 concepts + none)?

attribute levels of primary Z:
Z1: 0.3
Z2: 0.2

What was then chosen in 50% of the other cases? The None option was only chosen at 11% in my CBC.

Your manual says: "When an attribute has fewer levels than the number of concepts in a task, the sum of its proportions will be lowered.  When making comparisons across attributes it is useful first to adjust each set of proportions to remove this artifact.  One way to do so is to divide all the proportions for each attribute by their sum, giving them all unit sums."

Counts will be lower than 100% if you ever had any overlap in a question.  For example, if one task showed Brand A twice, and the respondent chose one instance of it, Brand A would show up as 50% (1 choice divided by 2 showings).  So, overlap in a question and a none option will make a counts report for an attribute be lower than 100%.
I understand the computation, but I am interested in the total number of times Z1 was selected in relation to the total number of choice tasks.

If z1 is shown twice in a choice task and one z1 is selected, then z1 still “wins“ in that choice task (even if one of them is not chosen).

So is it wrong to give them unit sums? And to say that in the case where none-option was not chosen, z1 was chosen in 60% (0.3/0.5) and z2 in 40% (0.2/0.5) of the cases?

Thank you
I'm not sure if it would tell you anything much different than our counts analysis.  You're just changing the denominator, which would be a constant change? But you can export out the design and answers to a CSV file in the Data Management area (Exports tab) and probably whip up your own counts analysis without too much trouble in Excel to count number of choices relative to number of tasks.