Have an idea?

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

Fixed Tasks in CBC


I have a CBC project (dual response none with only 2 scenarios per task to choose from) with one fixed task. This task had several aims:
1- Differentiate logical respondents from illogical as the choice is obvious
2- test how many respondents will prefer the scenario over the none response

My question is:
1- Is it appropriate to include or exclude this task when generating the model (logit-LC-HB) for the project? I understand that exclusion is more appropriate but I am not clear on the reason.
2- Is the item reported (none) in the results corresponds to the utility of the (dual response none option) and I can use it as an anchor for all the other utilities?
3- if 2 is true and I want this anchor to be =0, will it be appropriate to subtract the utility value from all the other utilities or how can I go about this?

asked May 7, 2020 by AMYN Bronze (3,000 points)

1 Answer

+1 vote
Best answer
Answers to your three questions:

1.  I think exclusion is more appropriate.  However, if  adding it in doesn't change the utilities much, I think a lot of folks would be comfortable adding it in as part of the estimation as well.  
2.  It is the utility of the none.  I can imagine how you could use it as the anchor to ONE of the attributes, but not all of them.  I do not know why you would want to do this, because it would not change your simulations or how you interpreted the utilities - all it would do is recenter the utilities of one of your attributes.
3.  How you use it as an anchor depends a little on  how the rest of your experiment is structured.  For example, if you had a labeled design, then you could just subtract the utility of None from the utility of all the alternatives' labels - the utility for the none option would then be zero and for all your other labels would be those label utilities minus the none utility.  If you have brand as an attribute, then you could do this with brand instead - the utility of each brand is that brand's utility minus the none utility and the None utility becomes zero.  For example, if you have brands with utilities of 1, 0 and -1 and a none utility of 2, then you would subtract 2 from each of these to get brand utilities of -1, -2 and -3 and a none utility of 0.   Note the you have used none to re-center the brand utilities, but not any of the other utilities.  I do not know how this will benefit your interpretation of the results.
answered May 7, 2020 by Keith Chrzan Platinum Sawtooth Software, Inc. (103,225 points)
selected May 8, 2020 by AMYN
Thank you very much for your clear and precise explanation.

My none option is (death), and the other 2 scenarios are 2 different health states described using 6 attributed (we call them dimensions & each having 5 or 6 different levels). The first step is to choose which one of the 2 health states is better; then, the second step is to compare the worse state with death. It is kind of similar to BWS CBC but in a sequential manner and still a CBC task.

I want the final utilities of each of the levels in the attributes to be directly anchored to death and death to be equal to 0 so that I can rescale the utilities of all the possible health states on a 0-1 scale were 0 is death & 1 is full health (health states are allowed to be with a negative value).

This is my optimal aim and why I am performing this anchoring/rescaling process.