Have an idea?

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

Randomized status quo / opt-out alternative

Is there a way to randomize the status quo in a CBC? In our study we have 2 alternatives (health care treatments) and 1 opt-out alternative (no attendance). Non-attendance is associated with an age-related risk. There are 3 different degrees of risk in the regarded population. Within a choice set with 12 choice tasks, the opt-out should be randomly present 1 of 3 risks in each choice task (as a graphical risk grid).

The idea behind this is: We want to examine the risk someone is willing to take if they do not participate in treatment and at what point the choice probability for or against opting out changes. For example, does an older person make a different decision if this person would face the lower risk of a younger person?

Thank you very much!

asked Jun 7 by Andrew Bronze (1,255 points)

1 Answer

0 votes
If I'm understanding you right, I think if you make this an alternative-specific design and if the opt-out constant alternative is one of the levels of the primary (first) attribute...which means not using the traditional "none" alternative...you can make this work.  This becomes like the "bus/train/car/walk" alternative-specific design example, where "walk" is a constant alternative, but is programmed in our software as one of the levels of the primary (first) attributes.  

If done in this way, the opt-out alternative will come in a randomized position into the task.  If you use fewer concepts than levels of that first attribute, the opt-out alternative will not always be present in the task.
answered Jun 7 by Bryan Orme Platinum Sawtooth Software, Inc. (186,865 points)

thank you very much for the quick answer!

I think I described the problem imprecisely. The opt-out should always remain fixed at the third position. Only the content/ value of the opt-out should vary, not the position.

One of the main attributes describes a certain risk after treatment. This risk varies greatly with non-treatment depending on age. So the opt-out should not show a fixed value but vary randomly across a choice set. It would already be sufficient if one of three different risk values is randomly displayed for the opt-out.

We first wanted to ask about the possibility of implementation. We still need to think about consequences in statistical analysis and interpretation.
Oh, in that case maybe you'll have to do something fancy as a power trick.  If it were my project, I'd probably go the alternative-specific design route, but I'd do some manipulation of Sawtooth's initial design to create independent rotation (relative to the other attributes) across the multiple versions of the "constant opt-out" that I wanted to do.  Those would each become new "levels" of my primary (first) attribute in the alt-spec design.  

You can customize/tweak the design by exporting the design from the design tab to .CSV, modifying the .CSV file, and re-importing the file (while also changing the dimensionality of the attribute list if you modify the .CSV file in a way that expands the attribute/level list).

Whenever you manually do some power tricks like this to modify the design, you should run the Advanced Design Test available within our CBC software to make sure the parameters of interest are all being estimated with good precision...and to examine your 1-way and 2-way frequencies.  As long as you do that testing to make sure the experiment is still sound, you should be in good shape.

Of course, also run through the survey multiple times as if you were a human respondent to make sure the tradeoffs are realistic and reasonable, as you want, for the experiment!
Thank you very much!
So far, I have little experience with alternative-specific designs. And I'm also a bit hesitant to manually change the experimental design. In any case, I will try it out. We are still preparing the pre-study.

Alternatively, is there a way to add a script in the CBC exercise settings in the "None option Text" that randomly displays 1 out of 3 images as None Option?
I'm not savvy like Zachary here is regarding adding script that randomly changes text of the None concept.  But, my worry is how this would get stored (or not) in the data for proper utility analysis.  Even if you were able to modify the text randomly of the None concept, you'd need to then do back-end data manipulation to change the format of the CBC exercise to an alternative-specific design for proper utility estimation.  Since that work would be required anyway, you might as well do it upfront.
Again, to be clear, what you'd need to do is an advanced power-trick, so it's the sort of thing that a typical user wouldn't be equipped to do very easily.  We have a consulting group called Sawtooth Analytics that for a fee can assist users with trickier designs implemented using our software.  You can write keith@sawtoothsoftware.com for a consultation.
thank you very much for your comment! We have also discussed again internally and decided to go another way for the time being, also due to the necessary backend data manipulation you mentioned. Instead, we will try to work with 3 CBCs that differ only in opt-out. The individuals will then be randomly directed to one of the CBCs. This would also give us a fixed reference point in each CBC which is easier to grasp analytically.

Anyway, thanks again for the time you invest in the forum and the valuable advice.

Kind regards,
Sounds like splitting into 3 CBCs can work.  You've probably seen that you can mix multiple CBCs into the same project using Lighthouse Studio...including moving the choice tasks around so that they interleave (first task from first CBC, second task from second CBC, third task from third CBC).  However, for best data estimation, you would want to merge them into a single utility estimation.  For that single utility estimation, you will need to do some data manipulation to create the proper attribute and level structure like we've been discussing.