Have an idea?

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

Can you augment designs in Sawtooth?


I'm running a study, and I would like to periodically augment my design to reflect new priors based on the responses I've received to date. So, for example, when I've received 1/3 of my responses, I'd like to augment the design to select a new D-efficient set of tasks for future respondents given the set of tasks and answers provided in prior tasks.

I'm also considering collapsing two of the levels for one attribute into one, and I'd like to know if I could do this and then augment the design. I'm not concerned about conducting individual-level utility estimation, so the fact that some respondents saw the two levels and some saw one isn't all that important from my perspective. Also, I'm only doing single comparisons of a product profile versus a null option, so I haven't compared those two levels of the attribute head-to-head.

Can you augment designs in either of the ways described above in Sawtooth? Or would I need to use another program and then import the augmented design?
asked Jun 23, 2017 by anonymous

2 Answers

0 votes
This kind of falls into the realm of possible but not built to be automated in the software.

For the first part, Lighthouse Studio does not incorporate utility balance when generating designs, so there is no way to set priors before creating the design.  The second gotcha is that when you create a CBC design in the software, the design has a unique ID.  This ID is recorded in the respondent's data record.  Only one design can be present in a survey at a time, so if your first 100 respondents saw design 1234 and the second 100 saw design 5678, the software will not be able to analyze all 200 people together by default because they saw different designs.

You can technically get around this by changing the design mid study, then combining your two designs into a single design and changing the design ID and version in the data, but it would be a pain.

The second part about combining levels together is again doable, but it's all or nothing.  Designs can be exported from Lighthouse Studio to a CSV spreadsheet, modified where you change all the levels from, say, 4 to 3, and re-imported back into Lighthouse.  You'll have to deal with the design ID issue mentioned above because importing will create a new design, but the answers will all still match up since you are only changing one attribute.
answered Jun 23, 2017 by Brian McEwan Gold Sawtooth Software, Inc. (37,410 points)
0 votes
What I have done on occasion is to create my design, in our software or outside of it, then model the first 100 or 200 respondents.  Those utilities become priors for making a new "Bayesian" design in a design software package that uses priors (e.g. N-gene).  I then create a new study file in our software and import the new design (again, the one made outside of our software).  

For analysis you can export the conjoint data as a single CSV export from each of the two study files, combine them and then run the analysis in CBC/HB.

As Brian says, if you really need to use a Bayesian design strategy, this is possible, but because it's not built into our software it requires some work-arounds.
answered Jun 23, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (50,675 points)
reshown Jun 23, 2017 by Walter Williams
Great. Thank you both!