Have an idea?

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

Monitoring CBC Version

My client wants us to specify lowest read in CBC version. How can i do this? If not can i keep track how many version are asked till now?
asked Oct 29, 2013 by sh_sawtooth (375 points)
'Lowest read in CBC version' is to get the CBC version which have min total completed counts.

3 Answers

+5 votes
Best answer
This is a nice question. As of now, selection of CBC version is automated and we do not have control on that.

I would recommend Sawtooth to give a provision that we can apply some constructed list for selection of version number to be shown. By default, automated feature can apply.

However, we can track the number of versions at run time.

Just create a free format question after first task which stores the value of version shown. Use function CBCVersion() for that.

Next, you can track the counts by either data reports on admin link or by creating a quota question
answered Oct 29, 2013 by Saurabh Aggarwal Gold (30,155 points)
selected Oct 29, 2013 by sh_sawtooth
+2 votes
;) I had same request in past. You have to more tricky here. I created  free format questions to show CBC task and used  lowest read to get CBC version. I was showing levels based upon my lowest read version assignment. For this approach, you have to export your design and create script manually for that. Hope you are using a few version like mine clients otherwise time consuming job. :)

Let me know for more details.
answered Oct 29, 2013 by Rajesh Rana Gold (23,530 points)
+1 vote
The reason why the software doesn't automatically try to exactly balance how many times each questionnaire version is completed by respondents is that it really doesn't matter very much from a statistical standpoint.  If you have 300 versions of the questionnaire (the default in the software), and some of the versions are filled out 3x by respondents and some only 1x or 0x, then for all practical purposes, the statistical efficiency of the design overall is not much different than if each version was completed by exactly the same number of people.  You can test this out with synthetic data to prove to yourself.

Even with a design with only 4 versions/blocks, it doesn't harm things very much (from a statistical standpoint) if one version is completed 2x as many times as another version.  These situations are very robust to imbalance in the number of times each version was completed.

(These statements are made under the assumption that Sawtooth Software designs are being used, where within each version of the questionnaire, each level appears nearly an equal number of times and each level appears with every other level nearly an equal number of times.  In other words, the designs are nearly balanced and near-orthogonal within each questionnaire version/block).

So, doing so much work to try to exactly balance the number of times each version is displayed is putting a lot of effort into something that typically does not matter much at all for the precise estimation of utilities under aggregate logit, latent class, or HB.
answered Oct 29, 2013 by Bryan Orme Platinum Sawtooth Software, Inc. (156,655 points)
edited Oct 29, 2013 by Bryan Orme