Have an idea?

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

Does the attribute order play a role in the design efficiency?

Hello,

I am currently programming an alternative-specific CBC exercise, with 6 attributes, and quite a few prohibitions between some of the attributes. I have already programmed this, and the design was not deficient (not as efficient as one would probably like, but no warnings came up at any point). I was then asked to change the order of the attributes so that for instance attribute 3 would appear first, followed by attribute 4 and then attribute 2 etc. I made these changes and checked that the prohibitions are correct (and have not changed since my last working programmed version of the exercise), and generated the design (300 versions, random generation method).

To my surprise, the design was deficient however. Is there a reason why this would happen? Are there any specific circumstances where this might be more likely to occur?

Thanks!
Orestis
asked Jan 18, 2012 by anonymous
retagged Sep 13, 2012 by Walter Williams

1 Answer

+1 vote
Which design test were you looking at the first time?

We have seen some instances in the past where order can have an effect for cases where the design is close to deficient and the attribute order is then changed. The basic test that runs when you generate a design is particularly susceptible to this. The designer itself can also be impacted if, in its randomized search of the design space, it gets stuck and can't move forward. If it worked before, you can try a different design seed or another design method to see if it can work around the dead end it hit. However, I suspect that your design has some issues, so make sure you test your design thoroughly.

If you continue to experience the problem with the one design and not the other, we'd like to take a look at it. There may be a minor algorithmic issue that we can program around to solve the problem. If you need to field quickly and are satisfied that the original design will work, you can always export the design (to a .csv) using the original attribute order, switch the order of the columns using Excel, and then import the newly-ordered design into a copy of your study where you've also reordered the attributes.
answered Jan 20, 2012 by Aaron Hill Silver Sawtooth Software, Inc. (8,065 points)
...