Have an idea?

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

The data is ill conditioned because of prohibition, what to do?

Hi,

I have a small question. At this moment I'm designing a CBC. I want to measure the influence of reviews on the purchase behavior of meal box subscriptions. However, I came across a small problem.

I will try and explain it to you. First I have the attribute star review. This can vary from 1, 3, and 5 stars. The second attribute is the review length. This can vary from short, long, and very long. As you might assume, a 1-star review text is not the same as a 5-star review. So, while I have 3 levels within review length they vary within which star rating they get. How should I transform this into Sawtooth, since a 1-star review can only be negative and vice versa.

However, I've used prohibitions which results in the following warnings:

**Warning: You have specified prohibitions/alternative-specific rules between two or more attributes.  You cannot automatically estimate an interaction effect between two attributes when a prohibition or alternative-specific rule is in place.

**Warning: We strongly encourage you to further investigate the efficiency of your design prior to fielding this study.

Also when testing it claims that the data are ill-conditioned and not showing errors for the interaction variables:

          Std Err    Attribute Level
  1       *******    1 1 1 star
  2       *******    1 2 3 stars
  3       *******    1 3 5 stars

  4       *******    2 1 Short - 1 Star
  5       *******    2 2 Long - 1 Star
  6       *******    2 3 Very long - 1 star
  7       *******    2 4 Short - 3 Stars
  8       *******    2 5 Long - 3 Stars
  9       *******    2 6 Very long - 3 Stars
 10       *******    2 7 Short - 5 Stars
 11       *******    2 8 Long - 5 Stars
 12       *******    2 9 Very long - 5 Stars

 13       0.02919    3 1 0 helpful votes
 14       0.02925    3 2 20 helpful votes
 15       0.02906    3 3 50 helpful votes

 16       0.02917    4 1 0 helpful votes
 17       0.02935    4 2 10 helpful votes
 18       0.02937    4 3 20 helpful votes

 19       0.02049    5 1 Yes
 20       0.02049    5 2 No

 21       0.02932    6 1 Foreign
 22       0.02917    6 2 Domestic
 23       0.02925    6 3 Local

 24       0.03849    NONE
asked Nov 18 by Olle

1 Answer

0 votes
Olle, yesterday I mentioned conditional display as one way around this problem, but another idea that would also work, now that I see more about your situation, would be to build this as an alternative-specific design.  Both of these options are supported in our software and you can learn more about them in the documentation or the help menu.  You can also contact support@sawtoothsoftware.com for help with this common problem.
answered Nov 18 by Keith Chrzan Platinum Sawtooth Software, Inc. (97,375 points)
...