Have an idea?

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

Is it possible to "normalize" relative importances (to make it independent of number of attribute levels)?

Dear Sawtooth team,

I really like the concept of relative importances to give an initial overview of results.
However, as the attributes in my CBC have different numbers of levels (from 3 to 7) the importances are of course difficult to "read", as the importance of attributes with more levels is automatically higher. Therefore I wonder if it is possible to normalize the importances so that it is independent of the number of levels?  Is it possible to do that andand would it even make sense methodically?

Thank you very much for your suport!
asked Jul 29 by Annkathrin (140 points)

1 Answer

0 votes
Importances are inherently normalized to sum to 100%.

The importance of attributes with more levels is NOT automatically higher.  I've done any number of projects even in the past month where attributes with fewer levels had more importance than attributes with more.  

To see this yourself, take a study with 7 levels of price for flights of $125, $135, $150, $160, $175, $190 and $200 and only two levels of a safety attribute (airline never crashes, airline crashes on 1 flight in 10).  Nothing in the math will automatically make price be more important than safety.
answered Jul 29 by Keith Chrzan Platinum Sawtooth Software, Inc. (114,400 points)