Have an idea?

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

How to interpret "Non" partworth in CBC/HB?


How should we interpret partworth of non option specially with respect to other attribute levels? Below is an example

Att. Levels    Part-worth    95% Confidence interval
Color-Red    1.660    0.308
Color-Blue    1.117    0.172
Color-Yellow    -0.083    0.267
Color-Black    -2.693    0.244
Dist.-Belk    0.786    0.085
Dist.- Curacao    0.266    0.075
Dist.-XYZ Website    0.063    0.062
Dist.-Ebay    -1.114    0.122
Price-209    0.812    0.061
Price-239    0.325    0.054
Price-269    0.103    0.048
Price-299    -1.240    0.097
Non-option    2.293    0.574
asked Aug 8, 2017 by Robin59 Bronze (565 points)

1 Answer

+1 vote
Best answer
The "none" parameter in CBC models is the utility of not choosing any of the concepts in a choice set.  Its precise meaning depends on how exactly you've worded it, but a general "none" wording means that the respondent prefers not to have any of the concepts offered.  In a simulation, which is the best way to interpret ANY of the utilities, when you allow "none" into the simulation, the % choosing it are those who choose not to choose any of the products you've specified.  Again, these could be people who choose to stay out of the market, to purchase elsewhere, to purchase later, etc., depending on how you've worded the "none" alternative in your questionnaire.
answered Aug 8, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (50,675 points)
selected Aug 10, 2017 by Robin59
Thanks Keith. I know that you can not compare level from one attribute to a level from another attribute. But is there any way to interpret partworth of non with respect to partworth of other attribute levels? ( please see above example)

Well, your None utility is pretty large relative to the other utilities, so you can see that you'll need to have a product with a sum of utilities that's greater than 2.293 (on average) to tip the scale and be appealing enough that the average respondent prefers it to not having one of the products at all.  That "on average" part can be really deceptive, though, which is why it's a much better idea to see what happens when you run simulations.