Have an idea?

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

Measuring overall preference with summation of utilities

For a standard CBC we almost always have a none option and then we use a partial vote to calculate preference for different profiles.  We recently did a message testing CBC with no none option and created a benchmark profile as the comparator.  And that worked fine.

Which made me wonder, if I have no none option and I don't want to use a benchmark competitive product - I just want to measure overall desirability of each profile; could I just use the straight sum of the zero centered utilities?  Something tells me that might be sidestepping the fundamentals of how they were derived in the first place - based on a choice-based format, so there needs to be some form of comparison profile to make it a valid calculation of preference?
asked Mar 29, 2016 by stevetlg.com (480 points)
retagged Mar 29, 2016 by Walter Williams

1 Answer

0 votes
You may indeed sum the part-worths to get a relative preference score for different product concepts.  But, the real trick is if you get a score of 180 for one concept and 90 for the other, what does this mean?  

Certainly we know that the first product is more preferred on average than the second.  But, is it twice as preferred?  Only a little bit more preferred?  

Part-worth utilities (whether raw or zero-centered) are not ratio scaled, so one cannot say that a total utility of 180 is twice as good as a total utility of 90.

Because researchers often want to be able to state if one product is twice as preferred as another (as opposed to just saying one is better than the other), researchers are usually looking for analysis approaches that lead to probability (ratio) scaled outcomes.

Running market simulations based on the RFC or Logit rule for CBC data can lead to such ratio-type interpretations.  But, summing the part-worth utilities (which are interval scaled) does not.
answered Mar 29, 2016 by Bryan Orme Platinum Sawtooth Software, Inc. (179,015 points)
Summing raw utilities by individual respondent