Have an idea?

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

estimated utilities

Hi,

So I understand that attribute utilities are by default zero-centered. But I am not sure if I interpret it correctly. - Is this assumption correct?

" an attribute utility that is close to zero shows that respondents on average are rather neutral regarding the attribute. In contrast attribute utilities that are deviating strongly from zero, show that respondents have a strong ( positive or negative) opinion on the attribute. "

Do attribute levels with a negative utility always suggest that they are negatively perceived. Or only that they are the least favorable attribute level compared to the other given attribute levels?

Thanks
asked Apr 19 by Charlotte

1 Answer

+2 votes
If you are using our CBC or ACBC software and if the utility estimation settings are at their default, then the raw utilities within attributes are indeed zero-centered.  The rescaled utilities (zero-centered diffs) coming out of all our conjoint systems (including ACA and CVA) are also zero-centered within attributes.

But, a negative doesn't mean people feel negatively about a level.  It just means they have relatively lower preference for a level relative to the other levels within the same attribute.   All levels in reality could be positively viewed; or on the other hand, negatively viewed.  It's just that some levels are relatively preferred more than others within the same attribute.

Standard conjoint methods do not have a way to measure if all levels are disliked or liked within an attribute; only the ability to estimate relative preferences within the same attribute.
answered Apr 19 by Bryan Orme Platinum Sawtooth Software, Inc. (164,490 points)
I could add that there is a flavor of conjoint analysis that does not lead to zero-centered utilities, where you can directly compare each level of every attribute to every other level of every other attribute in terms of utility.  That's "Best-Worst Case 2" or so-called "Best-Worst Conjoint Analysis" (a Jordan Louviere invention from the 1990s.  I was skeptical of it until I read papers on it by Louviere et al., Chrzan et al., and then did some of my own R&D and I also wrote a white paper on it.  https://www.sawtoothsoftware.com/support/technical-papers?id=1369
...