Have an idea?

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

How to get percentages of parth-worth utilities for the levels?

Dear Sawtooth Software Team,

I like Sawtooth very much, even if I am a little overwhelmed by the possibilities. I have already read really a lot of literature in last weeks (by Sawtooth and by other authors) and yet I have three basic questions …

1)    How is it possible to obtain percentages for the importance of the part-worth utilities for the levels after conducting a HB within Lighthouse Studio (see attachment). For the attributes, the analysis directly provides the importance. Does that even make sense?

2)    I am not allowed to compare the parth-worth utilities within one attribute, right? To say something like „level A from attribute 1“ increases the benefit almost twice as much as „level B from attribute 1“ is wrong, isn´t it? Does this mean that the part-worth values / average utilities in Lighthouse only provide me with information about which level leads to the greatest increase in benefit and which to the least benefit? Further interpretations are not permitted?

3)    One additional question: In Chapter 9 from Orme I read that the effects coding is used here and that all levels together result in 0. Is this the same when someone is talking in literature as normalization and scaling?

Best wishes
Tobias Dapper

A tip from my side:
The hands-up tutorials within Lighthouse help were great and extremely helpful, so I had little trouble creating a conjoint analysis. If there were an exemplary analysis for this tutorial afterwards with exemplary results, I would certainly be able to better understand the entire process of evaluation and interpretation. Just like Chapter 9 by Orme but with more details in relation to the interpreting and within a global setting / use case.
asked Mar 25 by Tobi

1 Answer

0 votes

Replies to your three questions:

1.  It really makes no sense to express the level utilities as percentages.  

2.  You are correct about the utilities themselves - higher number means better, within an attribute.  You'll find that the kind of statements you're wanting to make will work better if you're talking about simulator results than about the utilities themselves.  

3.  No.  Effects coding is a particular type of dummy coding that centers utilities on 0.0.  It is a carryover from the design-of-experiments thinking involving ANOVA, which historically also used effects coding.  It is not the same as standard normalization like Z scores.

And thank you for the suggestion.
answered Mar 25 by Keith Chrzan Platinum Sawtooth Software, Inc. (90,575 points)