Have an idea?

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

Can I compare path worth utlities from HB across respondents


first of all - I would like to appreciate all the help I got here so far. However, another question is striking my mind at the moment and I would be more than happy to receive some hints on how to proceed.

What I'm trying to do is investigate whether a certain personality trait is associated with the part worth of an attribute level:

As an example, lets assume that we want to find out the relevance of "pay" for job choice (besides other factors).

The attribute "pay" shall be categorial with four articial levels (1500 $, 2000 $, 2500 $, 3000 $).

Besides the ACBC, we surveyed a person's career ambition (the degree to which a person consideres their career when choosing a job).

The question I would like to answer is the following: Is the path worth that is attributed to the 3000 $ pay (highest pay) correlated with the career ambition (in other words: Does a high career ambition relate to a higher path worth for 3000 $?). However, in order to do this, the path worths need to be comparable across respondents - is that right? Will the ZC Diffs do the trick? Or is that not possible to do?

Thank you for a reply on this issue.
asked Aug 1, 2018 by Daniel

1 Answer

0 votes
As you suggest, the standard (non-Bayesian) approach to answering this question relies on using the zero-centered diffs (normalized) utilities.  These are the utilities that are rescaled such that the average difference between best and worst levels (within attributes) is 100 points.  Zero-centered diffs utility normalization is an approach to put each respondent on roughly the same scaling (to try to avoid problems in interpreting raw HB logit-scaled utilities across people).

You can use the standard F-test found in cross-tab packages or SPSS (etc.) to compare the utilities between groups of people on the zero-centered diffs.
answered Aug 1, 2018 by Bryan Orme Platinum Sawtooth Software, Inc. (198,715 points)