Have an idea?

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

Using Probability Scale for best-worst conjoint (profile case MaxDiff or best-worst Case 2)

Hello and have a lovely day,

I understand that the Probability Scale is mainly useful for Best-Worst Case 1 (Object Case Best Worse). Still, I wanted to check if I can use it as in Case 2 (best-worst conjoint (profile case MaxDiff, or best-worst conjoint) to calculate the ratio and relative importance for each level then by adding the values of the levels in a single dimension I will get the importance (or weight) of this dimension. Repeating this for all dimensions I then can get the importance of each Dimension on a ratio scale and can confidently say that a dimension of a score of 40 is twice important as a dimension of a score of 20 & vice versa. Will this be correct? if not then what is the best way to calculate the ratio or relative importance of each dimension & level using either the standalone software and the Lighthouse Studio 9.8

Thanks a lot
asked Feb 10 by AMYN Bronze (2,980 points)

1 Answer

+1 vote
Best answer
You can compute importance scores for BW-case 2.  But you can do so from the raw utilities, which will make more sense than computing them from the probability-transformed raw utilities.   You don't need the probability-transformed utilities to get importances that are ratio-scaled.
answered Feb 10 by Keith Chrzan Platinum Sawtooth Software, Inc. (90,475 points)
selected Feb 11 by AMYN
Thanks for your reply Keith,
Mind me if I asked how?
is it through direct addition or there are a certain transformation that I need to go through?
You calculate importance just line in conjoint analysis.  For each attribute subtract the utility of the worst (lowest-utility) level from that of the best (highest utility) level.  This creates a range for each attribute.  Now divide the range for one attribute by the sum of the ranges for all attributes, and that's your importance.  Again, this is the same calculation as for any other conjoint analysis.
I thought that the Best worst results give a better estimate of dimension importance than conjoint because all the levels are estimated on a common reference level and not each level is estimated within their own dimension thus the importance results as you described will be meaningful and comparable.
The results then can be interpreted as ratio or just interval (can we say that 2 is double 1 or jusr 2 is better than 1 ?)
Best-worst case 2 ("best-worst conjoint") DOES allow you to make cross-attribute level comparisons because the attributes are all on the same interval scale.  But the scale is still interval, so you cannot do ratio level operations on the utilities themselves.  BW-case 2 does not necessarily give you better importance estimates.