Have an idea?

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

Anchored MaxDiff what rescaling method to use?

What's the difference between probability rescaling and zero anchored interval scaling. Which of those rescaling procedures would you recommend under which circumstances? I see, that the ranking of the attributes is quite similar between both rescaling methods but the anchor becomes more important using zero-anchored interval scaling.
asked Nov 8, 2017 by RafalNeska (450 points)
In my experience I've found that if people are familiar with the scores that sum to 100 for a normal MaxDiff, the interval-scaled scores are easier to understand.  It might be because I usually describe the rescaled scores as a "relative desirability score."  This explanation I think holds well when you present a second set of scores to people where they in essence are still "relative desirability scores" but now you have a zero-anchor.  I've found people get confused by the anchor being 100, but your mileage may vary.  Also you need to keep in mind the interval scaling as Bryan mentions in his answer.

1 Answer

0 votes
Best answer

The probability rescaling has involved an exponentiation step, such that the scores are all positive and may be interpreted in a ratio sense (e.g., a 6 is twice as important as a 3).  The zero-anchored interval scale just involves a shifting of all the raw logit-scaled utilities such that the anchor is the zero point.  These utilities may not be interpreted in a ratio sense.

Depending on whether the anchor is, the probability rescaled scores can tend to squish the least desirable items all pretty close to zero, so if you are interested in discrimination among least-preferred items, this might not be helpful.  But, the ratio interpretation is nice.
answered Nov 8, 2017 by Bryan Orme Platinum Sawtooth Software, Inc. (169,815 points)
selected Nov 9, 2017 by RafalNeska
Bryan, Brian,
Thank you for your comments:)