Have an idea?

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

Anchored MaxDiff shows positive utility of costs

We used HB estimation for a MD data set.
The list included benefits and costs of a product.
Anchor definition was set in a way that all costs are below the threshold and benefits above the threshold.
The anchors are defined identically for all respondent.
To our surprise in the utilities several cost items receive positive values and vice versa for benefits for some of the respondents.
Is this a result a result of having identical anchors/thresholds for all respondents?
asked Jan 10 by alex.wendland Bronze (2,080 points)

1 Answer

0 votes
Alex, when you say "anchor definition was set in a way that all costs were below the threshold" does this mean you hard-coded the anchoring questions instead of letting respondents answer them?  Or do you mean something else?
answered Jan 18 by Keith Chrzan Platinum Sawtooth Software, Inc. (75,350 points)
Hi Keith, all cost items have the logic 1=2 and all benefits 1=1. So, yes, this is hard coded compared to have a rule depending on respondent stated information.
Does that makes sense?
Alex, that's what I suspected.  I've seen this, too, even when I put in hard anchors like this.  Say you have items 1-10, ordered from best to worst, and say you put in the hard anchor between items 5 and 6.  But when a respondent trades-off the items, he puts item 7 above item 5.  This contradictory information can shift items above or below the anchor you put in, because the model is trying to find the best-fitting place to put it amidst this contradictory information.  I assume your incorrectly placed items are not dramatically far away from the threshold?  If so, did responcdents understand them the way you intended?
Thanks, that makes sense from the estimation point of view. I hadn't read up on this and thought anchoring would work more like a constraint which is "reliable" and forcing the expected structure.

Item wording was fine I think but panelists will be panelists and observations were quite sparse (each item shown less than twice).

Could you please also point me to where I can find an explanation of the Zero-Anchored Interval Scale transformation?
We could only find some text about the properties but not on how the transformation is performed or how to possibly interpret the values.
Thanks in advance!
Alex,

I'm on the road today but if you email me at keith@sawtoothsoftware.com I'll send you the information when I arrive to my hotel.
...