Have an idea?

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

anchored Maxdiff application in Producing Partworth utilities


I wanted to know if it is possible to use Anchored MaxDiff in health economic studies and how could that be done?

I am using Case 2 MaxcDiff to describe some health states based on a certain tool and I want to use Death as the anchor value Zero, then compare other health states (utilities) to this anchor & the part-with utilities to it too.

Classically we use a different utility evaluating method that produces direct utilities (part-worths) between -1 & 1 then map the MaxDiff results to it, but I was wondering if this could be done within the MaxDiff itself without using another source of comparative utilities, since Originally yet utilities from MaxDiff are not bound to 0-1 (Death and Life"- is worst than death".

asked May 6 by AMYN Bronze (1,335 points)

1 Answer

+1 vote
This is a slightly different case of anchoring, one that's easily solved if I understand your intention.  

If you're using Best-Worst Case 2, then you have some options about how to code your utilities.  I'm assuming death is one of the levels of one of your attributes?

Probably the easiest way is to code the experiment just like any other MaxDiff study (with effects coding of all items as if they were a single attribute) but then realize that you could recode each attribute separately by subtracting off  the utility of the reference level from the utilities of all the other levels - this will give all attributes a (separate) zero-based score.  Just make death the reference level of its attribute.

Or perhaps you want to make death the reference level for the whole experiment and not just it's attribute.  In this case just subtract the utility of death from the utilities of all the items - death's utility will then be zero and every other item will have it as a reference level.  

But it would probably help if you explained more about your design and your intentions for interpretation, so that we don't have to live with my assumptions about them.
answered May 6 by Keith Chrzan Platinum Sawtooth Software, Inc. (75,350 points)
Hi Keith,

thank you for your insights. I will try to explain more about our study.

We aim to calculate the preference part-worth utilities of the general population for a general health economic tool. We are using 3 methods & want to calculate the utilities using each mothed then compare the results from each method as well as the results of every respondent to each method.

The 3 methods are CBC, MaxDiff case 2 & Time Trades-Off.

For the first point you mentioned, Death is not one of the attributes or a level of an attribute. So we need to use another method (Time Traders-Off) to calculate the utility of Death then include it in the MaxDiff (as what I understood from other researches).

So we want Death to be a reference level (our true Zero) for the whole experiment since it is not included in any attribute and it could be considered as a separate case.  

So If I understood correctly, Anchoring MaxDiff cant help me adjust the MaxDiff utilities on a scale of 0-1 (Death to Full life) with some negatives as worst than death cases, unless Death is one of the attributes and needs the help of another scaling method to rescale the results of it on the required scale.

In your case, I think we don't need to do one of the more complicated anchoring techniques.  We just do MaxDiff and include Death as one of the items.  Then when you get your MaxDiff utilities, subtract the utility of Death from the utility of each of the items.  With this transformation, utility of Death will be 0, those items worse than death will have negative utilities and those better than dying will have positive utilities.
thanks for the suggestion.
Originally Death is not part of any of the attributes in the tool we are evaluating. Do you suggest that I add it as an extra Attribute with only one level (death) or how could I add this extra item?
I do not think anchoring will do what you want it to do.  Unless Death is an item in the MaxDiff, I do not see how you will be able to use it as the reference level for estimating MaxDiff utilities.  And if you do this, then since Death is not in the CBC, you will necessarily be using a different design for the MaxDiff than for the CBC.
Well in the CBC I ask the respondent to evaluate the worst of each 2 concepts to death, so in one hand it is not an item by itself but on the other hand, it is evaluated by the respondents as a separate concept. would this qualify as including death in the CBC?
I'm having trouble envisioning the task you're having respondents do.  Can you send me an example at keith@sawtoothsoftware.com?
I tried to send you a zip file of the study to your e-mail from 3 different e-mails but all were rejected
It's a really big file, then?  Maybe try dropbox, but I think a screen shot of the question may be all I need right now.
The zip file is not that big, just 2.3 Mb. My concern is that the questionnaire is in Chinese and that you want to translate it to English to understand the content, I added a google translation tool so that is why I wanted to send you the whole study. but I will try to send the .pdf file print of the study.