Have an idea?

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

Is there a paper on running a MaxDiff on the fly to select attributes to include in a conjoint exercise?


We're potentially looking into running a study in which there could be as many as 35 attributes that would be evaluated (each with 2-4 levels). Since this is too many attributes, I was thinking of using MaxDiff on the fly to evaluate which 5 or 6 attributes are most important to respondents and feed those into a conjoint exercise (which I guess would have to be an ACA). Does this notion have any red flags? Are there any papers showing how I should implement and what I should look out for? Would it be possible to use a choice-based conjoint approach instead of a ratings based approach? Are there alternative approaches I should consider instead of the one I described above?
asked May 22, 2015 by Phil
Adaptive CBC is the only methodology that allows a customized attribute list per respondent.  We don't have a formal paper specifying what the best approach would be, unfortunately.  My initial thought is that of is there really 35 different attributes that everyone cares about?  I wonder if it would be better to try a two-stage approach where you run a MaxDiff on the 35 items to try to whittle things down.  We typically recommend each item being shown 4+ times for calculating MaxDiff scores on the fly, so you are potentially combining a very large MaxDiff exercise with an Adaptive CBC exercise.
Hi Brian,

You're actually describing what my initial thought was. I was just wondering if it was a legitimate approach and how I would go about doing that.
Although it is difficult to speculate without information regarding the content of attributes, I would suspect that a number of the attributes in a set of 35 would be correlated with one another.  One way of reducing the set might be to eliminate or somehow merge highly correlated items.

1 Answer

+1 vote
As Brian mentions, ACBC (Adaptive CBC) is our only software that could be used to take custom "MaxDiff scores on the fly" and use it to decide which attribute levels to display in a simultaneous conjoint exercise.

Here's the documentation on using MaxDiff to compute scores on the fly: http://www.sawtoothsoftware.com/help/lighthouse-studio/manual/index.html?computingscoreson_the_fly.html

You can use "Constructed Lists" logic in SSI Web to move the most preferred few levels into an ACBC study.

I have written a paper on comparing Counts, Individual-Level Logit, and HB for MaxDiff, see: http://www.sawtoothsoftware.com/support/technical-papers/maxdiff-best-worst-scaling/maxdiff-analysis-simple-counting-individual-level-logit-and-hb-2009
answered May 23, 2015 by Bryan Orme Platinum Sawtooth Software, Inc. (177,015 points)