MaxDiff Segmentation with excluced item

Question

When clustering maxdiff with 29 items using the latentClass functionality within maxdiff software: Is there a possibility to cluster with just 28 Items and exclude one?  

Any ideas how to do that? Unfortunantely replicating the segments with HB scores and k-means or other clusters does not work. Even for the 29 items different custers appear. (i guess the sawtooth software algorithm is more intelligent for this kind of data).

Thanks

Answer 1

If you toss out all choice sets where that attribute is chosen as best or worst, and then eliminate that attribute's column from your design matrix, the analysis should run for you.  

I think you create problems for yourself that way, especially if the attribute was chosen a lot, not least of which will be explaining the collateral damage of the recoding/data deletion.  

If you want to segment on only the 28 items, I'd cluster the HB scores instead.  As you note, you will of course get different clusters.

Comments

Keith, another approach that would lead to throwing away less good data (for sets in which the excluded items was chosen either best or worst) would be to recode all the MaxDiff tasks into their implied paired comparisons again for HB estimation.  As you know, a MaxDiff task among 5 items results in 10 implied paired comparisons.  Then, just throw away the implied paired comparisons that involve the item that Steve wants to omit from his latent class analysis.  So, if the item to be excluded from having an effect on the estimation was involved in a MaxDiff set, it would involve throwing away 4 of the 10 paired comparisons...so you'd still preserve 60% of the information from that choice set rather than losing all of it.

But, this approach is of course much more work than the easier solution you described.