Have an idea?

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

MaxDiff with 115 Attributes

I've been requested to advise on conducting a MaxDiff design with 115 attributes. I would normally display an item 3Xs, but the client has mentioned due to the # of items they'd only want to show each item 2Xs. With that said, I could design a MaxDiff with 23 tasks and 10 items per screen to yield the optimal and most appropriate design. Is there a better approach I should use and recommend. I've never shown more than 5 items per screen, 6 max, so not keen on 10 items per screen. I could of course the # of tasks to accommodate for less items per screen. Looking forward to your feedback.
asked Feb 13 by John

1 Answer

0 votes

I think 10 items per screen is too many - I don't think you'll get a great read on the many items in the middle of the distribution.  If you need utilities for all items, I recommend 23 sets of quints - then each respondent sees each item once (this is called a Sparse MaxDiff and it's been found to be the best way to handle large numbers of items).  At 115 you're close to the limit of what you can do with a Spare MaxDiff.  

Sparse MaxDiff produces excellent sample level utilities but as you imagine, the quality of the respondent level utilities suffers a bit (in an empirical study we published last year in The Journal of Choice Modeling, we found that respondent-level utilities coming from a Sparse MaxDiff were correlated at 84% with utilities generated from a standard 3x MaxDiff).  

If you only care about winners and you don't care much about the quality of respondent level data you can do something called Bandit MaxDiff, which easily handles large or even very large numbers of items.
answered Feb 13 by Keith Chrzan Platinum Sawtooth Software, Inc. (86,950 points)