Have an idea?

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

CCEA error message "minimum membership"


while running a consensus clustering analysis (2 to 10 groups; only k-means with mixed starting Points; one metric variable), I received the following error message:

"Could not find a solution with minimum membership"

The error message occurred while analyzing the 6-group-solution. Could you please explain what's behind that message?

Why is it not possible to derive a 6-group k-means solution (and higher)?

asked Dec 2, 2015 by anonymous

1 Answer

0 votes
Without seeing your project and data and just reading what you've put, let me try to guess what's going on.

First, I gather you are running in K-means mode, not cluster ensemble with consensus mode.  That means on your Settings tab, you've clicked "K-means, highest reproducibility replicate".  On that same tab, you'll see a box which states "Minimum group size".  By default, that should be set to "1", but you may have changed it.  That "Minimum group size" specifies what the smallest group size in terms of number of respondents belonging to it is allowed.

Next, you specify that you are using just one metric variable in your cluster solution, which seems rather strange to do, since most cluster solutions employ many more basis (input) variables than just one.

As a simple example of what may be occurring for you, let's imagine that your one input variable you are using for clustering just has values on a 5-point scale.  A 5-group solution would have no course of action but to take all people who answered a "1" and put them in group "1", all people who answered a "2" and put them into group "2", etc.  

You can see that if the algorithm goes to segment respondents into 6 categories in this situation, there is no way possible for it to find a way to classify people into more than 5 categories, so the "minimum group size=1" would fail to be satisfied and you'd receive this error message.

Don't know if that's what's going on here, but I thought I'd first ask you about your basic setup and start with this simple explanation.
answered Dec 2, 2015 by Bryan Orme Platinum Sawtooth Software, Inc. (132,290 points)
Thank you very much.

That is exactly the point. The reason behind just using one metric input variable is that we want to construct a diversified cluster Ensemble. Therefore, each (latent) construct is also used on a separate basis, while we finally merge them all in the final ensemble.

But it shouldn't be a problem for us. Thanks again for clarifying this.