I've been given a segmentation in which people are segmented into, say, 6 groups on the basis of some number of survey questions. The segmentation is based on a cluster analysis of some sort.

My task is to use a different, second set of information to see how well that can recover the assignment to segments. (This is because this second set is available for many more respondents.) I can measure the accuracy on this part by a confusion matrix.

BUT, my ability to do this is limited by the reliability of the initial segmentation itself. The people who generated the segmentation think of it as gospel, and have provided no metrics.

What should I ask them to do? What's the best way to measure how reliable that initial segmentation is?

  • 1
    $\begingroup$ Please, be more precise in your question: what information do you have about that old segmentation? In a comment to this question I collected some local links which might be helpful; and my answer there outlines landmarks of validity in cluster analysis, and I think your case is mostly about cross-validation. $\endgroup$ – ttnphns Sep 26 '17 at 19:18
  • $\begingroup$ @ ttnphns Right now, I only have the second set of information and the segmentation assignment. The first set of information (survey questions) is potentially available if I can get cooperation, which will depend, in part, on answering the question about "what are you doing to do with it?" I'll take a look at your linked question now. $\endgroup$ – zbicyclist Sep 26 '17 at 19:59
  • $\begingroup$ @ ttnphns I think I'm going to aim for a cross-validation check (generalizability), as described in your linked answer, if I can get the raw data. That will produce a confusion matrix so I can use the same type of confusion-matrix based measures to compare my later results. $\endgroup$ – zbicyclist Sep 26 '17 at 20:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.