k-medians optimal groups

[Nazli Uludere Aragon]

Dear Forum. There is a Stata Tip from 2012 by A. Makles on determining the optimal number of groups in a k-means clustering analysis. See: https://journals.sagepub.com/doi/pdf...867X1201200213. This is also cited in Stata manual section that deals with cluster k-means and k-medians. I think I can use the same routine for identifying optimal group number for a k-medians analysis (I have a large dataset and want to run 50+ iterations). Do you agree? This is not explicitly stated in the tip document. I just wanted to be sure. Thank you.

[Jared Greathouse]

Presumably. Email and ask the author, then post their response.

[Nick Cox]

I don’t see why this wouldn’t work.

The best test of a clustering in my view remains whether there is a two-dimensional summary graph of or close to the data on which clustering appears if not obvious, then at least helpful.

Back in the 19th century biologists classified themselves as “lumpers” and “splitters”. The urge to classify varies wildly in intensity….

[Nazli Uludere Aragon]

Thank you both for your replies. I have emailed the author too and will share the reply.

[Nazli Uludere Aragon]

From Dr. Makles (author of the Stata tip): "The WSS is based on means/averages and not on medians. By just changing k means to k medians and running the Stata tip will lead you to a comparison of apples with oranges."