Determining number of clusters in K-clustering analysis - Silhouette scores

[Nicola Cenacchi]

Hi all,

In the past couple of months I have discovered and used STATA's k-clustering code.
As I get more familiar with clustering analysis per se, I learned that there are a few methods to estimate the optimal number of clusters for a certain dataset.

One of the suggested methods is the calculation of the Silhouette scores (or coefficients). Other methods are the Elbow method and the Gap analysis.

I have done several searches but I was not able to find STATA commands to calculate any of these scores.
Does anyone know if anything is available?

Many thanks
Nicola

[John Moran]

Using version 18, under Methods and formulas (for Kmeans and kmedians cluster analysis) there is a reference to: Makles, A. 2012. Stata tip 110: How to get the optimal k-means cluster solution. Stata Journal 12: 347–351. This would appear to be the "Elbow method"; but I have limited experience with these routines

[Nicola Cenacchi]

Thank you very much John,

I will try to have a look. I did read something about the Elbow method. By my reading Silhouette scores might be preferred, but I will double check.
Thanks for the reference !

Best,
Nicola