# Can you have too many clusters in your standard errors?

I know the problems of there being 'too few' clusters when using clustered standard errors, but are there problems with using 'too many'?

For instance, I have 1million observations, and 2500 clusters.

• The number of clusters is built on your criterion. For example, one could use each observation as a cluster. If you have too many clusters, that mean how you build it can be optimised. – SmallChess Oct 27 '15 at 12:52

## 1 Answer

If data highly sparse (with too much variance), you may have even grater number of clusters. One way is to normalize your data and see if you can find some accuracy.