Accounting for clustering by facility

[JACQUELYNE ODAK]

I am conducting a study using 3 health facilities. Should I account for clustering by facility for the cross-sectional and case-control studies, and why/why not?

[Felix Bittmann]

In theory: if the sampling process is related to the clusters, yes, clustering is a good idea. However, 3 clusters is a very small number and usually at least 30 clusters are required. You can use cluster fixed-effects and use robust standard errors.

[Carlo Lazzaro]

Jacqueline:
welcome to this forum.
As an aside to Felix's helpful reply, you may want to take a look at Cameron_Miller_JHR_2015_February.pdf.