The Problem With Small Clusters

The problem with small number of clusters is that usually the standard errors are problematic. The estimation starts to get a little bit off it'd be like trying to do a regression with eight people and it's like well you really trust what you're getting out of that probably not. There are some ways to try to minimize the damage that that small sample size can do, which we've kind of hit on before where you get a dummy variable for each cluster and then you add those in as predictors in the models.