mixed cluster error

[Steven Archambault]

Hi Folks, I am estimating a three level model in Stata 14, and I am a little unsure of why we cannot cluster at the second level. I might be a little weak on the theory here. Can somebody explain this limitation? Using data from the mixed help file, you can reproduce the error (highest-level groups are not nested within state). Thanks.

clear all
webuse productivity
mixed gsp private emp hwy water other unemp || region: || state:, vce(cluster state)

highest-level groups are not nested within state

[Clyde Schechter]

It makes no sense to cluster at the state level here. By using a multi-level model with states nested in regions you are affirming that the observations are not independent at the state level because they are correlated within regions. When you use vce(cluster variable) you are not merely affirming that observations are non-independent within groups defined by variable, but you are also affirming that they are independent at and above that level. If that isn't true, (and because you have specified a region level in your model it presumptively isn't true) then the vce matrix calculated isn't valid anyway. This is one instance where Stata will not give you the rope with which to hang yourself.

[Steven Archambault]

This is clear, and I appreciate your response.