Clustering standard errors at state level - issues

[Jo Lidman]

Dear Statalists,

I have a question in regards to clustering standard errors at state level for panel data. As I have been told if you have less than "42" states (Angrist and Pischke), clustering on states may be an issue. In my case I have 16 states, but I also cluster on household level. So in total I have household: 3000, states 16 and in combination: 4000 clusters. Could this still pose an issue to my analysis or should it be resolved by the combinations of clusters with household clusters?

Kind regards.

[Maxence Morlet]

I would cluster at the state level, using a wild cluster restricted bootstrap (user-written boottest command). By clustering at the household*state level, you do not allow for correlation across households within states. MacKinnon et al. (2023) show that this leads to bias in standard errors.