Clusters, CCV or TSCB

[Helle Braathen]

Hi, first time posting in this forum.

I am currently working on a statistics exam and using panel data with fixed effects. I have detected heteroskedasticity, autocorrelation, and cross-sectional dependence. I'm not sure if I'm overcomplicating this for myself, but I've been reading up on how to best adjust for these issues for several days now.

I've read the article "When Should You Adjust Standard Errors for Clustering?" but I got a bit confused. At first, I considered using Driscoll-Kraay standard errors, but I see that my T is too small. My T = 10 and N = 205, so I have a total of N = 2050 observations. I'm analyzing Norwegian municipalities and looking at child welfare interventions per municipality using fixed effects.

So, my question is: Am I right to ignore cross-sectional dependence and just go with regular clustered standard errors (vce(cluster))? Or should I consider more advanced methods? I also understand that using the wrong type of standard errors could lead to incorrect estimates.

Any advice would be greatly appreciated!

[Maxence Morlet]

A case can be made to cluster on the municipality level.

According to MacKinnon et al. (2023): if all municipality have a similar or same number of observations, then you can use the classical CV1 standard errors. If not, you should use wild cluster bootstrap or CV3 standard errors.

[Helle Braathen]

Thank you very much for your relply!