robust vs. robust cluster(ID) in xtivreg2

[Abdan Syakura]

Hello all,

I have a general question on panel data estimation. I have panel data at country level for 1970-2022. The standard error is very different when using robust vs. robust cluster(ID) option. Almost none of my explanatory variables are statistically significant when using the robust cluster(ID) option. My question: is it absolutely necessary to cluster the SE at the ID level when doing panel data analysis? Why? Is it because the error term might correlate across time within the same ID? Is there any justification for not clustering at ID level?

Here is the explanation in the ivreg2 help file:
"robust specifies that the Eicker/Huber/White/sandwich estimator of variance is to be used in place of the traditional calculation. robust combined with cluster() further allows residuals which are not independent within cluster (although they must be independent between clusters)."

Thank you,

Abdan

[Andrew Musau]

You shouldn't base your decision on the significance of coefficients. The only reason you would choose not to cluster is if you have too few clusters, in which case the gains from clustering are outweighed by the bias from a small number of clusters. White standard errors are not appropriate for fixed effects panel data regression; see https://www.princeton.edu/~mwatson/papers/ecta6489.pdf.

[Abdan Syakura]

Thank you for your answer, Andrew.