Reghdfe Absorbed DoF

[ulas alk]

Hi Everyone,

When I use reghdfe and cluster my standard errors at the fixed effect level, my fixed effects become redundant and I got the following message

"* = FE nested within cluster; treated as redundant for DoF computation"

If I dont cluster, I get the same coefficients (of course with different standard errors) and do not get any warning about nested fixed effects.

Does anyone have an idea why I am seeing this? Thanks in advance.

[Andrew Musau]

You should not be clustering if you have fewer than 30 clusters. When clustering, you are assuming that observations are not independent within clusters, but they are between clusters. Therefore, each cluster is effectively a single observation, and you should therefore take this into account when calculating the model's degrees of freedom. There are different ways to adjust the degrees of freedom. For example, you get slightly different standard errors if you estimate a fixed effects model using areg compared to xtreg, fe, resulting from how the degrees of freedom are adjusted. The default for reghdfe is to treat fixed effects nested within clusters as redundant for degrees of freedom adjustment, avoiding some form of double counting. There is a detailed discussion on this and other options to adjust degrees of freedom in the documentation. See

help reghdfe

[ulas alk]

Thanks for your response Andrew Musau . In my actual dataset, I have more than 30 clusters but just wanted to give an example with a readily available dataset. Let's say I have two fixed effects (industry and year) and want to cluster at these two variables because I believe that observations are not independent within industry and year.

Should I be worried about * = FE nested within cluster; treated as redundant for DoF computation warning message and having redundant fixed effects after clustering? I see that all my industry and year fixed effects are shown as redundant as in the example above.

Ulas

[Andrew Musau]

No, there is no issue with how reghdfe does the degrees of freedom adjustment, so you should not worry. It is exactly the same way that xtreg,fe does it. The "warning message" is more information than warning.

[ulas alk]

Thanks a lot.