-bootstrap- with -cluster()- option to improve statistical precision?

[Peter Meier]

Dear all,

I am currently running the following regression command:

reghdfe growth c.var1##c.var2, absorb(panelvar timevar) cluster(panelvar)

This runs through but given a small sample size statistical precision is more limited than I'd wish for.

I was wondering whether this might improve if I bootstrap, and hence tried

bootstrap, reps(100) cluster(panelvar): reghdfe growth c.var1##c.var2, absorb(panelvar timevar)

Regardless of whether I add the -cluster(panelvar)- as option to the bootstrap or whether I add it as option to the actual regression, I get the following error

"Repeated time values within panel. The most likely cause for this error is misspecifying the cluster(), idcluster(), or group() option."

From the bootstrap help file I could not see how else I should specify the cluster option and I did not have either of the other two.

Would anyone happen to see my mistake, or more generally have an opinion on whether this type of bootstrapping might help with statistical significance at all?

Thank you so much!
PM

[FernandoRios]

Hi Peter
two points:
1. To solve the issue you describe you need to type: xtset, clear
This will address the repeated time values warning
2. Your point regarding "improving" Estimates. Bootstrap will get you an estimate that may be better in some cases, due to non-normality, or skew distributions. But if Clustering is the focus of the problem, perhaps Jacknife estimator is better (Jackknife by cluster variable). This is at least one of the latest recommendations on Clustered standard errors.
Now, this will not make things more significant. It will just get you SE that are more correct given the specification you have in mind.
Best wishes

[Joao Santos Silva]

Dear Peter Meier,

Going to the bottom of the question, bootstrap will not help you in this case because the two sets of standard errors are asymptotically equivalent.

Best wishes,

Joao

[Peter Meier]

Hi Fernando and Joao,

thank you both so much!
It is both helpful to know now how to have it run through without error message (just verified), and understand that it does not solve my current challenge but may help on future occasions when dealing with possible non-normality or skewed distributions!

Thank you so much and best wishes,
PM