[Gretl-users] Re: Jackknife Standard Errors for Clustered Regression
Am 17.08.2022 um 08:12 schrieb JOSE FRANCISCO PERLES RIBES:
Dear team:
I have recently read the paper of Prof. Hansen "Jackknife Standard
Errors for Clustered Regression"
(https://www.ssc.wisc.edu/~bhansen/papers/tcauchy.pdf) and found it
very interesting. I have checked that for the non-clustered scenario
the HC3a option in Gretl is the Jackknife Professor Hansen's favorite
option (according to the Gretl command reference Variant 3a is the
MacKinnon–White "jackknife" procedure).
My question is, for the clustered regression which would be the option
in Gretl (as stated in the paper, in Stata it would be
vce(jackknife,mse).?
Hi José, from section 22.5 in the guide (and also checking the doc for
the --cluster option for the ols command) I tend to think that in gretl
currently these are mutually exclusive options, so it wouldn't currently
be possible. Maybe someone with more experience in this area can weigh in.
I'm noting that combining the --cluster and --jackknife options does not
produce an error, however. While this would be correct according to the
spec (cluster just overrides), I wonder whether it's a good idea to
effectively ignore an unsuitable option silently. If a user wants to
override, couldn't she be expected to remove the jackknife option
explicitly? But that's a different issue not directly related to your
question.
thanks
sven