[Gretl-devel] Re: clustered standard errors, 'period' keyword, driscoll_kraay

Am 02.11.2023 um 13:51 schrieb Allin Cottrell: > On Thu, 2 Nov 2023, Sven Schreiber wrote: > >> in short: --cluster=period works as announced, but with two-way >> clustering --cluster=period,unit doesn't work, one has to create a >> series mimicking the period (e.g., genr time). >> >> I guess this is an oversight and not intended. > > It's not intended as such, but neither is it just an oversight. It's > not easy to combine an automatic structural feature with an explicit > series, given that many panels are unbalanced. But maybe we can find a > way of doing it. OK, if it's not an oversight, I would just document it and leave it as is. > >> Also, Allin, off-list you said that "set panel_robust driscoll_kraay" >> now is a further option, but with yesterdays's snapshot I'm getting >> an invalid value error. Is this in a different branch or not pushed >> or something? > > See the "set" doc for the current status: options are "arellano", > "pcse" and "scc". Exactly - but my question is what happened to the driscoll_kraay thing you mentioned a couple of weeks ago? Did it not work out, was just experimental, or something else?