Am 01.03.20 um 08:22 schrieb Sven Schreiber:
Am 01.03.2020 um 02:17 schrieb Allin Cottrell:
> On Sat, 29 Feb 2020, Artur Tarassow wrote:
>> I would like to add additional time-dimensions (n observations) for
>> each unit in a panel. I've tried to make use of the <dataset addobs>
>> command but it does, instead, add a whole new unit:
>> Is there a 'clever' way to expand the time-dimension while leaving the
>> unit-dimension untouched?
> 1) Save your current panel dataset, making sure you have unit and time
> variables defined.
> 2) Using "nulldata", construct a panel that's larger in the time
> 3) Use "join" to bring in the data from the original, saved dataset.
> 4) Use whatever method you fancy to populate the remaining empty
> observations in the time dimension.
Yes, I also had this workaround in mind.
I also had a case where the raw data set was in the "by variable" format
(as it's called in the manual). In this case it's simple to add n
additional series which also enter the stack() function.
Thanks Allin for the workaround, but I tend to think this should
a feature request - which doesn't say anything about when it would be
implemented, of course. But the workaround could be mentioned in the
ticket there. (Where personally I would use 'append' instead of 'join'
here, but that's not the point.)
Would this be a feature request on sourceforge.org? I could put a new FR
(The problem is obviously related to the fact that gretl always uses
"stacked time series" format internally. Related to that: 'dataset
insobs' is not allowed for panel datasets.)
But I still find it surprising given the behavior of other gretl
functions, so I would propose a documentation like with 'obs': "Note
that the result is invariant to subsampling." And yes, I could insert
that if there are no objections.