On Mon, 22 Jan 2024, Sven Schreiber wrote:

The dataset is monthly in the time dimension and recognized as such by gretl, n=1..435 and t=2023:01..2023:12. (N*T = 5220) Missings are present somewhere in most of the included variables.
[...]

Sven, I tried replicating what you describe but failed to replicate the problem (not very surprising). Could you give us a little more detail, please?

* What sort of manipulations preceded your saving of the panel dataset to disk?

I don't remember exactly in what order I took the steps. What I can say is that I started with two panel datasets for the years 2022 and 2023 (so T=24). 

Then I "reduced" both datasets to the year 2023, discarding the first (in a time sense) halves of each. 

Then (I think) I "joined" two variables of the second dataset into the first, using "--ikey=stationids,iso" as the panel index variables. ("stationids" are non-consecutive but unique integers for the units, "iso" is the ISO-format date number of the month.)

The "ichk" series in my example was formed by running the pnobs() function separately on two or three variables of the dataset, and then combining those results with the appropriate boolean pattern to obtain an indicator dummy for the obs of those cross-sectional units which don't have any valid obs (in that year).

* When you saved the restricted dataset, did you use the "store" command or the GUI? And if the latter, did you take the option (it's offered in a dialog) of reducing the dataset in memory to that saved to disk?

I used the GUI for that, and gretl asked whether I want to stick with the reduced dataset, and I answered yes. Or actually, I think it asked "do you want to revert to the full range", and I answered no.

thanks

sven