#300 add feature for panel data model clustered by firms does not work

[JTN]

I don't know why, but "add" command (via script or menu) does not work, when adding variable to the model (in order to get F change). I have panel data model clustered by firms and I get info that "requested statistics is not available".

[Allin Cottrell]

Please show us your data and model specification, and tell us what version of gretl you're running.

I tried something that I think should be similar using current gretl: (1) load the Arellano-Bond data file abdata.gdt; (2) estimate a fixed-effects specification with dependent variable n and independent variables wand k, with robust standard errors in the variant Arellano (clustered by firm) selected; (4) in the model window menu, go to /Tests/Add variables and select ys as the variable to add. This gave me a robust F-test for adding the selected regressor.

I also tried the following script without any error:

open abdata.gdt
panel n 0 w k --robust
add ys

[JTN]

Thank you for your response.
My data is very large. How can I share it? The version of gretl is the latest one (2023c).
I use pooled ols method.
I have opened abdata.gdt and tried on this file something similar to my work. I tried following script:

ols n 0 w k --cluster=unit
add ys

And after "add ys" line I'm getting info "requested statistics is not available".
It is awkward because some of previous attempts based on the same data and with the same script succeded.
Thank you for your time.

[Artur Tarassow]

My data is very large.

What do you mean by large? How many variables and rows do you have? Maybe we can simulate such an dataset if you can't share it.

You may want to upload a zip-file here including the data. Simply press the "Add attachements" below your message here.

You can store a compressed gretl-dataset (https://gretl.sourceforge.net/gretl-help/cmdref.html#store) as follows:

list VarsToExport = SERIES YOU WANT TO EXPORT
store FILENAME.gdt VarsToExport --gzipped=9

Best

[Sven Schreiber]

I can confirm the error message, on a January snapshot. I'd say that econometrically it doesn't make much sense to estimate a pooled OLS model and clustering for the units at the same time, and maybe that's why this problem has gone unnoticed so far, but that's arguably a different issue and a different discussion. I would agree that it's a bug, I guess it should work.

[Allin Cottrell]

Thanks for the report, JTN. No need to share your data now that we see how to replicate the bug. This should be fixed shortly. I'll post again and close the ticket once fixed snapshots are available.

[Allin Cottrell]

This should now be resolved in git, and the snapshots for Windows and Mac. But I'll await confirmation of that before closing this ticket.

[JTN]

Thank you very much. Now it is working just fine. I very much appreciate it.