#174 Not able to add 2SLS regression to model table

[Arne Gramstad]

When trying to add 2SLS regressions to the model table I get the error: "Sorry, this model can't be put in the model table".

The model table feature works fine with OLS. I haven't attempted other types of non-OLS regressions.

OS: Windows 7

[Sven Schreiber]

[This action is part of a little round of maintenance, sorry for the delay.]

This behavior still in 1.9.14; I suspect (but don't remember exactly) that this is by design, namely that only OLS (and probably single-equ. models) are allowed. I agree that this limitation feels overly restrictive and IMHO therefore the 'modeltab' business is not super-useful. But that's a matter of taste and in any case would be a feature request and not a bug.

However, I found out that the docs are not explicit about this and also somewhat circular: the User Guide refers to the 'modeltab' Command Reference, and the command ref. says "See the Gretl User's Guide for details." This is a minor bug (and I'm lowering the severity/priority). If Allin confirms that the OLS limitation is the state of the design, I could add such a note to the docs and then close this bug.

thanks,
sven

[Sven Schreiber]

Still in 1.9.90 -- Allin, what's the future of modeltab anyway? Is it supposed to only work with OLS by design, or should it be extended to cover other model types in the future?
Thanks,
Sven

[Sven Schreiber]

Ping -- the circular references between the User Guide and the Command Ref. are still there.

I see two options / design decisions for the "modeltab":

• It's only made for OLS-estimated models, and that's the way it is.
• The plan is to extend it to (some) other estimators, but it's not there yet.

In any case IMHO the docs should be updated to communicate the situation.

Allin, could you announce your opinion on this? Otherwise eventually I would try to add to the docs somewhere the statement that the modeltab is only meant for OLS.

thanks,
sven

[Sven Schreiber]

Another ping, I will also alert on the devel-mailing list to reach a conclusion.
thanks,
sven

[Allin Cottrell]

OK, 2SLS models can now go in the model table, in today's
CVS. Many types of model can go in, but we exclude certain
"fancier" types where we don't have the facility to display
important information in the model-table format. With 2SLS
we can't show what the instruments are, but I guess that's
acceptable if one just wants a simple view of the final
estimates.

[Allin Cottrell]

P.S. look at gui2/model_table.c for the low-down. You'll
see that in the function model_table_precheck() we screen
the model type, disallowing NLS, MLE, GMM, ARBOND and some
others. If not explicitly banned, a model type is accepted.

[Sven Schreiber]

Hmm, isn't this a recipe for bugs in the future, when new funny models
are introduced in gretl and it is forgotten to explicitly ban them here?

That is, wouldn't a whitelist approach be better than a blacklist?

thanks,
sven

Am 12.03.2015 um 21:55 schrieb Allin Cottrell:

P.S. look at gui2/model_table.c for the low-down. You'll
see that in the function model_table_precheck() we screen
the model type, disallowing NLS, MLE, GMM, ARBOND and some
others. If not explicitly banned, a model type is accepted.

[Allin Cottrell]

On Thu, 12 Mar 2015, Sven S. wrote:

Hmm, isn't this a recipe for bugs in the future, when new funny models
are introduced in gretl and it is forgotten to explicitly ban them here?

That is, wouldn't a whitelist approach be better than a blacklist?

Maybe so. The blacklist approach does require that we remember about
the model table when adding new built-in estimators. However, I
guess our general policy is to back off on adding C-coded estimators
in favour of function packages.

Allin

Am 12.03.2015 um 21:55 schrieb Allin Cottrell:

P.S. look at gui2/model_table.c for the low-down. You'll
see that in the function model_table_precheck() we screen
the model type, disallowing NLS, MLE, GMM, ARBOND and some
others. If not explicitly banned, a model type is accepted.

[Sven Schreiber]

Ok thanks, so that's the feature request part, the circular reference in the docs is the remaining minor bug. I would change the docs as promised, but as nowadays I'm not on Linux regularly, I cannot do it right away.
thanks,
sven

[Sven Schreiber]

I've put a note into the user guide about the limitations of the model table, and thus I'm closing this.