Re: [Gretl-devel] gretl 1.8.5?
IMHO these are all worthwhile improvements but given the resource
constraints, I of course understand some or all of them will have to
wait. Thanks very much for taking them into consideration.
> b)- I still don't see why saving fitted values, residuals
and
> squared residuals need to be disabled while the various tests can be
> very neatly done without reverting to the model sample.
That's because you haven't looked at the code ;-). Or in other
words, you haven't had to think through the full implications of
adding new variables in the case where we're looking at two
different sub-sampled datasets: the one the model was estimated
on, and the one currently in force. (It's not so difficult if the
full dataset is currently in force and only the model's dataset
was sub-sampled, but even that is tricky to get right.)
I don't know the code (at least for now ;-) but still I think the
operation here should not be related to what the current sample is no?
What gretl needs to do here is:
1)- Temporarily restore the model sample (internally),
2)- Create a vector with the size of the full-sample, where the rows
valid for the model sample are filled with the required values from
the model and the rest are just missing values. This is what gretl
would produce if the user first selected the model sample, saved the
residuals etc. and returned to the initial sample.
3)- Return back to the current sample which may or may not be the full
sample. If the user now chooses to display the newly created variable,
what he will see may or may not be a smaller vector with or without
missing values depending on what the current sample is. The newly
created variable will probably be useless unless the user selects the
model sample but that is not important.
Please correct me if the above approach is wrong.
Cheers
A. Talha Yalta
--
“Remember not only to say the right thing in the right place, but far
more difficult still, to leave unsaid the wrong thing at the tempting
moment.” - Benjamin Franklin (1706-1790)
--