In the course of working on the bug reported by Ignacio
I took a look at some other aspects of add and omit.
These commands have a long history, and from the beginning
they have had a dual aspect: they run a test but also
"return" a new model, which becomes the current model
in place of the model being tested. I don't think I'd
make them that way if I were writing them today, but
anyway...
A relatively recent addition is the --wald option, in
which case the current model is not replaced. In addition,
there has been undocumented behavior whereby if you
give the --quiet option the current model is not
replaced. This seems to me unruly, and in CVS I've
modified add and omit thus:
* The --quiet flag infuences the printout only, and
has no effect on whether or not the current model is
replaced.
* A new --test-only option can be used to prevent
replacement of the current model. This can be applied
for all variants of "omit", although it's already
implicit in the --wald option.
Does this seem OK?
And one more thing: with a regular "omit", without the
--quiet flag, the full estimation results for the
reduced model are printed first, then the test. This
can confuse the unwary (e.g. my students), because
the first F-test they see in the output is the
"overall" F-test for the reduced model, not the
"omit" test itself. To get around this I've modified
the model printout so that it does the equivalent of
--simple-print, leaving off the auxiliary statistics
including the overall F-test. (If you save the model
in the GUI and then view it in its own right, you
get the full printout as usual.)
Again, does this seem OK?
Allin
Show replies by date