10:22 a.m.
...just after sending the last message gretl crashed on me, and it's
reproducible (in the console):
- open abdata and estimate something with OLS
- run hausman
- execute 'print $test' (which produces NA)
- run hausman again -> crash
Recent Windows snapshot.
thanks
sven
3:10 p.m.
On Sun, 11 Jul 2021, Sven Schreiber wrote:
...just after sending the last message gretl crashed on me, and
it's
reproducible (in the console):
- open abdata and estimate something with OLS
- run hausman
- execute 'print $test' (which produces NA)
- run hausman again -> crash
Thanks for the report. That's now fixed in git. I haven't updated
the snapshots just yet because some new stuff needs more testing
first.
Allin
4:52 p.m.
On Sun, 11 Jul 2021, Sven Schreiber wrote:
- open abdata and estimate something with OLS
- run hausman
- execute 'print $test' (which produces NA)
Not having $test available in this case seems a bit lame. In git
I've made $test and $pvalue available as 3-vectors following
"hausman". The three tests are in the order shown in the printout:
1 Wald test for poolability in relation to fixed effects.
2 Breusch-Pagan test for poolability in relation to random effects.
3 The Hausman test itself.
If this behavior seems OK I can document it. (Maybe there's a case
for just giving the Hausman test result?)
Allin
6:39 a.m.
Am 11.07.2021 um 22:52 schrieb Allin Cottrell:
On Sun, 11 Jul 2021, Sven Schreiber wrote:
> - open abdata and estimate something with OLS
> - run hausman
> - execute 'print $test' (which produces NA)
Not having $test available in this case seems a bit lame. In git I've
made $test and $pvalue available as 3-vectors following "hausman". The
three tests are in the order shown in the printout:
I think in principle having these accessors here is good, yes. (This
looks a bit like phasing out the usage of the special accessor $hausman
at least in the panel context, which I'd see as further harmonization, a
good thing IMO.)
1 Wald test for poolability in relation to fixed effects.
2 Breusch-Pagan test for poolability in relation to random effects.
Couldn't these be done as "modtest --poolability" or something?
3 The Hausman test itself.
If this behavior seems OK I can document it. (Maybe there's a case for
just giving the Hausman test result?)
Yes, I tend to think it's better to separate those things. The Hausman
test principle is very general, but Wald or LM tests are different, so
I'd say it gets confusing when those have to be invoked by a "hausman"
command.
The other way around: the Hausman result in the narrow sense could also
be returned as part of a new "modtest --poolability".
thanks
sven
2:23 a.m.
Am 11.07.2021 22:52 schrieb Allin Cottrell:
On Sun, 11 Jul 2021, Sven Schreiber wrote:
> - open abdata and estimate something with OLS
> - run hausman
> - execute 'print $test' (which produces NA)
Not having $test available in this case seems a bit lame. In git I've
made $test and $pvalue available as 3-vectors following "hausman". The
three tests are in the order shown in the printout:
1 Wald test for poolability in relation to fixed effects.
2 Breusch-Pagan test for poolability in relation to random effects.
3 The Hausman test itself.
If this behavior seems OK I can document it. (Maybe there's a case for
just giving the Hausman test result?)
Hi Allin,
in case the decision is to keep this feature, I suggest to add row
labels. Currently the output is just
<hansl>
open grunfeld
ols invest const value
hausman
eval $test
? eval $test
15.974
273.16
3.8110
</>
and the user (ok, at least me :-) ) would have to consult the document
each time checking which row refers to what.
Best,
Artur
1223
days inactive
1225
days old
4 comments
3 participants
participants (3)
-
Allin Cottrell -
atecon -
Sven Schreiber