12:36 p.m.
Hello again,
forgive me if this is a stupid question, but why are robust covariance
matrix estimations not supported for a panel 'between' regression? Since
this is just cross-section OLS it should be easy. Or am I missing something?
thanks,
sven
12:43 p.m.
Am 23.07.2014 14:36, schrieb Sven Schreiber:
Hello again,
forgive me if this is a stupid question, but why are robust covariance
matrix estimations not supported for a panel 'between' regression?
And this is not really related, but I don't want to start too man
threads....
This gives a data error:
<hansl>
open abdata
panel n 0 ys --between
omit --auto=0.2
</hansl>
thanks,
sven
1:27 p.m.
Am 23.07.2014 14:43, schrieb Sven Schreiber:
And this is not really related, but I don't want to start too man
threads....
This gives a data error:
<hansl>
open abdata
panel n 0 ys --between
omit --auto=0.2
</hansl>
actually, 'omit ys' also produces the error, it's not about the 'auto'
option, but apparently about the between estimation.
-s
2:24 p.m.
On Wed, 23 Jul 2014, Sven Schreiber wrote:
Am 23.07.2014 14:43, schrieb Sven Schreiber:
>
> And this is not really related, but I don't want to start too man
> threads....
>
>
> This gives a data error:
>
> <hansl>
> open abdata
> panel n 0 ys --between
> omit --auto=0.2
> </hansl>
>
actually, 'omit ys' also produces the error, it's not about the
'auto'
option, but apparently about the between estimation.
There should be a better error message, but the point is that the panel
between model is just provided "by courtesy" and doesn't offer the
facilities that other models do. This is because it's not estimated on the
dataset proper but rather on a panel-means dataset constructed on the fly
and then destroyed.
We could work on offering more, but in the meantime you can use the p*
functions to construct a suitable dataset yourself, as in
<hansl>
open abdata.gdt
panel n 0 ys --between
matrix bn = pshrink(pmean(n))
matrix bys = pshrink(pmean(ys))
N = rows(bn)
nulldata N --preserve
series n = bn
series ys = bys
ols n 0 ys
omit ys
</hansl>
Allin Cottrell
3:08 p.m.
Am 23.07.2014 16:24, schrieb Allin Cottrell:
On Wed, 23 Jul 2014, Sven Schreiber wrote:
There should be a better error message, but the point is that the panel
between model is just provided "by courtesy" and doesn't offer the
facilities that other models do. This is because it's not estimated on the
dataset proper but rather on a panel-means dataset constructed on the fly
and then destroyed.
Fair enough, but then I would suggest:
- yes a better error message, and ideally also a note in the docs,
- if 'between' regression is not a first-class gretl member, then
perhaps it should not get its own menu entry (model -> panel -> between
groups); perhaps it could go under OLS if the dataset has a panel structure.
thanks,
sven
8:53 p.m.
Am 23.07.2014 um 16:24 schrieb Allin Cottrell:
We could work on offering more, but in the meantime you can use the
p*
functions to construct a suitable dataset yourself, as in
<hansl>
open abdata.gdt
panel n 0 ys --between
matrix bn = pshrink(pmean(n))
matrix bys = pshrink(pmean(ys))
N = rows(bn)
nulldata N --preserve
series n = bn
series ys = bys
ols n 0 ys
omit ys
</hansl>
Just for future reference, an alternative way would be to exploit the
fact that 'pmean()' stores the values repeatedly in all time periods so
one could just pick the first time period and then run the regression
without leaving the original workfile:
<hansl>
open abdata.gdt
panel n 0 ys --between # just for comparison
series ave_n = pmean(n)
series ave_ys = pmean(ys)
genr time
smpl time==1 --restrict
ols ave_n 0 ave_ys # can also use --robust here
omit ave_ys
</hansl>
BTW, it would seem natural to extend pmean() (and the other p*
functions) to accept lists just like diff(), ldiff(), lags() etc.
The template for naming the created series could be 'pm_varname' or
somesuch.
cheers,
sven
3803
days inactive
3804
days old
5 comments
2 participants
participants (2)
-
Allin Cottrell -
Sven Schreiber