9:07 p.m.
Dear Gretl Team,
Is it possible to allow the accessors $test and $pvalue after the
correlation (corr) command?
It is important to me because I'm developing some programs that tries to
automatically choose regressors.
Best,
Henrique Andrade
11:08 p.m.
On Mon, 4 Apr 2016, Henrique Andrade wrote:
Dear Gretl Team,
Is it possible to allow the accessors $test and $pvalue after the
correlation (corr) command?
It is important to me because I'm developing some programs that tries to
automatically choose regressors.
How about this?
<hansl>
function matrix hanslcorr (series x, series y)
r = corr(x, y)
n = sum(ok(x) && ok(y))
test = r * sqrt((n-2) / (1 - r*r))
pval = pvalue(t, n-2, test)
return {r, test, pval}
end function
# example usage
open data4-10
corr CATHOL ENROLL
m = hanslcorr(CATHOL, ENROLL)
print m
</hansl>
Allin
2:09 p.m.
2016-04-04 20:08 GMT-03:00 Allin Cottrell <cottrell(a)wfu.edu>:
On Mon, 4 Apr 2016, Henrique Andrade wrote:
Dear Gretl Team,
>
> Is it possible to allow the accessors $test and $pvalue after the
> correlation (corr) command?
>
> It is important to me because I'm developing some programs that tries to
> automatically choose regressors.
>
How about this?
<hansl>
function matrix hanslcorr (series x, series y)
r = corr(x, y)
n = sum(ok(x) && ok(y))
test = r * sqrt((n-2) / (1 - r*r))
pval = pvalue(t, n-2, test)
return {r, test, pval}
end function
# example usage
open data4-10
corr CATHOL ENROLL
m = hanslcorr(CATHOL, ENROLL)
print m
</hansl>
Dear Allin,
Thanks a lot for your help. It solved my problem, but I think this is a
situation where we should try to make accessors as universal as possible. I
see no reason to not provide $test and $pvalue in this case ;-)
One more comment... I think "n = sum(ok(x) && ok(y))" is only correct
when
x and y have the same size. Am I right?
Best,
Henrique Andrade
2:21 p.m.
Am 05.04.2016 um 16:09 schrieb Henrique Andrade:
2016-04-04 20:08 GMT-03:00 Allin Cottrell <cottrell(a)wfu.edu
<mailto:cottrell@wfu.edu>>:
Thanks a lot for your help. It solved my problem, but I think this is a
situation where we should try to make accessors as universal as
possible. I see no reason to not provide $test and $pvalue in this case ;-)
In principle I'm also for a universal coverage of the $-accessors, but
OTOH the 'corr' command is certainly not a test per se -- otherwise you
could also argue to have something provided by $test after doing 'ols'.
One more comment... I think "n = sum(ok(x) && ok(y))" is only correct
when x and y have the same size. Am I right?
Yes but the length (including the missings) is determined by the current
sample for both x and y, so that shouldn't be a problem.
cheers,
sven
2:22 p.m.
On Tue, 5 Apr 2016, Henrique Andrade wrote:
One more comment... I think "n = sum(ok(x) &&
ok(y))" is only correct when
x and y have the same size. Am I right?
They're supposed to be series. How can they be of different sizes?
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
2:30 p.m.
On Tue, 5 Apr 2016, Henrique Andrade wrote:
Dear Allin,
Thanks a lot for your help. It solved my problem, but I think this is a
situation where we should try to make accessors as universal as possible. I
see no reason to not provide $test and $pvalue in this case ;-)
Well, these would only be applicable in the case where you pass just
two series to the "corr" command (unless we wanted to make this quite
complicated, with $test and $pvalue both being matrices).
One more comment... I think "n = sum(ok(x) &&
ok(y))" is only
correct when x and y have the same size. Am I right?
No, this should pick out the number of observations where we have
valid values for both x and y.
Allin
8:10 p.m.
2016-04-05 11:22 GMT-03:00 Riccardo (Jack) Lucchetti <r.lucchetti(a)univpm.it>
:
On Tue, 5 Apr 2016, Henrique Andrade wrote:
One more comment... I think "n = sum(ok(x) && ok(y))" is only correct
when
> x and y have the same size. Am I right?
>
They're supposed to be series. How can they be of different sizes?
Sometimes our time series start at different point of time ;-)
Dear Allin, Sven and Jack, thanks for your help and comments!
Best,
Henrique
3205
days inactive
3206
days old
6 comments
4 participants
participants (4)
-
Allin Cottrell -
Henrique Andrade -
Riccardo (Jack) Lucchetti -
Sven Schreiber