4:51 p.m.
Hello
I was wondering if the Gretl API offers a paired t-test or the Kruskal–Wallis test for
more than two variables.
Thank you
Ryan
5:36 p.m.
Am 05.05.2021 um 18:51 schrieb riles(a)triton.net:
Hello
I was wondering if the Gretl API offers a paired t-test or the Kruskal–Wallis test for
more than two variables.
Hi, about the Kruskal-Wallis test: AFAIK the contributed package
"KruskalWallis" is the only offering, and hence not in C, but in hansl.
But in there the KruskalWallis2 function appears to support several
variables.
(@Jack, BTW: I guess the issue is practically moot now, but I think the
package requirement would have to be 1.9.12 instead of 1.9.4.)
The paired t-test is built into the GUI in the test statistic
calculator, so there's a C backend "somehow". But do you already know
how to do a simple mean test through the C API? Then I guess the
recommendation is to form the difference of the two (paired) series and
proceed as usual.
cheers
sven
6:53 p.m.
ok, thats what I thought,
I don't see the paired t-test option in the gui, but since I am not using the GUI that
doesn't matter. Yea looks like I just call the means_test() function in
descriptions.c, so I will follow your recommendation for doing the paired t-test in the
API.
Thank you for your help
Ryan
10:13 p.m.
On 05.05.21 20:53, riles(a)triton.net wrote:
ok, thats what I thought,
I don't see the paired t-test option in the gui, but since I am not using the GUI
that doesn't matter.
I guess you're right, it's not really there, only a nonparametric paired
analogue. The reason is probably the same, namely that it's so easy to
just form the difference and do the simple mean test.
cheers
sven
7:33 a.m.
On Wed, 5 May 2021, Sven Schreiber wrote:
(@Jack, BTW: I guess the issue is practically moot now, but I think
the
package requirement would have to be 1.9.12 instead of 1.9.4.)
Do you think it worthwhile to update the package just for this? I had a
brief look at the package to see if there was something else worth adding
to justify raising the version number, but I couldn't think of anything.
-------------------------------------------------------
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
-------------------------------------------------------
7:59 a.m.
Am 06.05.2021 um 09:33 schrieb Riccardo (Jack) Lucchetti:
On Wed, 5 May 2021, Sven Schreiber wrote:
> (@Jack, BTW: I guess the issue is practically moot now, but I think the
> package requirement would have to be 1.9.12 instead of 1.9.4.)
Do you think it worthwhile to update the package just for this? I had a
brief look at the package to see if there was something else worth
adding to justify raising the version number, but I couldn't think of
anything.
As I said, such an old version requirement is probably irrelevant now.
There may be a couple of minor points, though:
- the help text could perhaps use some reformatting (and typo: returs)
- About the grouping variable: is this supposed to be a binary dummy to
define the samples? Or at least discrete? If so, it doesn't seem to be
checked, I could execute the GUI function with arbitrary variables there
apparently. If not, maybe some more explanation could help.
- In general it's not quite clear to me why the second function combines
a different interface (list instead of series) with a different
statistic (median instead of mean). Again, maybe some explanation might
be good.
thanks
sven
11:04 a.m.
On Thu, 6 May 2021, Sven Schreiber wrote:
As I said, such an old version requirement is probably irrelevant
now.
There may be a couple of minor points, though:
- the help text could perhaps use some reformatting (and typo: returs)
- About the grouping variable: is this supposed to be a binary dummy to
define the samples? Or at least discrete? If so, it doesn't seem to be
checked, I could execute the GUI function with arbitrary variables there
apparently. If not, maybe some more explanation could help.
- In general it's not quite clear to me why the second function combines
a different interface (list instead of series) with a different
statistic (median instead of mean). Again, maybe some explanation might
be good.
I just uploaded for moderation a new version that should address these
issues. Thanks, Sven.
-------------------------------------------------------
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
-------------------------------------------------------
11:23 a.m.
Am 06.05.2021 um 13:04 schrieb Riccardo (Jack) Lucchetti:
On Thu, 6 May 2021, Sven Schreiber wrote:
> As I said, such an old version requirement is probably irrelevant now.
>
> There may be a couple of minor points, though:
> - the help text could perhaps use some reformatting (and typo: returs)
> - About the grouping variable: is this supposed to be a binary dummy to
> define the samples? Or at least discrete? If so, it doesn't seem to be
> checked, I could execute the GUI function with arbitrary variables there
> apparently. If not, maybe some more explanation could help.
>
> - In general it's not quite clear to me why the second function combines
> a different interface (list instead of series) with a different
> statistic (median instead of mean). Again, maybe some explanation might
> be good.
I just uploaded for moderation a new version that should address these
issues. Thanks, Sven.
OK, thanks, but now that you use the isdiscrete() function, your
requirement jumps a bit farther to 1.9.92 (2014c) according to the
changelog.
cheers
sven
12:27 p.m.
On Thu, 6 May 2021, Sven Schreiber wrote:
> I just uploaded for moderation a new version that should address
these
> issues. Thanks, Sven.
>
OK, thanks, but now that you use the isdiscrete() function, your
requirement jumps a bit farther to 1.9.92 (2014c) according to the
changelog.
You're right. Amended version on its way. While I was at it, I made the
criterion a bit more permissive: a series is deemed to be suitable as a
factor if it's marked as discrete or the number of values it takes is less
than sqrt($nobs).
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
1:58 p.m.
Am 06.05.2021 um 14:27 schrieb Riccardo (Jack) Lucchetti:
On Thu, 6 May 2021, Sven Schreiber wrote:
>> I just uploaded for moderation a new version that should address these
>> issues. Thanks, Sven.
>>
>
> OK, thanks, but now that you use the isdiscrete() function, your
> requirement jumps a bit farther to 1.9.92 (2014c) according to the
> changelog.
You're right. Amended version on its way. While I was at it, I made
the criterion a bit more permissive: a series is deemed to be suitable
as a factor if it's marked as discrete or the number of values it
takes is less than sqrt($nobs).
OK, the fixed version has been moderated and is online for everybody's
enjoyment.
thanks
sven
1089
days inactive
1090
days old
9 comments
3 participants
participants (3)
-
Riccardo (Jack) Lucchetti -
riles＠triton.net -
Sven Schreiber