2:31 p.m.
Dear all,
for an update and simplification of the FEP package I would to replace a
user-own function by xtab. However, xtab lacks currently a few features
for using it within a function:
1) accessor for p-values and test-statistics; in my case, for instance,
I just need information about Pearson's chi^2-test.
2) --silent option
Is there a way to access these statistics somehow?
Here is an illustration:
<hansl>
set verbose off
open denmark.gdt -q
series mup = (diff(LRM)>0)
series yup = (diff(LRY)>0)
list L = mup yup
xtab L
eval $pvalue
eval $test
</hansl>
Thanks,
Artur
3:35 p.m.
On Fri, 5 Jan 2018, A. T. wrote:
Dear all,
for an update and simplification of the FEP package I would to replace a
user-own function by xtab. However, xtab lacks currently a few features
for using it within a function:
1) accessor for p-values and test-statistics; in my case, for instance,
I just need information about Pearson's chi^2-test.
2) --silent option
Is there a way to access these statistics somehow?
Here is an illustration:
<hansl>
set verbose off
open denmark.gdt -q
series mup = (diff(LRM)>0)
series yup = (diff(LRY)>0)
list L = mup yup
xtab L
eval $pvalue
eval $test
</hansl>
Hope this helps:
<hansl>
set verbose off
open denmark.gdt -q
series mup = (diff(LRM)>0)
series yup = (diff(LRY)>0)
list L = mup yup
xtab L
matrix A = mxtab(mup, yup)
matrix c = sumr(A)
matrix r = sumc(A)
scalar n = sumc(c)
matrix B = c*r/n
P = (A - B).^2 ./ B
PearChi2 = sumc(vec(P))
</hansl>
-------------------------------------------------------
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
-------------------------------------------------------
4:10 p.m.
Am 05.01.2018 um 16:35 schrieb Riccardo (Jack) Lucchetti:
On Fri, 5 Jan 2018, A. T. wrote:
Hope this helps:
<hansl>
set verbose off
open denmark.gdt -q
series mup = (diff(LRM)>0)
series yup = (diff(LRY)>0)
list L = mup yup
xtab L
matrix A = mxtab(mup, yup)
matrix c = sumr(A)
matrix r = sumc(A)
scalar n = sumc(c)
matrix B = c*r/n
P = (A - B).^2 ./ B
PearChi2 = sumc(vec(P))
</hansl>
This is of course very nice and welcome, but just to be clear, I don't
think Artur was claiming that it's very difficult to calculate the chi2
independence test manually. Just that a slightly extended interface to
the existing gretl routine would be useful.
cheers,
sven
4:46 p.m.
> Hope this helps:
Thanks Jack, this works nicely an perfectly replicates my own version
which is much more complicated. Would is be ok for you if we use your
version for the FEP package?
This is of course very nice and welcome, but just to be clear, I
don't
think Artur was claiming that it's very difficult to calculate the chi2
independence test manually. Just that a slightly extended interface to
the existing gretl routine would be useful.
Exactly. Fully support this! ;-)
Best,
Artur
8:44 p.m.
On Fri, 5 Jan 2018, A. T. wrote:
Dear all,
for an update and simplification of the FEP package I would to replace a
user-own function by xtab. However, xtab lacks currently a few features
for using it within a function:
1) accessor for p-values and test-statistics; in my case, for instance,
I just need information about Pearson's chi^2-test.
2) --silent option
Is there a way to access these statistics somehow?
Not at present. And given that the "xtab" command accepts lists for
both x and y arguments it's not obvious that it could be easily
arranged. I guess we could offer a --silent option and storage of
(Pearson) $test and $pvalue conditional on getting just one x and y.
Jack, thoughts on that?
Allin
7:16 a.m.
Am 09.01.2018 um 21:44 schrieb Allin Cottrell:
On Fri, 5 Jan 2018, A. T. wrote:
> 1) accessor for p-values and test-statistics; in my case, for
instance,
> I just need information about Pearson's chi^2-test.
> 2) --silent option
>
> Is there a way to access these statistics somehow?
Not at present. And given that the "xtab" command accepts lists for both
Hm, BTW more warning messages might be useful:
<hansl>
open credscore.gdt
list x = Acc Age # Acc is binary, Age is positive integers
xtab MDR x # w/o comment only reports the combination MDR vs. Acc
discrete Age # necessary because Age has many values
xtab MDR x # reports both combinations
</hansl>
x and y arguments it's not obvious that it could be easily
arranged.
I guess it could be as symmetric matrices, similar to the arrangement of
a covariance matrix. The diagonal could be left empty or the p-values
there could be set to 0. ($test and $pvalue are already sometimes
matrices as per the doc.)
guess we could offer a --silent option and storage of (Pearson) $test
and $pvalue conditional on getting just one x and y.
'--silent' sounds good.
While we're at it, getting the Fisher test results (if applicable, 2x2
crosstab) would also be nice. I guess this could really be made
conditional on just a single x and y variable.
thanks,
sven
6:35 p.m.
On Wed, 10 Jan 2018, Sven Schreiber wrote:
Am 09.01.2018 um 21:44 schrieb Allin Cottrell:
> On Fri, 5 Jan 2018, A. T. wrote:
>> 1) accessor for p-values and test-statistics; in my case, for instance,
>> I just need information about Pearson's chi^2-test.
>> 2) --silent option
>>
>> Is there a way to access these statistics somehow?
>
> Not at present. And given that the "xtab" command accepts lists for both
Hm, BTW more warning messages might be useful:
<hansl>
open credscore.gdt
list x = Acc Age # Acc is binary, Age is positive integers
xtab MDR x # w/o comment only reports the combination MDR vs. Acc
discrete Age # necessary because Age has many values
xtab MDR x # reports both combinations
</hansl>
OK, in git you now get notified if an "xtab" argument is dropped as
non-discrete.
> x and y arguments it's not obvious that it could be easily
arranged.
I guess it could be as symmetric matrices, similar to the arrangement of a
covariance matrix. The diagonal could be left empty or the p-values there
could be set to 0. ($test and $pvalue are already sometimes matrices as per
the doc.)
> guess we could offer a --silent option and storage of (Pearson) $test and
> $pvalue conditional on getting just one x and y.
'--silent' sounds good.
While we're at it, getting the Fisher test results (if applicable, 2x2
crosstab) would also be nice. I guess this could really be made conditional
on just a single x and y variable.
My feeling is that the above is getting too complicated. For now,
here's what we have (git/snapshots):
* In the simple bivariate case (only) $test and $pvalue are available
for the Pearson chi-square test and its p-value -- provided the
minimum expected value condition is met. The Fisher test is sui
generis (we print 3 p-values but no test statistic); I think that if
there's a demand for that we might add a function to do it.
* There's now a --quiet option (which really only makes sense in the
bivariate case).
(We offer --silent only if --quiet is not silent, but in this case I
don't think there's any need for degrees of quietness.)
I've also fixed a long-standing issue that apparently hasn't been
noticed: if you give the --matrix=name option to xtab we were
insisting on a list argument, which was then ignored. So far as I can
see there's no call for a list arg in the --matrix case, so I've
dropped the requirement (and giving a list will now provoke an error).
Allin
9:25 p.m.
Am 10.01.2018 um 19:35 schrieb Allin Cottrell:
On Wed, 10 Jan 2018, Sven Schreiber wrote:
> Hm, BTW more warning messages might be useful:
OK, in git you now get notified if an "xtab" argument is
dropped as
non-discrete.
good
My feeling is that the above is getting too complicated. For now,
here's
what we have (git/snapshots):
* In the simple bivariate case (only) $test and $pvalue are available
for the Pearson chi-square test and its p-value -- provided the minimum
expected value condition is met.
Yeah, I think this is probably at least 90% of what might be needed.
The Fisher test is sui generis (we
print 3 p-values but no test statistic); I think that if there's
a
demand for that we might add a function to do it.
I recently put a function for 2x2 tables into the "extra" package. I
guess the return vector of that function could be extended by the Fisher
test p-values. Might be a hard-to-find place, though.
I've also fixed a long-standing issue that apparently hasn't
been
noticed:
yep, certainly not noticed by me!
thanks,
sven
2578
days inactive
2583
days old
7 comments
4 participants
participants (4)
-
A. T. -
Allin Cottrell -
Riccardo (Jack) Lucchetti -
Sven Schreiber