Re: [Gretl-users] Casewise Diagnostic
--- Sven Schreiber <svetosch(a)gmx.net> wrote:
Ignacio Diaz-Emparanza schrieb:
> El Wednesday 21 November 2007 13:38:52 Sven Schreiber escribió:
>
>> Having said that, in PcGive there is a post-estimation feature
that
>> reports observations with large residuals (per default I think
it's 3.5
>> standard devs, but it's adjustable). I have used that feature many
times
>> and maybe it's something to think about for gretl (after 1.6.6 of
course).
>>
>>
>
> In gretl we have the "Influencial observations" analysis. It is in
the model
> output menu under the "Tests" item.
>
>
>
Sometimes it's a bit embarrassing how little I know about gretl's
capabilities, even though I translated all of them...
Is that available for VARs/VECMs as well? (I should see for myself,
but
don't have time right now, so it's more a bit of a reminder to myself
to
check later...)
-sven
hahaha...
Yes I strongly agree with u Mr. Sven. I think I'm to hurry asking
someone for help before explore the biggest ability of gretl.
Yes I agree too that outlier detection is an option when we have
cross-sectional data.
However, there is an theory call A Winsorized Mean
(http://www.eminencegris.com/Winsor.htm) that is not to delete the
outlier observation but change the posision of outlier observation to
closesly to regression line. So, with this theory and techniq, we can
find the outlier too in time-series and panel data. Am I right :)
Yes Mr. Ignacio
I agree with u
But, Influencial observations too advance for me.
And there is just a little information in the result of Influencial
observations. I think the author should give * too in influence and
DFFITS beside leverage. Then the reader should have information for the
cutting point of the influence and DFFITS.
However :) now i know how to detect the outlier with simple techniques
just like casewise diagnostic in spss program.
In Model analysis > klik Analysis > Display Actual, fitted, residual
then u will see * that denotes a residual in excess of 2.5 standard
errors (that's outlier with 2.5 standard deviation) Why 2.5??? Not 3???
or 2??? :)
Then I go to data then delete the outlier observation one by one from
the biggest number to the smallest one (warning, from the biggest to
smallest observation only. You have to know why!) using Sample >
Restrict, based on criterion...
If we have 3 outlier in observation 23, 60, and 85, then i make
criterion like this:
obs!=85 then click ok
obs!=60 click to "add to current restrict" then click ok
obs!=23 click to "add to current restrict" then click ok
Affer that, i run the OLS again, the sample size will then minus 3
observation. Viola! my beta is more powerfull now :D
Note: deleting outlier does not always makes your beta more powerfull
:)
A joke:
I'm not an outlier; I just haven't found my distribution yet!
*Thanks to Ronan Conroy in Dublin, Ireland for this real cute one. I
have also been searching for my distribution throughout my entire
career.
http://www.cos.name/bbs/read.php?tid=6618&page=e&fpage=1
Best Regards For All
Novan AS
____________________________________________________________________________________
Be a better sports nut! Let your teams follow you
with Yahoo Mobile. Try it now.
http://mobile.yahoo.com/sports;_ylt=At9_qDKvtAbMuh1G1SQtBI7ntAcJ