Re: [Gretl-devel] Proposal of new feature
On Wed, 2 Mar 2011, Marcin B�^Bażejowski wrote:
in PDF there is very short proposal for feature similar to Ox's
"outliers detection". Idea is very simple: for each observation when
residual exceeds given broad value (i.e. u_{\alpha}\cdot S_e) then we
create dummy variable. The maximum number of dummies is called "dummy
saturation" and is limited only by degrees of freedom.
Marcin, in your mock-up you show this procedure as somehow tied to
a selection of variables. I don't understand that; can you explain
what you have in mind?
The procedure you're suggesting is not a test (I mean, there's no
null hypothesis), so it probably should not go under the Tests
menu. And I'm not sure that creation of a (possibly large) set of
single-observation dummies is very helpful to the GUI user. Might
it not be more helpful to offer something under Edit/Modify model,
along the lines of "Exclude observations with residuals greater
than <select N> * sigma in absolute value"?
Allin