[Octave-patch-tracker] [patch #8584] statistics/stepwisefit selection on partial p values

[Mikael Kurula]

URL:
  <http://savannah.gnu.org/patch/?8584>

                 Summary: statistics/stepwisefit selection on partial p values
                 Project: GNU Octave
            Submitted by: alcarola
            Submitted on: Sun 21 Dec 2014 07:36:44 PM GMT
                Category: None
                Priority: 5 - Normal
                  Status: None
                 Privacy: Public
             Assigned to: None
        Originator Email: 
             Open/Closed: Open
         Discussion Lock: Any

    _______________________________________________________

Details:

I made some rather trivial modifications to Nir Krakauer's stepwisefit.m in
the statistics package, in order to select predictors to consider for
inclusion in the models based on partial p values, rather than on correlation
with the residual. This is as far as I can understand also suggested in
Section 2.9 of Draper & Smith: Applied Regression Analysis, second edition,
Wiley 1981.

Drawbacks:
- Computationally probably more intense than correlation
- I am not a statistician, so not so confident that I got it all right

Advantages:
+ Partial p values should be a much more reliable measure of the significance
of a predictor than correlation with the residual
+ stepwisefit_p.m seems to perform much better than stepwisefit.m on at least
some examples arising from CCD experimental designs

The Octave community is of course free to do what it finds fit with this small
development. Thanks to Nir for suggesting that I post the modification here.



    _______________________________________________________

File Attachments:


-------------------------------------------------------
Date: Sun 21 Dec 2014 07:36:44 PM GMT  Name: stepwisefit_p.m  Size: 6kB   By:
alcarola

<http://savannah.gnu.org/patch/download.php?file_id=32698>

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/patch/?8584>

_______________________________________________
  Message sent via/by Savannah
  http://savannah.gnu.org/