Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!godot.cc.duq.edu!news.duke.edu!news.mathworks.com!newsfeed.internetmci.com!in1.uu.net!news.interpath.net!sas!newshost.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: What are p-values (t-stats?)
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <DquvIu.6Dn@unx.sas.com>
Date: Sat, 4 May 1996 00:47:18 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References: <19960501084927.BKAMP@pi0192.kub.nl> <31876880.5149@clark.net>
Organization: SAS Institute Inc.
Lines: 39


Bart Kamp asked:
> 
> does anyone know what 'p values' are? It has something to do with
> the validity testing of NN. I guess it is something like t-statis-
> tics.

In article <31876880.5149@clark.net>, "David H. Uthe" <uthed@clark.net>
replied:
|> P-values, when used in testing statistics, are nothing more than the 
|> proportion of persons that choose a particular response. P-values are 
|> tracked for each response, A-B-C-D, but the "item's" p-value refers to 
|> the keyed response.

I have never heard of the term "p-value" used in that sense. P-values
are an aspect of statistical hypothesis testing covered in most any
introductory statistics text. But you have to understand sampling
distributions, null hypotheses, and error rates before you can
understand p-values, and I don't have time to explain all of that
right now. There is a brief explanation of these concepts on pp. 5-26
of the _SAS Procedures Guide_, version 6, 3rd edition.

The application of these concepts to neural nets raises problems
with nonidentifiability and violation of the usual regularity
conditions. If you really want p-values for neural nets, you should
probably try bootstrapping:

   Efron, B. and Tibshirani, R.J. (1993), _An Introduction to the
   Bootstrap_, London: Chapman & Hall.

   Hjorth, J.S.U. (1994), _Computer Intensive Statistical Methods:
   Validation, Model Selection, and Bootstrap_, London: Chapman & Hall.


-- 

Warren S. Sarle       SAS Institute Inc.   The opinions expressed here
saswss@unx.sas.com    SAS Campus Drive     are mine and not necessarily
(919) 677-8000        Cary, NC 27513, USA  those of SAS Institute.
