Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!news.ultranet.com!news.sprintlink.net!redstone.interpath.net!sas!mozart.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: bias
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <D865H7.Mz@unx.sas.com>
Date: Sat, 6 May 1995 18:17:31 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References: <1995Apr28.195623.7533@cm.cf.ac.uk> <D7t3E1.ECz@unx.sas.com> <799200709snz@longley.demon.co.uk> <D7x5u6.Fvo@unx.sas.com> <799411757snz@longley.demon.co.uk>
Organization: SAS Institute Inc.
Lines: 84


In article <799411757snz@longley.demon.co.uk>, David@longley.demon.co.uk (David Longley) writes:
|> ...
|> As  I tried  to elaborate  in a  series of articles  here on  the  net
|> 'Fragments of Behaviour' 1 - 9 ...25/4/95) my ultimate concern is the relative
|> merits of actuarial vs. clinical judgment.

As pointed out recently on the stat-l list, dentists seem to be better
at clinical judgments than are physicians and psychiatrists. It was
hypothesized that the better judgment of dentists is due to the fact
that dentists regularly examine healthy people as well as sick people,
hence the dentists have better training data.

   Alanen P, Hurskainen K, Isokangas P, et al.  Clinicians ability to
   identify caries risk subjects. Comm Dent Oral Epidemiol 1994; 22: 86-9.

|> ...
|> As I see it, the  *key* difference between NNs  and Regression techniques such
|> as Logistic Regression (which also  uses a  squashing function to  generate  a
|> probability) is that the latter is designed to give  you weightings  for  your
|> input (independent) variables, whilst  NNs do not,  and can not  in  principle

It is true that there are some methods, such linear or logistic
regression and perceptrons, that assign a single weight to each input
variable, and there are other methods, such as multilayer perceptrons
and polynomial regression, that involve weights that do not correspond
to single inputs. But regression techniques belong to both classes, and
so do NNs.

|> ...
|> Here, I think conventional  statisticians  are 'hiding  their  lights..'.  The
|> ability to step through a multiple regression equation programme & say exactly
|> what each step is doing algorithmically (effectively) is the real value of the
|> technology,

It is certainly a great advantage if one can fit a model that can be
understood in that manner, and that is one reason that many people
prefer tree-growing methods (CART, CHAID, C4.5, etc.) to MLPs.  But
sometimes the data are not amenable to such models and it becomes
necessary to use models such as MLPs or kernel regression that are not
so easily interpretable.

|> as doing so, along with extracting & analysing  residuals, looking
|> at measures of fit and so on

One can do those things with NNs just as well as with linear regression.

|> is analysing *extensionally* ie according to  the
|> explicit  principles embodied  in the  predicate  calculus  (substitutivity of
|> identicals is of course essential for solution of simultaneous equations).

You lost me there.

|> Having said that, I  *do* think  that  conventional  stats technology could be
|> marketed better.

Yes, the neural net people have clearly out-marketed the statisticians.
"Learning concepts" makes much better ad copy than "estimating discriminant
functions".

|> This is just a quick response to say that I'd like to pursue the comparison of
|> conventional regression technology, cluster analysis, discriminant etc and NNs
|> further - thanks again.

The articles in the following files, available by anonymous ftp from
ftp.sas.com (Internet gateway IP 192.35.83.8) in the directory
/pub/sugi19/neural, go into considerable detail on those issues:

 neural1.ps     Sarle, W.S. (1994), "Neural Networks and Statistical
                Models," Proceedings of the Nineteenth Annual SAS Users
                Group International Conference, Cary, NC: SAS Institute,
                pp 1538-1550. (Postscript file)

 neural2.ps     Sarle, W.S. (1994), "Neural Network Implementation in
                SAS Software," Proceedings of the Nineteenth Annual SAS
                Users Group International Conference, Cary, NC: SAS
                Institute, pp 1551-1573. (Slightly revised version,
                postscript file)

-- 

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.
