IUBio

The fatal nicks in Occam's Razor

John M Price jmprice at calweb.com
Fri Apr 17 19:24:45 EST 1998


In sci.psychology.psychotherapy Etherman <etherman at mdc.net> wrote:

: John M Price wrote in message <353643cc.0 at calwebnnrp>...
: >And history is a humanities, not a science.
: >
: >And you still fail to grasp the concept of necessity.  Of course, your
: >straw Occham is a lot simpler than the real one, so I can understand your
: >mistake.


: One problem I see is as follows: How do you know what's necessary?

Depends.  I look for examples in the multivariate world, as that is how
most of the complex systems are addressed. 

For factor analysis, see the scree test.  Or, one could set the Eigenvalue
directly.  After a point, the additional factors don't add didly to the
solution, and the predictions derived from the solution don't become more
meaningful or accurate.

For multidimensional scaling, then the stress test.  Actually, more
specifically, the change in stress following adding a dimension to the
solution.  There is a point at which it becomes useless.  For most human
stuff, the solutions seem to resolve at three dimensions.  This also is
seen in the old Semantic Differential.

For multiple regression, then the change in R^2 for the whole solution.  
Tiny changes won't do much.

: Sure, we can think of trivial examples that look something like this:
: X or X plus some untestable hypothesis.  Obviously the former is
: favored because the untestable hypothesis is unnecessary.

What?  An hypothesis is generated by a theory.  It is not the number of
hypotheses we are discussing, rather the elements and relationships within
them as defined by the theory.

 But
: what if you have two theories that come at the same phenomena
: from completely different directions? Perhaps they both have
: the same number of hypotheses, so the doctrine of "use only
: what's necessary" isn't of any use.

See above.  Then you are out of the realm of Occham, and into the realm of
fitting the data again.  At that point, an analysis of the models is
necessary.  Only if they are equivalent in explaining the observations
would the razor be applied.

See the example of the Ptolemeic and Copernicon solar systems.

Or the Newtonian and Relativistic mechanics.

: Other criteria would have to be introduced.  For example, how
: well do the theories fit with the rest of science? I would also think
: that a theory that can be generalized easier would also be preferred.

Depends.  If it explains everything, it really explains nothing at all.


--
John M. Price, PhD                                     jmprice at calweb.com
Life: Chemistry, but with feeling!   |   PGP Key on request or by finger!
  Email responses to my Usenet articles will be posted at my discretion. 
Comoderator: sci.psychology.psychotherapy.moderated          Atheist# 683

"We will not be driven by fear into an age of unreason if we
remember that we are not descended from fearful men, men who
feared to write, to speak, to associate, and to defend causes
which were, for the moment, unpopular."
           - Edward R. Murrow, "See It Now", 7 March 1954




More information about the Neur-sci mailing list

Send comments to us at biosci-help [At] net.bio.net