data mining [Re: Robert X. Cringely's baby ...]
Mon, 29 Apr 2002 11:28:36 -0700
I am very sorry to hear such news,
and wish Mr. Cringely the best in
his endeavour. Part of that well-
wishing, I am afraid, comes in the
form of minor criticism.
> And rather than
> follow the traditional scientific method of first stating a hypothesis that
> we prove or disprove, I want the data to speak for itself. I want to get
> the best rocket scientists on Wall Street to apply their neural nets and
> other tools to divining from all this data a real leading indicator for SIDS.
Gathering such data may prove to
be very valuable; but one should
remember why we make hypotheses:
if a model has enough degrees of
freedom, it can fit any data.
If we are lucky, there will turn
out to be a strong signal in the
data. If we are unlucky, there
will be several spurious trends.
For example, it may turn out that
none of the infants who succumbed
to SIDS were in households with an
Arcimboldo, and all who were, were
healthy -- but one shouldn't then
recommend Arcimboldo acquisition
without further investigation.
(on a loosely related topic)
According to Larson's EH.net review,
Sutton asks: "Is it possible to find
economic models that work?" (such as
mechanics does for planetary motion,
or at least for tidal variations)
I'm not planning on tracking it down
myself, but here's its info:
John Sutton, _Marshall's Tendencies: What Can Economists Know?_
Cambridge, MA: MIT Press, 2000. xvi + 122 pp. $22.95 (cloth), ISBN:
0-262-19442-2; $14.95 (paperback), ISBN: 0-262-69279-1.