Medicine's Ten Greatest Discoveries

Steven B. Harris sbharris at ix.netcom.com
Sun Sep 6 03:54:17 EST 1998


In <EQrI1.2130$c3.3697154 at tor-nn1.netcom.ca> "David Lloyd-Jones"
<dlj at pobox.com> writes: 

>In somewhat the same sense, multiple regression analysis, which was
>destroyed epistemologically by Locke and Hume a couple of centuries
before
>it actually came into general use, has no claim whatsoever to Platonic
>logical rigor -- but is one of the most useful forms of explanation
that we
>have.  Usually right, on rare occasions wrong.


   In your dreams.  Logistic regression is mainly good for ruling out
causation (in which case it really is usually right, and only on rare
occations wrong).  But for cases where variables correlate, the usual
case is that direct causal relationships cannot be inferred, and quite
often do not exist.  Rather, correlations are usually caused by third
factors which are causal to both variables, which have no actual
relationship with each other.  Such confounders or proxy variables
bedevil every multiple logistic regression ever done.  If not,
prospective studies and experimentation itself would hardly be
necessary.  All we'd have to do is epidemiology.  If black people score
less well on intelligence tests, why then, being black must cause brain
problems.  If gay men use a lot of nitrite recreational drugs and get
AIDS, and the more nitrities they use the more likely they are to get
AIDS, why then nitrites must cause AIDS.  If married people live
longer, why then marriage must make you live longer.  If people who jog
every morning live longer, why then jogging every morning must make you
live longer.  If there is more juvanile crime during those months in
which ice cream sales are highest, it must be that ice cream causes
deliquency.  And so on and so on, into foolishness ad infinitum.



More information about the Bioforum mailing list