Herman Rubin wrote:
>
> In article <[EMAIL PROTECTED]>,
> Frank E Harrell Jr <[EMAIL PROTECTED]> wrote:
> >To suggest that ANY method of modeling or data analysis can reliably infer causation
> >from non-experimental data is bogus. Even in a parallel group double blind
> >randomized controlled trial, in which there is tight experimental manipulation,
> >an assumption is required to infer causation due to treatment effects, for
> >individual patients.
>
> This is hardly the case. One does infer causation in such
> fields as geology, astronomy, and meteorology, and also in
> archaeology and paleontology. It is even done in linguistics.
>
> How reliable it is does vary from field to field, but it
> not only can be done, but should. It takes a more careful
> decision-theoretic analysis, but one can only act on the
> available information, and often one must act.
It's hard to disagree with either statement. The problem is
what we infer from "infer". Once upon a time, a key reference
in this area was David Kenny's Correlation & Causality. Today,
I believe many practitioners look to Kenneth Bollen's Structural
Equations With Latent Variables. It's been a few years since
I read it, but the two things that stick with me are Figure 3.9
showing 10 different models for three observed variables that
have perfect fit to a particular covariance matrix and Bollen's
statement on page 72, "We can only reject a model--we can never
prove it to be valid." Thus, I agree with Harrell that
statistical methods alone can never "reliably infer" (prove)
causality, but I also agree with Rubin that statistical evidence
is an important piece of the puzzle when trying to infer (seeing whether
the data are consistent with) a causal relationship. Try
as I might I can't stop the urge to rephrase the last part of the
last sentence to, "(seeing whether the data are not inconsistent
with all of the implications of)". In my experience, I tend to
see people use the word infer in the sense of "prove". My
canned response is that statistical methods alone can never prove
*anything*.
never prove caus
===========================================================================
This list is open to everyone. Occasionally, people lacking respect
for other members of the list send messages that are inappropriate
or unrelated to the list's discussion topics. Please just delete the
offensive email.
For information concerning the list, please see the following web page:
http://jse.stat.ncsu.edu/
===========================================================================
