Re: Mossman's confusion

["Dukelow, James S Jr" <jim.dukelow@pnl.gov>]

Ken Mossman wrote:
------------------------------

Date: Fri, 17 Mar 2000 11:24:12 -0700
From: Ken Mossman <Ken.Mossman@asu.edu>
Subject: Cohen's ecological data: a test of LNT?

This my last shot!

Cohen and I agree that LNT is a causal relationship. Howevever, I am now
totally confused about whether Cohen's data tests for a causal relationship.
In his April 1997 paper in Health Physics (page 625), Cohen indicates that
his data do not test for a causal relationship but has a much more limited
objective. Now Cohen says that his data tests for a causal relationship.
Which is  it? Either the data test for a causal relationship or thery don't!

If the data test for a causal relationship, as Cohen now states, then the
ecological fallacy becomes operative. If the data do not test for a causal
relationbship, then the data are not a test of LNT because LNT is a causal
relationship (as Cohen and I now agree).


Kenneth L. Mossman
Professor of Health Physics
Director, Office of Radiation Safety
Arizona State University

==============================

I agree that Professor Mossman is totally confused on this point.

On the other hand, Professor Cohen's explanations of what he was doing with his
data, dating back to the 1995 Health Physics paper and in his recent responses
to Field, Lubin, and Mossman have been straightforward and clearly expressed in
standard English.

The LNTH is one possible causal relationship among an infinite family of
possible causal relationships.  LNTH is important because some people actually
believe that it represents the scientific reality of the relationship between
exposure to radiation and subsequent cancer.

One of the things that Cohen could have done with his data was to construct a
regression or a maximum likelihood estimate of the relationship between radon
concentrations and excess cancers and used that to propose a causal
dose-response relationship.  As he has noted many times, doing that would be
subject to the ecological fallacy, and he DID NOT do that.

What he did do was take advantage of the fact that, as a putative dose response
relationship, the LNTH makes certain predictions about how many lung cancers
ought to correspond to how much exposure to radon.  Cohen compared the predicted
number of cancers to the actual number of cancers.  He discovered that, for most
of the range of values of radon exposures, the predicted values are nowhere
close to the actual values and that this result is extraordinarily statistically
significant.  This is a classical Popperian test of hypothesis, and absent
explanations of the data related to confounding, requires the rejection of the
hypothesis, at least in the radon/lung cancer context.

His test of LNTH does not suffer from the ecological fallacy precisely because
the LNTH is such a "special" hypothesis.  The linear coefficient that defines
the relationship can be moved in or out of the integrals or summations used to
calculate the predicted number of cancers, independently of how the radiation
dose is distributed.  It is the only proposed dose-repsonse relationship for
which the concept of a person-rem and a dose-response relationship of so many
cancers per person-rem (no matter how the exposure is distributed among the
persons) makes scientific sense.  Verifying this is an exercise in freshman
calculus.

I might point out, in passing, that anyone who is terrified of the ecological
fallacy, to the extent that they will give no credence whatsoever to data that
might suffer from the ecological fallacy, needs to make sure their health never
depends on use of one of the many diagnostic techniques based on tomography.
They all suffer equally from the ecological fallacy.  

In addition, Professor Gary King of Harvard won the 1995-96 Gosnell Prize of the
American Political Science Association for the best methodological work during
that biennium.  The ecological fallacy was first identified by political
scientists and they have remained interested in finding some way around it,
since so much of their fundamental data is aggregate data -- the privacy of the
voting booth, and all that.  King's method for avoiding the ecological fallacy
in aggregate data, which borrows some of the mathematical techniques used in
tomography, is described in his book _A Solution to the Ecological Inference
Problem: Reconstructing Individual Behavior from Aggregate Data_, published by
Princeton University Press in 1997, available in paper for around $17.  King has
put the computer code implementing his algorithm on his web page
<GKing.harvard.edu>, so anyone interested can download it and play with it.  For
a more recent update on King's method and responses to it, you can read the
paper "Ecological Inference" by Alexander Schuessler of New York University in
the 14 Sept 1999 issue of the Proceedings of the National Academy of Sciences,
which is now available online at  <www.nas.edu>  under Publications.

It is past time for epidemiologists to quit whining about the ecological fallacy
when confronted with ecological data and join the 21st century -- or whatever
century we are in right now.

Best regards.

Jim Dukelow
Pacific Northwest National Laboratory
Richland, WA
jim.dukelow@pnl.gov

These comments are mine and have not been reviewed and/or approved by my
management or by the U.S. Department of Energy.
************************************************************************
The RADSAFE Frequently Asked Questions list, archives and subscription
information can be accessed at http://www.ehs.uiuc.edu/~rad/radsafe.html