Re: Adjustment for confounding

[Bernard L Cohen <blc+@pitt.edu>]

Bernard L. Cohen
Physics Dept.
University of Pittsburgh
Pittsburgh, PA 15260
Tel: (412)624-9245
Fax: (412)624-9163
e-mail: blc+@pitt.edu


On Thu, 9 Mar 2000 FIELDRW@aol.com wrote:

> In a message dated 3/9/00 11:09:39 AM Central Standard Time, blc+@pitt.edu 
> writes:
> 
> <<  --My study avoids the problem of non-linearity by using
>  "Stratification". For example, if one worries about population density as
>  a confounding factor, I do a complete analysis utilizing only data from
>  counties that have nearly the same population density. I do this for many
>  different population densities, and I always get the same result, without
>  assuming that anything is linear with population density.
>     The reason I am able to do this is because I have 1600 counties in
>  my data files, whereas 160 counties is enough for doing a complete
>  analysis with little statistical uncertainty. 
>     I also have a treatment of multiple confounding factors, but that
>  is too complex to summarize here.  >>
> 
> 
> Dr. Cohen, You are not able to adjust for multiple differences within or 
> between counties at the same time such as differences is smoking rates, age, 
> education etc, etc.  Adjustment using more summary averaged data is 
> inadequate to control for confounding.

	--I don't "adjust" for anything. The dependence on smoking
prevalence, S, in my papers arises from the fact that BEIR Reports treat
smokers and non-smokers as two different species with different lung
cancer risks, whereas county lung cancer statistics do not differentiate
between smokers and non-smokers. Thus, S represents the fraction of each
species that are included in the county statistics. There is no assumption
of linearity, and if S-values, are accurate, there is no uncertainty in
this procedure.
	Your issue of simultaneously treating variations in age,
education, etc is covered in Sec. I of my 1995 paper. But if it is not
convincing, you should be able to concoct a potential explanation for our
results based on its shortcomings. Your potential explanation need not be
correct; once you propose it in a paper accepted for publication in Health
Physics, you get the $1000 reward. It is then my responsibility to analyze
it for plausibility. If I cannot show that it is very highly implausible,
I will publicly concede that my conclusions about failure of LNT were not
justified.


Bernard L. Cohen
Physics Dept.
University of Pittsburgh
Pittsburgh, PA 15260
Tel: (412)624-9245
Fax: (412)624-9163
e-mail: blc+@pitt.edu


************************************************************************
The RADSAFE Frequently Asked Questions list, archives and subscription
information can be accessed at http://www.ehs.uiuc.edu/~rad/radsafe.html