[ RadSafe ] Cohen's radon results differently
Bernard L. Cohen
blc at pitt.edu
Tue Sep 20 12:04:36 CDT 2011
I feel strongly that the scientific method does not allow one to play
around with smoking prevalence to get some result, but that it must be
derived from some logical independent source. If this restriction is
removed, by playing around with smoking prevalences I can get any result
you might specify, but that is mathematics, not science.
On 9/20/2011 12:44 PM, Chris Hofmeyr wrote:
> Prof Cohen,
> It is to be welcomed that you identify the determination of smoking prevalence
> as a basic problem.
> If my information is correct, you did consider using LC to deduce smoking, but
> that would obviously have obscured the effect you were trying to see. The use
> of cigarette sales was therefore probably a safer choice, but the result when
> plotting LC against smoking shows a rather much larger spread (Figure 2e of my
> notes) than it would have been if smoking were defined in terms of LC! A more
> realistic assumption might lie somewhere between these, but it would remain an
> I honestly believe that the smoking prevalence figures might not be completely
> above reproach, but they are probably not totally wrong. After all, they do
> show the same trend as the LC plots against the county variable ranks, although
> not as prominently. I would suggest that a stronger correction function should
> be investigated.
> One can discuss what a reliable smoking correction to lung cancer (LC)
> mortality would depend on, e.g.
> 1. good smoking prevalence statistics at the county level
> 2. a trustworthy correction formula,
> 3. some indication that treating all smokers equally is warranted, and
> 4. some kind of reality check to prove consistency.
> 5. The large increase in female LC between the two periods of observation cries
> for an explanation. What carcinogen apart from smoking could conceivably be
> However, I would welcome some comment on my proposal that the plotting of LC
> against county RANK of a variable like radon or elevation can provide
> sufficient proof of LC independence of the variable, and that a smoking
> correction is not absolutely required under the given circumstances.
> chris.hofmeyr at webmail.co.za
> On Mon, 19 Sep 2011 18:55:01 -0400 "Bernard L. Cohen"<blc at pitt.edu> wrote
>> To do anything, I would need some credible source of smoking prevalence
>> data. I tried several and they did not change my results. Please suggest
>> an alternative.
>> On 9/18/2011 3:31 PM, Chris Hofmeyr wrote:
>>> On 19th June Howard Long challenged my reminder that in Cohen's US-wide
>>> of domestic radon and lung cancer, there was, on average, a negative
>>> correlation between Cohen's smoking prevalence and the average county radon
>>> concentration. Howard wanted numbers.
>>> On 20th June prof Cohen invited further discussion of his work. Then away
>>> an extended stay, I told dr Long that I will have a look when I get back
>>> B Cohen and Wes van Pelt had kindly made their data available in
>>> form almost a decade (and a few computers)ago. At the time and later I did
>>> numerous plots and wrote some notes, which I now found on an archive disk.
>>> I put some of it together in a .pdf file of about 1.5MB (attached); please
>>> excuse some repetition in the text. If the attachment does not survive
>>> then please request directly from me at webmail address below.
>>> Some background for the younger crowd:
>>> Cohen set out to study lung cancer rates as a function of county average
>>> domestic radon concentration (ARC)in 1601 US counties. His null hypothesis
>>> presumably to justify LNT (Linear No-Threshold model of carcinogenesis).
>>> However, fate stacked the cards in that smoking prevalence, as determined
>>> him, was to a degree anti-correlated with average county radon. Due to the
>>> carcinogenic strength of smoking, the raw data naturally reflected an
>>> anti-correlation with radon concentration.
>>> This meant that the lung cancer data would have to be carefully corrected
>>> extract a 'real' correlation, which Cohen probably only realized when he
>>> analyzing the data. Cohen tried more than one avenue to assess smoking
>>> and he was obviously reasonably satisfied with his final choice, although
>>> correction strangely left the anti-correlation practically unchanged.
>>> Having investigated lung cancer (LC) - Cohen's and Van Pelt's data - also
>>> relation to elevation and county population size, I am practically forced
>>> the conclusion that there appears to be a strong smoking residual even
>>> the correction. LC shows an even stronger anti-correlation with altitude
>>> 1260ft than with radon (ARC) below about 3pCi/L,yet (LC) proves
>>> independence of both.
>>> Cohen tested several hundred possible confounders, but very few managed to
>>> turn the LC
>>> correlation with Average radon positive.
>>> I would respectfully ask prof Cohen: SUPPOSE the smoking correction was
>>> insufficient and left an implicit smoking component in the LC data?
>>> In the notes I explain how one can use a plotting trick on Cohen's data to
>>> evaluate LC dependence (or rather independence) on ARC even without a
>>> correction to the LC data - and consequently dispose of LNT.
>>> chris.hofmeyr at webmail.co.za<mailto:chris.hofmeyr at webmail.co.za>
>> Bernard L. Cohen
>> Physics Dept., University of Pittsburgh
>> Pittsburgh, PA 15260
>> Tel: (412)624-9245 Fax: (412)624-9163
>> e-mail: blc at pitt.edu web site: http://www.phyast.pitt.edu/~blc
> South Africas premier free email service - www.webmail.co.za
> Save on insurance with OUTsurance
Bernard L. Cohen
Physics Dept., University of Pittsburgh
Pittsburgh, PA 15260
Tel: (412)624-9245 Fax: (412)624-9163
e-mail: blc at pitt.edu web site: http://www.phyast.pitt.edu/~blc
More information about the RadSafe