Re: The Puskin Paper on radon and smoking

[BERNARD L COHEN <blc+@PITT.EDU>]

>

> Cohen's Response:



> Cohen also offers a "test" of the Puskin hypothesis (and I don't agree with

> it). It goes like this:

>

> Cohen admits that there is a negative correlation between radon

> concentration and smoking, but claims that the correlation is not strong

> enough to account for the excess lung cancers found in low radon areas.



	--Puskin claims that the explanation of his observation is that my

smoking data are erroneous, missing a strong negative correlation between

smoking prevalence, S, and radon exposure, r. I show that there is no

possible set of S-values that will lessen the problem posed by the Puskin

observation.



> He goes on to say that, even if he ordered the counties according to smoking

> prevalence and assumed a perfect negative correlation with the average radon

> concentration, that would not be sufficient to explain the negative

> correlation between average radon concentration and lung cancer. In other

> words, there is not that much difference in smoking prevalence between

> counties to explain the differences in lung cancer, regardless if the low

> radon counties happen to be at the top or the bottom of the smoking scale.



	--With the Puskin type analysis,  differences in S-values between

counties does not matter. The results listed in the column headed

"Corr(S-r)=-1.0" in Table 1 of Item #7 on my web site show that there is

no indication that the dependence of cancer rate on radon exposure is more

positive for lung cancer than for the other smoking related cancers; if

anything, the contrary seems to be the case, so the Puskin hypothesis

(strong negative S-r correlation) does nothing to explain the Puskin

observation.  The results in that column are independent of the width of

the S- distribution. For example, if this width is doubled, the values in

that column remain unchanged. I also consider changing the shape of the

S-distribution, and this does not help.



>

> This does not make any sense to me: Either you have good smoking data, then

> your ranking will be pretty good and you don't have to reorder the

counties,

> or your data is bad and then your ranking may or may not be correct, but the

> width of the distribution is certainly biased toward the null. Consider the

> following example:

>

	--I can understand your puzzlement about the fact that the width

of the S-distribution does not matter. In my original treatment, I show

that it does matter, but that the width required to explain the data is

highly implausible. However, in the Puskin treatment, S and r are treated

as independent variables in a multivariate regression, and for reasons I

don't understand, values in the column labelled "Corr(S,r)=-1.0" are

independent of the width.

	To check on this, I used two uniform distributions of S-values,

one ranging from 90-110, and the other ranging from 70-130, three times as

wide.With a perfect negative correlation with radon in each case, the

values in the column labelled "Corr(S,r)=1.0" are the same.





> There are two counties. County 1 has 100% non-smokers and county 2 has 100%

> smokers. You decide to determine smoking status by asking everyone to

> complete a survey. Half of everyone in each county does a good job filling

> out your survey. The other half randomly checks off yes or no, without

> reading your question. Your survey will then show county 1 has 75%

> non-smokers and 25% smokers while county 2 has 75% smokers and 25%

> non-smokers. Your ranking is still correct (county 1 has more non-smokers)

> but the measured width of your distribution is only half of the correct

> width. The true width of the distribution might be enough to account for

> differences in lung cancer incidence, while the measured width might not be.



	--See my above discussion showing that, in the Puskin analysis,

the width does not matter.



> Cohen also explains the sources of his smoking data and points out that all

> 3 independent methods of estimating smoking produce similar results. To me,

> that is more convincing than the reordering "test". It is also impressive

> that no one has yet proposed a reasonable smoking distribution, which would

> make Cohen's results compatible with LNT.

>

> Conclusion:

>

> I hope that more people follow Puskin's example and offer new input into the

> debate. Far too few people are involved in the scientific part of the radon

> debate. (There are plenty of people on the standard setting side of the

> debate.)

>

> I hope that no one (including the NCRP) will consider the findings of the

> NCRP or any other group as final, until the theory is shown to be compatible

> with the data. I have no problem with NCRP or ICRP recommendations being

> used for setting standards, however. After all, the purpose of committees is

> not to do any science, but rather to build consensus by applying reasonable

> and conservative principles. (Remember what the RP stands for!). They are on

> the standard setting side of the debate.

>

> Kai

> http://www.eic.nu

>

>



************************************************************************

You are currently subscribed to the Radsafe mailing list. To unsubscribe,

send an e-mail to Majordomo@list.vanderbilt.edu  Put the text "unsubscribe

radsafe" (no quote marks) in the body of the e-mail, with no subject line.

You can view the Radsafe archives at http://www.vanderbilt.edu/radsafe/