Re: Cohen's ecologic study questions

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

On Tue, 22 Jan 2002, Rad health wrote:



> Drs. Field and Smith have repeatedly "specifically suggested" in their Forum

> paper and follow-up papers in Health Physics that one of the major factors,

> among others, most likely causing your paradoxical findings is cross-level

> bias or inter county variability.



	--That is not what I mean by "specific" and not in accordance with

definitions I have given. By "specific"  I mean something that

can be calculated if data are available. My requirement is that it be

specific enough to be calculeted; I do not require that data be available.

My approach, which I call "plausibility of correlation", is to calculate

the correlations with radon or with lung cancer that are necessary to

explain my results, and then evaluate the plausibility of that

correlation. This is all explained in my paper on "Treatment of

confounding factors in an ecological study" which is posted on my web

site. Please read that paper and base your comments on it.



> ---------------------------------------

> Dr. Cohen, I have specific questions concerning Field's, Smith's, and

> Lynch's paper for you.

>

> I think one of their focuses has been on the quality of your data.  They

> previously suggested that you could improve your data by using lung cancer

> INCIDENCE data from the SEER states.



	--I will look into this when time permits, but I have other more

urgent things to do. You could accelerate the process if you would write a

letter to the editor raising the issue; that would give me an opportunity

to get the analysis published, which would greatly increase my priority on

this. You are asking me to spend a lot of time on a project just to

respond to your inquiry on RADSAFE. I cannot publish this as a response to

Field et al because the Editor decreed that they would have that last

word.  I personally do not believe that using SEER data will make a major

change, and it is not obvious to me that SEER data are more relevant than

mortality data; that is why I have not personally given that project a

very high priority.



> 1) Smith and Field showed that by using SEER data in Iowa your large inverse

> association disappeared. Have you ever explained in specific scientific

> terms why the association went away when SEER data was used?



	--I did not report a large inverse correlation for Iowa; it

differed from zero association by 1.1 SD (standard deviation) for males

and by 0.3 SD for females (see Table 4 of item #2 on my web site). If

there is an important difference in using SEER data and mortality data for

Iowa, that is not something for me to explain; I said what I did and what

the results were.

	The results for one particular state are not very significant. My

data gave positive correlations between lung cancer and radon for 10

states for males and 11 states for females. The distribution of regression

coefficients for the various states was centered at -5.0 with a width of

8.0. It is very easy to see why there can be such variations due to

confounding factors. We have been thru this discussion before.



> 2) They continue to say that cross-level bias and inter county variability

> caused your findings.  Do you really have the data to show this is not the

> case?



	--I need a specific example -- see above. If a specific example is

offered, it is up to me to supply the data required.



> My most important question about the Field and Smith paper:  Are adjustments

> made using aggregate data (smoking, education, etc.)usually used to adjust

> for confounding at the individual level - sufficient for use in adjusting at

> the county level?



	--I do not "adjust" my data for anything. I treat confounding by

my method of stratification which is very much superior.



>  From reading their papers they say you may never find

> your answer because your data are lacking.  Can you really say you have

> tested all the possibilities if you do not have the data you need to adjust

> for aggregate confounding?



	--Providing the data is my problem. What I need is a

specific suggestion for which data are needed.











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

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/