Re: summary versus individual data

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

On Fri, 23 Feb 2001, Harry Hinks wrote:

> 
> Prof. Cohen wrote -
> 
> 	--Of course you have to collect enough data to get a true average, and 3 
> houses is not enough. But aside from this, in your example, if LNT is 
> correct, the only thing important for determining the number of cancers is 
> the 9 pCi/L average. No further useful information is obtained by using 
> measurements in particular houses.
> 
> 
> ---------But the important point is that my test has been done, and LNT
> -fails, unless someone can come up with an alternative explanation. Failing 
> one experimental test is enough to validate a theory. Of course other tests 
> are always welcome.
> 
> 
> 
> ---------------
> That is my point - how do you know the 9 pCuries/liter is correct?
> 
> How many houses are enough?  How do you determine how many houses in a 
> specific area need to be sampled to provide a representative sampling?  

	--That is a subject of considerable research involving completely
independent surveys by different groups. These matters are described in my
papers


> Wouldn't it depend on the variance of radon concentrations in a specific 
> area?  This is not like the atomic bomb data where there is decreasing 
> exposure with distance that is somewhat predictable.  I just looked at one 
> of your papers, it looks like you only have a few measurements representing 
> an area with a population of 30,000 people or more in many cases.  Do you 
> truly feel your limited number of measurements give you a true concentration 
> (exposure) average?  In consideration of what Ruth said, how can these 
> limited number of measurements for a given area be considered dose 
> estimates?  How can you apply collective dose assumptions to concentration 
> data generated from a few radon measurements obtained for a particular area. 
>   I don't think the point is how many total measurements you have for the 
> states, but rather how representative are the limited meausurements in a 
> certain area.

	--These are matters covered in my papers, but basically the answer
is the same as given above

> 
> If you do not think individual level measurements are needed, how well can 
> you predict what the radon concentrations are in Radsafe subscribers houses? 
>   We can actually test this.  If 20 radsafers (more if you like) can send me 
> their radon concentrations (should we say living room) and zip code, I can 
> collect the data and put it in a table.  I will send you the zip codes and 
> we can look at the agreement between the actual measurements and your 
> predictions.  Are you willing to do this?
> 

	--If I can get enough data in that way, I would be happy to
analyze it. But my papers contain much more elaborate tests

> 
> Also, are you saying that because no one can explain what you find, the LNT 
> fails?  I asked a professor I have in class about this and they told me that 
> you could not test for cross-level bias in summary data studies.  So how do 
> you know your study does not suffer from cross-level bias? This goes back to 
> the ability to correct for confounding information with summary data.  My 
> possible explanation for your findings is that they suffer from cross-level 
> bias.  Can you show us that your studies do not suffer from cross-level 
> bias?

	--The issue of cross level bias is discussed in my paper in Health
Physics 76:437-439; 1999.

	I am leaving on a trip and will not be able to correspond further
on this for a week. But if messages are received, I will respond when I
return.

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