Ecological studies and population shifts

[David Scherer <scherer@uiuc.edu>]

Wade Patterson wrote, in part:
Dr. Kaurin and others:
I'd like an explanation of why the Japanese survivor data is not an
"ecological" study. No Japanese wore a dosimeter. Doses are assigned solely
on the basis of "entire" radial increments. How are radial increments
different than counties? By the way, the errors in doses so assigned are
30%-50% due to potential inverse quare law differences. 
<end quote>

With the A-bomb survivors, we are at least sure that all the subjects were
in fact exposed to some radiation, approximated by their distance from
ground zero.  In the radon study there is no individual verification that
the subjects were actually exposed to the radon levels suggested by the
county in which they died.

It could be that people from high-Rn counties moved to low-Rn counties
after their cancers had been initiated but before they were diagnosed.  It
is also possible that people spent most of their life in low-Rn counties
and moved to high-Rn counties later in life, thus avoiding the Rn exposures
assumed by using the county average.  I know of no way to numerically model
this effect.  Perhaps one could look at bulk changes in population or age
distributions over time, but they are only suggestive.

Perhaps because of the large database involved in Dr. Cohen's study this
effect is not significant.  For example, one could presume thia happened
with some fraction of the population (N percent) and see if this changes
the inverse dose-reponse relationship.  If you need a 25 percent population
shift to overcome the result, it seems pretty unlikely that this is
important.  On the otherhand, if a 2 percent population shift is sufficient
to alter the result, then ....

Regards,
Dave Scherer
scherer@uiuc.edu