[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Concrete Ecologic Example Myth
I would be happy to send any Radsafers an email version of a paper by another
author that supports my points below and provides further details.
Regards, Bill Field
>
> On Sat, 21 Jun 2003 epirad@mchsi.com wrote:
>
> > In Cohen's ecologic study, this is what I see:
>
> --I have responded to all of these point in the last few days. If
> anyone wants repeats of one of these responses, please specify
> >
> > 1.) An inverse relationship between estimated county smoking information and
> > estimated county radon concentrations. Maybe the reason for this should be
> > explored first.
> >
> > 2.) Estimated county smoking rates explaining little of the variation in
> > county lung cancer mortality.
> >
> > 3.) Uses smoking data that can not adjust for duration of smoking or intensity
> > of smoking.
> >
> > 4.) Estimated lung cancer incidence rates for the time period of interest
> > being poorly estimated by lung cancer mortality data from a different time
> > period.
> >
> > 5.) An inverse association between estimated county radon concentrations and
> > other smoking related cancers not related to radon, which to me and most
> > others points to confounding by smoking.
> >
> > 6.) The negative association goes away in Iowa (a state with 1% of the U.S.
> > population but 1/16 of the U.S. ecologic units) if I merely use better SEER
> > lung cancer incidence data. Dr. Cohen's only answer for this is that there
> > must be some problems associated with ethnic minorities in Iowa.
> >
> > etc.
> >
> > Given these facts, I am not motivated to perform studies analogous to little
> > green men.
> >
> > Perhaps Cohen could set up 1600 ecologic units starting with an inverse
> > association that he finds between smoking and radon, counties unevenly
> > weighted by population and geographic area and randomly sample (yes, he would
> > have to get better data by surveys of counties)the joint distribution of
> > smoking and related socioeconomic variables in a Monte Carlo analyses?
> >
> > Email me directly and I would be happy to point you to existing papers that
> > document points 1-4 below. If you would read the papers that document my
> > statements below, I think you would understand the limits of ecologic analyses
> > and the fact it has little to do with dose response.
> >
> > Bill Field
> > > Bill,
> > >
> > > I think you are confusing trying to establish a dose response relationship
> > > from ecological data with testing a particular theory against an ecological
> > > observation. These issues are not limited to epidemiology. It tends to be
> > > difficult (impossible?) to formulate a theory of how small things interact
> > > by just looking at the end result of the interaction, but any valid theory
> > > must still be compatible with the observation.
> > >
> > > If I look at Mars and it looks red, that is not proof that Mars is populated
> > > by a bunch of compulsive Martians who go around painting everything red.
> > > However, if you have a theory that states that Mars is populated by a bunch
> > > of compulsive Martians who go around painting everything green, then my red
> > > observation is a problem for your theory. That doesn't mean your theory is
> > > wrong, but it would sure help if you could come up with an explanation that
> > > reconciles your theory with my observation. (e.g. Martians have wives that
> >
> > > go around covering the green paint up with red blankets.)
> > >
> > > To me, no one has come up with an explanation that makes LNT compatible with
> > > the ecological observation. Until that happens, I can't accept LNT as a
> > > valid scientific theory. (Using LNT as a basis of regulation is a different
> > > story.)
> > >
> > > You state that many examples have been given on how Dr. Cohen's data can be
> > > biased. I don't recall any that seem reasonable to me. All I need is one
> > > NUMERICAL example. Put 1600 hypothetical counties on a spreadsheet. Put in
> > > some confounders (Martian wives) and reproduce the ecological observation.
> > > (You correctly state that it may be impossible to IDENTIFY empirical sources
> > > of ecologic bias from aggregate data alone, but that should not stop you
> > > from POSTULATING them.)
> > >
> > > Kai
> > >
> > > ----- Original Message -----
> > > From: <epirad@mchsi.com>
> > > To: "Kai Kaletsch" <eic@shaw.ca>
> > > Cc: "BERNARD L COHEN" <blc+@PITT.EDU>; <radsafe@list.vanderbilt.edu>
> > > Sent: Friday, June 20, 2003 12:54 PM
> >
> > > Subject: Concrete Ecologic Example Myth
> > >
> > >
> > > > Kai,
> > > >
> > > > I know we have gone through discussions about the limitations of ecologic
> > > > studies many times but there are some points that a few radsafers continue
> > > to
> > > > miss.
> > > >
> > > > You suggested the other day putting an upper limit on the error from an
> > > > ecologic study. Unfortuantely, this is not easy to do since the error in
> > > > ecologic studies is unbounded.
> > > >
> > > > Others problems to keep in mind -
> > > >
> > > > 1) It is not possible to identify empirical sources of ecologic bias from
> > > > aggregate data alone. Researchers must rely on prior knowledge of
> > > intergroup
> > > > variation in the distribution of other risk factors and effect modifiers.
> > > > We don't know this inter group variation for Dr. Cohen's data nor has he
> > > > adjusted for it.
> > > >
> > > > 2) Factors responsible for ecologic bias may not be confounders or effect
> > > > modifiers at the individual level and that identifying the bias is even
> > > more
> > > > difficult because factors may not even "appear" to be confounders or
> >
> > > modifiers
> > > > at the ecologic level.
> > > >
> > > > 3) Ecologic biases can even reverse the direction of an observed
> > > association,
> > > > especially when the range of average exposure levels across groups is
> > > small or
> > > > the exposure under study is not a strong risk factor regardless even if
> > > you are
> > > > using an ecologic study to test the LNT.
> > > >
> > > > 4) If cross level bias is occurring (which we know it is), ecologic data
> > > can
> > > > not be successfully used to adjust for ecologic bias.
> > > >
> > > > Many examples have been given on how Dr. Cohen's data can be biased and
> > > since
> > > > Cohen will use his own faulty data to test the "concrete" example;
> > > providing
> > > > a "concrete" example is doing no more than providing an explanation Cohen
> > > will
> > > > use his own faulty data to negate. So, the circle will never end.
> > > > > Friends,
> > > >
> > > > > We are all aware that ecological studies have some limitations - these
> > > issues have been discussed (some would say ad
> > > > > infinitum) on this board and elsewhere.
> > > >
> >
> > > > ************************************************************************
> > > > 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/
> > > >
> > >
> >
> >
>
************************************************************************
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/