RE: If you do Science, use the Scientific Method!

["Fritz A. Seiler" <faseiler@NMIA.COM>]

Dear Peter,



I replied to your mail directly, not noting that you sent it to both lists.

Here is part of what I

wrote to you directly and then some:



I think that you SHOULD be doing 'that'! It is by polite scientific

discussion that we learn how

to do things.  I have been in the radon discussion for more than twenty

years, as I was a staff

member at the ITRI (Inhalation Toxicology Research Institute) in

Albuquerque, New Mexico

from 1980-1990.  You have the correct idea as to what the two sides are

claiming.  No problem

with that.  As I said in my rant, I am just a plain old risk assessor

(Nuclear physicist by trade),

so I want to use information to estimate the risks of radon exposure.

    I think that your statement, that Bernie's data cannot be used to

estimate the personal risk

of an individual, is not quite correct.  Using Bernie's data, we can make

some conditional risk

statements.  Joe Alvarez and I have shown that, IF you treat the individual

as the average

member of the general American population, you can estimate its risk and its

error, and IF

you treat that person as an individual in that group, you get the same

numerical value but a

much larger error.  What you cannot do is to use the general population risk

and its error for

one individual.

    I take it that you are from Australia, so what does ARPANSA mean? And

what is your

snail mail address? If you are interested, I can send you some of our

reprints that explain

these problems in detail.



P.S.: If you fit a model to some data, let's assume a straight line model,

you do a simple

linear regression and obtain a slope and a y-axis intercept and their

errors.  A look into

any book on statistics will show you that you get two kinds of errors for

the slope:  One

is the error for the average slope and the other is the standard deviation

of the average

slope.  The first is the error of the average risk for the population (slope

= risk coefficient,

in the linear model),  and  the second is simply the distribution estimated

for an assumed

(n+1)st data point (that would yet have to be determined, but can be

estimated from the n

data points already available.  That value is the risk for an individual in

that group.  Those

are the conditional risk values that can be determined for a model fit

(linear or nonlinear!)

to the n data points.  Again, if any one needs exact references or some

reprints (If I still

have some!), contact me off list.



Again, havea nice weekend



Fritz



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

Fritz A. Seiler, Ph.D.

Sigma Five Consulting:       Private:

P.O. Box 1709                   P.O. Box 437

Los Lunas, NM 87031         Tome', NM 87060

Tel.:      505-866-5193         Tel. 505-866-6976

Fax:      505-866-5197         USA

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



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

"This is the hour when democracy must justify

itself by capacity  for effective decision, or risk

destruction or desintegration. Europe is dotted

with the ruins of right decisions taken too late."



"America's Responsibility in the Current Crisis"

Manifesto of the Christian Realists. May, 1940.

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