Dr. Don Hendricks
-4-
August 28, 1972
We have arbitrarily selected C=.7 as the value to use here.
Other evidence
suggests that this will not be a bad guide for other radioisotopes, and that
the results should apply reasonably well to those islands contaminated mainly
by fallout.
If the sampling is done along a very distinct gradient in
concentrations, then the above guides don't hold, as for example as shown in
the Rocky Flats data.
If one has some evidence on whict to sketch in
several levels of contamination (as was available at the GMX site on NTS)
then the variability within such a subdivision should again be roughly as
indicated.
The consequences for the Eniwetok study seem to be that we can
feel reasonably secure in predicting confidence limits (as %) for situations
other than on those islands where ground zeros existed, where rather larger
samples are indicated for comparable confidence limits.
The above calculations can be summarized in the following table:
n (sample size)
Approximate confidence limits
as % of mean concentration
10
+ 44%
20
+ 31%
30
+ 26%
. 40
50
+ 22%
“
+ 20%
60
+ 18%
100
+ 14%
200
+ 10%
These results are simply calculated as 1.4/59 so one can easily find values
for other sampleé sizes, or can increase the numerator (which is 2(.7) or
twice the coefficient of variation) to some larger value to reflect supposedly
greater variation.
From this evidence, we don't recommend that sample sizes
as smal? as 10 be regarded as anything but rough guides to relative concentrations.
Samples of 30 or more may be sufficient to make some reasonably
satisfactory. estimates.
However, if one wants to be fairly precise, even
100 samples is not too many, since we still have a fdarly wide range (+14%)
for the true mean.
Perhaps it is worth repeating here that these confidence