700
quality control but a bad standard because the standard deviations of
the two middle groups are not significantly different
statistically.
We
have not had the opportunity to investigate this type of possible cause
and by now,
since the data are 2 years old,
tions used on these data may not be
the standards and calibra-
recoverable.
Had the possibility of
600
this situation been suspected in advance of the actual cross-check test,
sufficient information could have been collected to investigate this and
other possible causes.
It is obvious that if any reasonable quality control standard were
applied to the total data set of this example, represented by the broken
Estimated Population of
Sampled Data
Figure 3,
400
300
DATA VALUES
500
line curve
in Figure
3,
the result would be an unreasonable set of
acceptance limits as wall as a loss of information contained by the
data. Perhaps our current uneasiness with acceptance rules is due to
our intuitive recognition of this unreasonableness. We err in placing
the blame on the acceptance rule rather than on the type of data we work
with.
CONCLUSION
This is just one of about 50 quality control data sets involving 10 different radioisotopes that have so far been examined in ovr laboratory,
all but one showed a statistically significant (maximum likelihood ratio
test) mixture of distributions.
Similar statistical work has also found
such mixtures in thermoluminescent dosimeter data, plutonium in soil data,
and air pollutant data. The plutonium in soil mixture-of-distributions
has been identified as caused by two distinctive sizes of plutonium
particles by using autoradiographs of soil samples; about 85 percent of
the plutonium in the soils found at the Nevada Test Site is in the form
of an extremely fine dust and 15 percent is particles large enough to be
counted under a dissecting microscope.
This has been called the "hot
particle problem.” Mixtures of distributions are a well known and
accepted phenomena in medical statistics where the components distributions are typically identified as sick and well patients, male and
female,
and so on.
Our quality control data differ from the hot particle problem and the
medical examples cited above in the reasons for the structure in the
data. The structures in the hot particle situation and in the medical
examples are due to a true structure in the media sampled.
In our
quality control data,
aliquots of a single homogeneous
laboratory-
prepared medium are sent to the participants, and the structure is
introduced during the sample analysis procedure.
There is nothing in
the present statistical analysis to @istinguish the various causes of
a heterogeneous data structure.
whatever the cause.
The
599
HOH 40 JAUND IWWHON 40 31¥NIGHO
598
statistical analysis is the same
The cause can only be defined by follow-up
investigations.