other gave a concentration above the reporting
tistical evaluation. Suffice it to say that such re-
limit, the ratios were often outside the factor of
sults occur with disturbing frequency.
3.00 criterion. By far the poorest metal in this re-
Another common extreme is represented by
spect was Cd, where 25 of 40 such ratios failed
several analytes in VS233 where the QA labora-
tory reported < 1.1 g/kg and the QC laboratory
this test. The failure rate for Ag was 6 in 16 pairs,
reported < 7000 g/kg. Such examples suggest
for Hg it was 4 in 27, and only 2 of 21 Se pairs
failed to be within a factor of 3.00. In general,
the need for standardization of reporting proce-
most values above reporting limits were from the
dures for samples requiring dilution to permit de-
QA laboratories, although several were flagged
termination of one or more high concentration
with "J," meaning they were estimates but below
analytes. If such guidelines are already in place
reporting limits. The QA laboratories may be pro-
but are misunderstood or ignored by laborato-
viding a slightly more reliable analysis than the
ries, a training program might improve the situa-
QC laboratories, but there still is insufficient real
tion. Alternatively, if low concentration analytes
evidence to defend this statement.
really are of no interest when one or more analytes
are present at very high concentration, data com-
pilations could be greatly simplified.
VOCs in soils
For the 69 sets of VOCs in soils, there was a
For the 134 QC/QA ratios with concentrations
total of 134 analyte pairs having both a QC and
above reporting limits, ratios ranged from 0.015
QA concentration above reporting limits. Ethyl-
to 2830! The distribution of these ratios by analyte
benzene with 25 pairs, toluene with 35 pairs, and
and for the total is shown in Table B4. We note
total xylenes with 36 pairs represented the bulk
that the distributions are very similar for indi-
of the data. The remaining 38 pairs were distrib-
vidual analytes and for the group collectively. This
uted among benzene (eight pairs), chloroform (one
suggests that statistical analysis applied to the
pair), total 1,2-dichloroethenes (five pairs), me-
entire group would adequately represent results
thyl ethyl ketone (two pairs), tetrachloroethene
for VOCs in soils. It is also apparent that extreme
(11 pairs), and trichloroethene (11 pairs). Carbon
values are excessive. Over 40% of the ratios are
tetrachloride, chlorobenzene, 1,2dichloroethane,
outside the limits 0.254.00 and 24% are outside
1,1-dichloroethene, and vinyl chloride yielded no
limits of 0.1010.0. While we know of no stan-
pairs. Although samples VS260 through VS266
dard for acceptability of analytical results on split
used GC methods rather than GC/MS, there was
or co-located samples sent to different laborato-
no apparent reason to treat those results sepa-
ries, it stretches the boundaries of credibility to
rately.
think that analysts would describe agreement
Most analytes also had examples where either
within a factor of 10 as acceptable quantitation.
the QC or the QA result was a real number but
Nonetheless, we will base statistical characteriza-
the other was below reporting limits. In many
tion on the 76% of ratios within 0.1010.0.
cases it was impossible to judge how well (or
A histogram of the logs of all 134 ratios is
badly) these pairs compared because reporting
shown in Figure 4. Logs of the 102 ratios between
limits were based on dilutions made to accom-
0.10 and 10.0 form a tolerable approximation to a
modate a high concentration analyte. For example,
normal distribution. Further support for the hy-
benzene in sample VS217 was reported as 59.7
pothesis of a lognormal distribution is provided
g/kg by the QA laboratory (reporting limit of
by the linear cumulative probability plot of these
6.1 g/kg) while the QC laboratory reported <
logs on normal probability paper (Fig. 5). When
2500 g/kg. These results may agree, but it is
separate lognormal plots were prepared for
impossible to tell. Possibly this issue is unimpor-
ethylbenzene, toluene, total xylenes, and the re-
tant given the very high concentration of xylene
maining analytes collectively (Figs. 6a6d), all pro-
in this sample. One thing is clear: reporting these
vided reasonable fits to this model.
concentrations to three significant figures is a gross
Based on the lognormal model, the geometric
misrepresentation! In other cases, such as methyl
mean of the 102 ratios between 0.10 and 10.0 was
ethyl ketone in sample VS218, the QA labora-
0.95 and the 95% confidence limits on this mean
tory reported 271 g/kg and the QC laboratory
were 0.77 to 1.16. The 99% tolerance intervals for
reported <12 g/kg. Although the disagreement
individual ratios were 0.062 to 14.4. The fact that
of these results could be represented by a ratio if
these intervals are considerably wider than the
we assigned the QC result a value of 12 g/kg,
boundaries used for editing suggests that some
we decided not to include such results in the sta-
values outside the range of 0.10 to 10.0 may also
8
Previous Page
