side). There were no samples with Ag concentra-
their analyses. Nonetheless, we believe that this
tion above reporting limits for both QA and QC
range is reasonable for metals in soils and we
laboratories.
recommend that it be used. Rejection of 10% of
We also determined whether the outliers were
contract laboratory results would likely trigger
for several metals in a few samples and whether
demands to provide proof that the QA concentra-
they occurred with certain laboratories. However,
tions are sufficiently reliable to be used as referee
the outliers were distributed among 25 different
values. We believe that this position is justified
samples. Two samples had outliers for three met-
regardless of what regulatory limits are chosen.
als and three samples had two metals outside of
A further very practical question remains:
the limits. None of the latter five samples was
would an entire set of results for eight metals be
from the same project. We lacked necessary infor-
rejected just because the QC/QA ratio for one
mation to state with certainty that different labo-
metal was outside tolerance limits? We doubt it,
ratories were associated with each of the five
especially if the results for the offending metal
samples. In any case, we feel that outliers are oc-
were low compared to concentrations of concern.
curring more or less randomly rather than as any
We suspect that rejection would occur only when
systematic problem. Sample handling and/or
two or more ratios are outliers for a sample and
analysis errors are likely causes.
only when concentrations are high. For the
Geometric means for all four metals were be-
archived data, only five samples out of 124 (4.0%)
low the expected mean ratio of 1.00, but the 95%
had more than one ratio per sample outside of
tolerance bands. Consequently, we believe ratios
The tendency of all the ratios to be below 1.00
should be flagged if they are outside 0.402.50
indicates that QC laboratories are reporting
and examined by a QA manager to determine the
slightly lower concentrations than QA laborato-
fate of the results. Better yet, an artificial intelli-
ries, but this pattern is neither marked nor clearly
gence computer program could probably be de-
established. The 99% tolerance intervals for Ba,
vised to do this task. With so few samples having
Cr, and Pb were very similar. The somewhat wider
multiple outliers, it seems that the 0.402.50 range
limits for As is a reflection of its large standard
for ratios is justified.
deviation. Only five additional values beyond the
Although most results for Ag, Cd, Hg, and Se
original outliers are outside of these tolerance in-
were less than the reporting limit (ND<), some
tervals.
insight can be gained by tabulating the reporting
It is instructive to consider the effects of im-
limits (Table B3). A few values were unusually
posing three possible sets of logarithmically dis-
high (27 of 722, or 2.4%) and were deleted with-
tributed rejection limits. In Table B2 we see that
out further consideration. Of the remaining val-
the arbitrarily chosen limits of 0.303.00 produced
ues, eleven of the 17 unacceptable reporting lim-
6.3% exclusions with nearly equal numbers of
its were associated with only three samples, none
outliers below 0.30 and above 3.00 and reason-
of which were samples with excluded As, Ba, Cr,
ably similar numbers for each metal. If the limits
and Pb ratios. The concentrations of other metals
are relaxed to 0.254.00, only 4.4% of the ratios
in these three samples were not unusual. While
are excluded. Two-thirds of the exclusions are on
matrix effects may be involved, we continue to
the low side, i.e., cases where the QC con-
believe that an occasional faulty analysis is the
centration is much lower than the QA concentra-
major culprit. After these exclusions, means and
tion. In the other direction, if we use limits of
medians were in reasonable agreement, indicat-
0.402.50, 10.2% of the ratios would be rejected.
ing that the distributions were not seriously
Once again two-thirds are on the low side, but
skewed. Furthermore, the ranges and the means
the pattern is quite different for As than it is for
for QA and QC laboratories were also in good
the metals determined largely by plasma emis-
agreement. Even under these conditions, the ra-
sion. The tendency for low QC results is absent
tios of QC/QA reporting limits were often out-
for As but quite pronounced for Ba, Cr, and Pb.
side 0.303.00 for specific samples. Whether this
The most recent results (again, 32 project loca-
is a problem depends on the relationship between
tions) contained 11.9% of the ratios outside 0.40
reporting limits and regulatory concentration val-
2.50 with similar percentages of low and high
ues. When reporting limits are well below regula-
outliers.
tory limits, there is less need for the reporting
Clearly the 0.402.50 limits would put pres-
limits to be in close agreement.
sure on laboratories to improve the accuracy of
When one laboratory reported ND< and the
7