To perform a proper errors-in-variables regres-
erence should be very good. If, however, the con-
sion requires consideration of the measurement
centration in most samples is 5 to 100 mg/kg , the
errors in both variables. The appropriate methods
two methods will be much more likely to disagree.
are outlined in Mandel (1984). These methods re-
This must be kept in mind when interpreting re-
quire estimating the ratio of the random error vari-
sults, especially when comparing across different
ance for the on-site method to that of the refer-
studies that may have collected samples at con-
ence analytical method. With split sample data,
siderably different analyte levels.
suitable estimates of these ratios may generally be
obtained by using variance estimates from Grubbs'
test or the related tests mentioned above.
SUMMARY OF ON-SITE ANALYTICAL
If the variance ratio is not constant over the
METHODS FOR EXPLOSIVES IN SOIL
range under study, more complicated models than
those analyzed in Mandel (1984) must be em-
There is considerable interest in field methods
ployed. Alternatively, transformations of the data
for rapidly and economically determining the
might stabilize the variance ratio. Note that it is
presence and concentration of secondary explo-
the variance ratio, not the individual variances,
sives in soil. Such procedures allow much greater
that must remain constant. The ratio of variances
flexibility in mapping the extent of contamination,
for two methods with nonconstant absolute vari-
redesigning a sampling plan based on near-real-
ances but constant relative variances will still have
time data, accruing more detailed characterization
a constant variance ratio.
for a fixed cost, and guiding continuous remedial
Two other caveats about the use of regression
efforts. Ideally, screening methods provide high-
techniques also are appropriate. First, standard
quality data on a near-real-time basis at low cost
regression methods produce bias regression pa-
and of sufficient quality to meet all intended uses,
rameters estimation and may produce misleading
including risk assessments and final site clear-
uncertainty intervals. Similarly, the interpretation
ances, without the need for more rigorous proce-
of R-squared values also is affected. Second, per-
dures. Although the currently available screening
forming regressions on data sets in which samples
procedures may not be ideal (not capable of pro-
with concentrations below the detection limit (for
viding compound-specific concentrations of mul-
one or both methods) have been eliminated may
tiple compounds simultaneously), they have
also result in biased regression estimates, no mat-
proved to be very valuable during the character-
ter which regression analysis method is used.
ization and remediation of numerous sites. Cur-
rently, available field methods that have been
evaluated against standard analytical methods
Comparison to regulatory thresholds,
and demonstrated in the field include colorimet-
action limits, etc.
When the purpose of sampling is to make a
ric and immunoassay methods (Table 4). Each
decision based on comparison of results to a spe-
method has relative advantages and disadvan-
cific value such as an action level for cleanup, on-
tages, so that one method may not be optimal for
site and reference analytical method results may
all applications. To assist in the selection of one or
be compared simply on the basis of how well the
more screening methods for various users' needs,
two methods agree regarding the decision. The
Table 3 (modified and expanded from EPA 1997)
appropriate statistical tests are based on the bino-
provides information on on-site test kits for de-
mial distribution and include tests of equality of
tecting explosives in soil. Selection criteria are dis-
proportions and chi-square tests comparing the
cussed in the following sections.
sensitivity and specificity (or false positive and
The two types of currently available on-site
false negative rates) of the on-site method relative
methods, colorimetric and immunoassay, are fun-
to the reference analytical method. Note that any
damentally quite different. Both methods start
measure of consistency between the two methods
with extracting a 2- to 20-g soil sample with 6.5-
is affected by how close the true values in the
to 100-mL acetone or methanol for a period of 1 to
samples are to the action level. The closer the true
3 minutes, followed by settling and possibly fil-
values are to the action level, the less the two meth-
tration. The basic procedure in the CRREL and
ods will agree, even if they are of equal accuracy.
EnSys RISc colorimetric methods for TNT is to add
For example, if the action level is 30 mg/kg and
a strong base to the acetone extract, producing the
most samples have levels of above 1000 mg/kg,
red-colored Janowsky anion. Absorbance is then
the agreement between the on-site method and ref-
measured at 540 nanometers (nm) using a spec-
14
Previous Page
