Environmental Engineering Reference
In-Depth Information
relation between each procedure and COMP as well as the average value of tested
protocol compared to the average COMP value in the distribution area.
“
Repro-
ducibility
cient of variation
or relative range of the three samples taken in one property. The ability of the
sampling protocols to detect
”
of the sampling protocols was determined by the coef
was assessed by analyzing the
proportion of properties where the attached lead measuring device (COMP)
exceeded the MCL of 10
“
problem properties
”
were also analyzed, i.e. the pro-
portion of a sampling protocol that were greater than 10
µ
g/L;
“
false positives
”
µ
g/L when the measured
lead was in fact less than 10
µ
g/L.
“
Consumer acceptance
”
was assessed by the
consumer
s willingness to cooperate in undertaking a sampling protocol in their
home while
'
several aspects of the procedure (e.g. is
the procedure easily applicable, are skilled samplers needed, does the procedure
need speci
“
practicality
”
was assessed by
“
c tools, etc...)
”
(EU Report, p. 56). Both
“
consumer acceptance
”
and
“
was assessed using
a hypothetical wage rate, and an estimated time in each sampling protocol was used
to determine the total cost for each protocol.
In the EU Report
practicality
”'
have no quanti
able evaluation schemes.
“
Cost
”
s performance evaluation of all the sampling protocols, the
30MS and the RDT methods met the representativeness criteria while the FF
sample did not. One of the ways in which representativeness was assessed was by
examining the linear relationship between lead from a sampling protocol and lead
from the sample obtained from the lead measuring device (called COMP). Both
slope and R-squared statistic were used to make judgments. From the EU Report
(see p. 35), the RDT sample overestimated the measuring device sample (slope 1.27
and r-squared of 0.61) while the 30 min
'
first liter sample and the second sample
both underestimated the COMP sample (slope of 0.80 and r-squared of 0.50 and
0.56 respectively). The average of the 2 L from the 30MSA, however, has a similar
outcome for slope of 0.80 but a slightly improved r-squared of 0.58. FF strongly
underestimated the measure of lead from the measuring device (COMP) with a
slope of 0.57 with an r-squared of 0.29. If one were to choose a sampling protocol
based on model
fit, then the RDT protocol should have been chosen. If one were to
choose a sampling method based on accuracy then the 30MS should be chosen but
caution should be applied. Since the slope of the linear relationship between
measured lead (COMP) and 30MS sample is 0.80, the 30MS sampling protocol
consistently underestimates lead in drinking water. Therefore, on average, the
30MS sampling protocol underestimates the true value of lead by 20 percent. This
has severe implications for determining the percentage or number of households
with lead above the MCL of 10
g/L, as the number of properties that actually have
lead in drinking water is much more than expected under this protocol. One way to
overcome this problem is to lower the MCL of 10
µ
g/
L). The RDT sampling protocol has the opposite problem. On average it overes-
timates the true value of lead by 27 percent.
The EU Report also did not consider the average of the RDT and the FF
sampling protocol to be any better than the 30MS sample or RDT sampling pro-
tocols, stating that it
µ
g/L by 20 percent (i.e. to 8
µ
(EU
Report, p. 38). However, the average of the RDT and the FF sample does in fact
“
does not improve relation or give additional information
”
Search WWH ::
Custom Search