United States
Environmental Protection
Agency
Environmental Monitoring
Systems Laboratory
Las Vegas, NV 89193-3478
Research and Development
EPA/600/S4-90/013 Sept. 1990
4>EPA Project Summary
A Rationale for the
Assessment of Errors in the
Sampling of Soils
J. Jeffrey van Ee, Louis J. Blume, and Thomas H. Starks
The sampling of soils in RCRA and
Superfund monitoring programs
requires associated quality
assurance programs. One objective
of any quality assurance program is
to assess and document the quality
of the study data to ensure that it
satisfies the needs of the users. The
purpose of this report is to describe
the nature and function of certain
quality assurance samples in the
assessment and documentation of
bias and precision in sampling
studies of inorganic pollutant
concentrations in soils. A foundation
is provided for answering two basic
questions:
How many, and what type of,
quality assurance (or, to be more
specific, quality assessment)
samples are required to assess
the quality of data in a field
sampling effort?
How can the information from the
quality assessment samples be
used to identify and control the
principal sources of error and
uncertainty in the measurement
process?
This document has been developed
to provide people who plan,
implement, or oversee RCRA or
Superfund soil sampling studies with
information on quality assessment
samples so that they will have a
better basis for decisions concerning
the employment of such samples in
their quality assurance programs.
This Project Summary was
developed by EPA's Environmental
Monitoring Systems Laboratory, Las
Vegas, NV, to announce key findings
of the research project that is fully
documented in a separate report of
the same title (see Project Report
ordering information at back).
Introduction
The four principal contributions of this
document are as follows:
(1) a list of names and descriptions of
quality assessment samples;
(2) a rationale for determining the
number of quality assessment
samples to employ in a study;
(3) a description of the function of
various types of quality assurance
samples in determining estimates
of components of measurement
error variance; and
(4) a basis for the development of a
computer program for
computation of components of
measurement error variance.
The list of names and descriptions of
types of quality assessment samples in
Table 1 is considered as a major
contribution in that there is great need for
standardization of nomenclature and
terminology in soil-sampling quality
assurance. The named sample types in
Table 1 are classified as double blind,
single blind, or non blind. A double-blind
sample is one that the laboratory chemist
will not recognize as a quality assurance
(QA) sample. A single-blind sample is
one that the laboratory chemist will
recognize as a QA sample but will not
know the pollutant concentration in the
sample. A non-blind sample is one that
the laboratory chemist recognizes as a
QA sample and one for which the
chemist knows the reference value for the
pollutant concentration. Principal
emphasis in this document is given to the
double-blind and single-blind types of
quality assessment samples.
-------
Table 1. Type of Quality Assessment Samples or Procedures
Double-Blind Samples
1. Field Evaluation Samples (FES)
These samples are of known concentration, subjected to the same manipulations as routine samples and introduced in
the field at the earliest stage possible. They can be used to detect measurement bias and to estimate precision.
2, Low Level Field Evaluation Samples (LLFES)
These samples are essentially the same as field evaluation samples, but they have very low or non-existent
concentrations of the contaminant. They are used for determination of contamination in the sample collection, transport,
and analysis processes. They can also be used for determination of the system detection limit.
3. External Laboratory Evaluation Samples (ELES)
This sample is similar to the field evaluation sample except it is sent directly to the analytical laboratory without
undergoing any field manipulations. It can be used to determine laboratory bias and precision if used in duplicate. We
recommend using the same sample as the FES to allow isolation of the potential sources of error. Spiked soil samples
have been used as external laboratory evaluation samples in past studies for dioxin, pesticides, and organics, and
natural evaluation samples have been used for metals analysis in soil and liquid samples.
4. Low Level External Laboratory Evaluation Sample (LLELES)
This sample is similar to the LLFES except it is sent directly to the analytical laboratory without undergoing any field
manipulations. It is used to determine method detection limit, and the presence or absence of laboratory contamination.
We recommend using the same sample as the LLFES to allow isolation and identification of the source of contamination.
5. Field Matrix Spike (FMS)
This is a routine sample spiked with the contaminant of interest in the field. Because of the inherent problems associated
with the spiking procedure and recovery it is not recommended for use in field studies.
6. Field Duplicate (FD)
An additional sample taken near the routine field sample to determine total within-batch measurement variability. The
differences in the measurements of duplicate and associated samples are in part caused by the short-range spatial
variability (heterogeneity) in the soil and are associated with the measurement error in the field crew's selection of the
soil volume to be the physical sample (i.e., two crews sent to the same sampling site, or the same crew sent at different
times, would be unlikely to choose exactly the same spot to sample).
7. Preparation Split (PS)
After a routine sample is homogenized, a subsample is taken for use as the routine laboratory sample. If an additional
subsample is taken from the routine field sample in the same way as the routine sample, this additional sample is called a
preparation split. The preparation split allows estimation of error variability arising from the subsampling process and
from aH sources of error following subsampling. This sample might also be sent to a reference laboratory to check for
laboratory bias or to estimate inter-laboratory variability. These samples have also been called replicates.
Single-Blind Samples
r. Field Rinsate Blanks (FRB)
These samples, also called field blanks, decontamination blanks, equipment blanks, and dynamic blanks, are obtained by
running distilled, deionized (DDI) water through the sampling equipment after decontamination to test for any residual
contamination.
2. Preparation Rinsate Blank (PRB)
These samples, also called sample bank blanks, are obtained by passing DDI water through the sample preparation
apparatus after cleaning in order to check for residual contamination.
3, Trip Blank (TB)
These samples are used when volatile organics are sampled, and consist of actual sample containers filled with ASTM
Type II water, and are kept with the routine samples throughout the sampling event. They are then packaged for shipment
with the routine samples and sent with each shipping container to the laboratory. This sample is used to determine the
presence or absence of contamination during shipment.
Non-Blind Samples
These samples (e.g. Laboratory Control Samples (LCS)) are used in the Contract Laboratory Program to assess bias and
precision. For convenience, these samples are described in Appendix E of the report with the definitions being adapted
from the CLP Inorganic Statement of Work #788.
-------
The Rationale
A basic recommendation in the report
is that the number of quality assessment
samples of any given type that are
employed in a study should be a function
of how the information from the samples
will be employed. For example, if a type
of QA sample is to be used in the
estimation of the total measurement error
variance for the study (i.e., to determine
the measurement precision), then it is
important that a sufficient number of the
samples be employed so that the
estimate of the total measurement error
variance can be used with confidence in
the evaluation of study results and in
defending possible court challenges to
study conclusions. The quality of the
common estimate, s2 of a variance
depends on the number of degrees of
freedom (a number directly related to the
number of QA samples used in the
computation) for the estimate. For
example, from Table 2, if the number of
degrees of freedom for a variance
estimate, s2, is only 2 and the distribution
of the measurements is normal, a 95
percent confidence interval for the true
variance, 02, is from 0.27s2 to 39.21s2
whereas, if the estimate had been based
on 20 degrees of freedom, the 95 percent
interval for the true variance would be
0.58 s2 to 2.08 s2.
Now suppose the data quality objective
(DQO) for total measurement error
variance is that this variance be not
greater than 5, and after the survey the
total measurement error variance
estimate s2 is based on only 2 degrees of
freedom and s2 = 1. One could not be
confident that the true variance is not
greater than 5 since the upper 95 percent
confidence limit for the true variance is
39.21. However, if the estimate, s2= 1,
had been based on 20 degrees of
freedom, one could be reasonably
confident that the true total measurement
error variance is considerably less than 5
since the upper limit for the 95 percent
confidence interval is 2.08. In establishing
soil survey DQOs, one should establish
data quality objectives for the estimates
of data quality and use these quality
objectives in determining the number of
QA samples required to reach the
objectives.
While estimates of measurement
precision may be obtained by estimating
appropriate variance components,
estimates of bias are much more difficult
to obtain. Bias caused by contamination
is a positive bias, but may occur in such
a small proportion of samples as to have
little chance of being detected in the QA
samples used to detect contamination.
Bias related to causes other than
contamination, such as incomplete
recovery, will often depend on the
individual sample pollutant concentration
and on the individual soil sample matrix.
Hence, estimates of bias based on field
evaluation samples or external laboratory
evaluation samples may only reflect the
amount of bias for the reference
concentrations and types of soil matrix of
those samples and not the bias in the
routine samples. Bias detection rather
than bias estimation should be the
primary purpose of assessment of bias.
The quality assurance process should
eliminate bias rather than try to estimate
and adjust for it.
Unfortunately, variance estimates from
any one type of QA sample will seldom
provide an estimate of a variance
component of interest (e.g., the variance
caused by variation in the process of
actually taking the physical soil sample
from the earth). It often takes a
combination of variance estimates from
different types of QA samples to obtain
an estimate of a variance component of
interest. For illustrative purposes,
consider a study in which soil samples
are taken, submitted, prepared, and
analyzed in n (> 2) batches each
containing r routine samples, one field
duplicate sample, and two field evaluation
samples. The field duplicate and its
associated routine sample encounter all
the possible sources of error from that of
locating the physical volume of soil to be
extracted to the analytical error. However,
the error component associated with the
batch effect is the same for both
samples. Hence, the difference of the
measurements on the two samples
contains no information about the batch
effect but is a sum of the differences in
errors from all other sources of error. It is
for this reason that the variance estimate
calculated from the pair differences over
all batches,
n
- S(FDI-RSl)2/(2n),
is an estimate of the total measurement •
error variance minus the between-batch
error variance, (o2m-o2b). It is important to
have a good estimate of the total
measurement variance, o2m, but it cannot
be obtained from the field duplicate QA
samples alone if the study contains more
than one batch and there are nonzero
batch effects (i.e., a2b > 0). The
difference between the measurements of
the pair of field evaluation samples (FES)
is a sum of differences of all error
components except batch-effect error
and the error associated with the physical
taking of a sample since they are in the
same batch and were not obtained by the
sample taking mechanism. Hence, the
variance estimate based on the
differences of these pairs over the n
batches,
n
S [FES11-FES21]2/(2n)(
t-1
is an unbiased estimate of the total
measurement error minus the sum of the
sample-taking variance and the between-
batch variance, 02m-a2s-o2b. To get the
between-batch variance, it is necessary
to calculate the sample variance of the
batch averages, FESj =
(FES1i + FES2i)/2, of the pairs of field
evaluation samples.
This sample variance,
n
si™,
is an unbiased estimator of (o2m-a2s
Now unbiased estimates s2m, s2s, and
s2b, of total measurement error variance,
sample-taking error variance, and
between-batch error variance may be
obtained by solving for them in the three
equations,
to obtain
*-
and
These equations explain why pairs of
field evaluation • samples were employed
-------
in each batch; for if only one FES had
boon placed in each batch, it would have
been impossible to obtain an unbiased
estimate, sam, of the total measurement
error variance. It should be noted that if
one wants a reasonably accurate
estimate of the total measurement error
variance, it is necessary to have a large
number of degrees of freedom for each
of the estimates, s2FD, S2wFEs. ar>d
Further, it would be wasteful to
Some 95 Percent
Confidence Intervals for
Variance
Table 2.
Degree
of
Freedom
Confidence Interval
2
3
4
5
6
7
a
a
10
11
12
13
14
is
16
17
18
19
20
21
22
23
24
25
30
40
50
100
0,27s2
0.32s2
0.36s2
0.39S2
0.42s2
0.44s2
0.46S2
0.47s2
0.49s2
0.50S2
0.52s2
0.53S2
0.54S2
0.54S2
0.56s2
0.56S2
0.57S2
0,58s2
0.58s2
0.59s2
0.60s2
0.60s2
0.67s2
0.62s2
0.64s2
0.67s2
0.70s2
0.77S2
£ a2 £
£ a2 £
£ az £
£ a2 £
£ a2 £
£ a2 £
£ a2 £
£ a* £
£ az £
£ a2 £
£ a2 £
£ a2 £
£ a2 £
£ a2 £
£ a2 £
£ a2 £
£ a2 £
£ a2 £
£02£
£ a2 £
£02£
£ a2 £
£ a2 £
£02£
£ a2 £
£ a2 £
£ a2 £
£<# £
39.21s2
13.89s2
8.26s2
6.02s2
4.84s2
4.14s2
3.67S2
3.33s2
3.08s2
2.88s2
2.73s2
2.59s2
2.49s2
2.40s2
2.32s2
2.25s2
2.79s2
2.73s2
2.08s2
2.04s2
2.00s2
7.97s2
7.94s2
7.97S2
7.78S2
7.64s2
7.67s2
7.35S2
have a large number of degrees of
freedom for one of these three estimates
and a small number for another, since the
estimate of the total variance cannot be
more precise than the least precise term
in its formula. In the above example, the
variance estimates s2FD. s2WFEs, ancl
s2BFES' h30" n, n, and (n-1) degrees of
freedom respectively. The choice of n will
depend on the DQO for the precision of
these variance estimates. The report
suggests that for estimation of total
measurement error, an n of at least 20
might be a reasonable DQO requirement
for most studies.
An Alternative Method for
Assessing Variability without
Field Evaluation Sample
The basic use of the FES in the
preceding section was to estimate
between batch variance. As an
alternative, it is suggested that additional
field duplicates may be employed for this
purpose. One may go back to a particular
sampling location (i.e., a point at which
one sample of soil is taken), and take a
fresh (collocated) sample to include with
each batch (or with at least 21 randomly
selected batches if there are a larger
number of batches). If it is difficult to take
so many collocated samples from one
sampling location, one might use two or
three such locations and take collocated
samples to include in the batches,
alternating between locations (e.g., for
two sampling locations A and B, batch 1
has a collocated sample from location A,
batch 2 from location B, batch 3 from
location A, ...). By comparing the
variability between collocated samples
that are collected and analyzed in
different batches, with the variability
within the field-duplicate-and-associated-
routine-sample pairs, one can estimate
the variability contributed by changes in
the measurement process between
batches. These collocated samples are
actually field duplicates, but because
they are used in a different way than the
field duplicates encountered in the
previous section, they will be identified as
batch field duplicates (BFD).
As with the process utilizing field
evaluation samples,
n
s?, - S(FDt-RSt)V(2n),
is an estimate of the total measurement
error variance minus the between-batch
error variance, (o2m-o2b).
The estimate of variance obtained from
field duplicates inserted in each batch,
m
provides an assessment of measurement
error variance, 02m.
(1) This equation is appropriate when
the m batch field duplicate samples
are all taken from one sampling
location. BFD is the sample mean
of the m samples.
(2) This equation is appropriate when
the batch field duplicate samples
are from L locations with rrij (>1)
BFDs coming from sampling
location j. BFDj is the sample mean
of the nrij samples taken for location
j-
The difference between the variance
estimates utilizing field duplicates and
batch field duplicates (s2BFD~s2FD
provides an estimate of between batch
variance,
-------
-------
The EPA authors, J. Jeffrey Van Ee (also the EPA Project Officer, see below)
and Louis J. Blume, are with the Environmental Monitoring Systems
Laboratory, Las Vegas, NV 89193-3478. Thomas H. Starks is with the
Environmental Research Center, UNLV, Las Vegas, NV 89154.
The complete report, entitled "A Rationale for the Assessment of Errors in the
Sampling of Soils," (Order No. PB 90-242 306; Cost: $17.00, subfect to
change) will be available only from:
National Technical Information Service
5285 Port Royal Road
Springfield, VA 22161
Telephone: 703-487-4650
The EPA Project Officer can be contacted at:
Environmental Monitoring Systems Laboratory
U.S. Environmental Protection Agency
Las Vegas, NV 89193-3478
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
Official Business
Penalty for Private Use $300
EPA/600/S4-90/013
------- |