United States
               Environmental Protection
               Agency
                 Office of Prevention, Pesticides
                 and Toxic Substances
                 Washington, DC 20460
EPA 747-R-93-009
August 1993
&EPA
               Office of Prevention, Pesticides, and Toxic Substances
SAMPLING GUIDANCE FOR SCRAP
METAL SHREDDERS

Field Manual
                  Grid superimposed over
                  material to be sampled
                     I
          Take samples from approximate
           centers of squares in the grid.
                                               XXX
                                               XXX
                                               XXX


                                             All samples
                                              combined
                                               in one
                                               bucket

-------
    SAMPLING GUIDANCE FOR SCRAP
          METAL SHREDDERS

              Field Manual


              August 1993
  U S. Environmental Protection Agency
  Region 5, Library g»H2J)
  77 West Jackson Boulevard, 1201 rw*
  Chicago, IL 60604-3590
United States Environmental Protection Agency
      Office of Prevention, Pesticides
           and Toxic Substances
         Washington, DC 20460

-------
                      DISCLAIMER

      This document has been  reviewed and approved for
publication by the Office of Pollution Prevention and Toxics, U.S.
Environmental Protection  Agency. The use of trade names or
commercial products does  not constitute Agency endorsement or
recommendation for use.

-------
                Authors and Contributors
Westat Project  Staff:

       James Bethel
       Ralph DiGaetano
       Mary Peppier
EPA  Project  Staff:

       Edith Sterrett, Project Officer
       Susan Dillman, Task Manager
                        -111-

-------
                                 Acknowledgments

             The Office of Pollution Prevention and Toxics wishes to thank everyone involved
with this project at Westat and at the Evironmental Protection Agency, as well as others that have
contributed to the document.

             In particular, we acknowledge the technical help of John Rogers and William
Devlin of Westat in preparing this document.  Mary Lou Pieranunzi, Angelia Murphy, Maida
Montes and Anna Page also helped to prepare and proof-read the manuscript.

             We are grateful to the many reviewers in the Office of Pollution Prevention and
Toxics and in other branches of the Environmental Protection Agency for reading the document.
In particular, we thank Brad Schultz and Dan Reinhart of the Exposure Evaluation Division.

             We owe special thanks to Mitchell D. Erickson of Argonne National Laboratory,
David N.  Speis of Environmental Testing and Certification Corporation, and Herschel Cutler of the
Institute of Scrap Recycling Industries for reviewing this document and contributing many useful
comments about it Their review does not constitute approval.
                                     -IV-

-------
                         TABLE OF CONTENTS


Section                                                                Page

  1.      Introduction  	   1

  2.      Sampling Procedures   	   7

         2.1     Basic Sampling Guidelines	   7
         2.2     Sampling Ruff	  13
         2.3     Quality Assurance	  18

  3.      Preparation for Analysis	  20

         3.1     Preparing Fluff Samples for Laboratory Analysis	  20
         3.2     Compositing 	  21

  4.      Evaluating  Sample   Results	  24

         4.1     Possible Sources of Error	  24
         4.2     Confidence Intervals	  24
         4.3     SampleSizes	  29
         4.4     Analytical Methods  for other  Objectives	  32
         4.5     Additional Reading	  32


                              LIST OF TABLES

Table                                                                   Page

  Worksheet 1: Calculation of Average and Standard Deviation	  26

  Worksheet 2: Calculation of Confidence Intervals	  27

  1       t-values for confidence intervals	  28

  2       Relative errors for estimating PCB levels with sample sizes of 2 to 25	  31


                              LIST  OF  FIGURES

Figure                                                                  Page

  1       Schematic illustration of shredder process	   2

  2       Illustration  of grid sampling	   9

  3      Replicated grid sampling	  10

  4      Sampling over time	  12
                                    -v-

-------
                       LIST OF FIGURES (Continued)

Figure                                                                 Page

  5      How to sample stored fluff	 16

  6      Guidelines for compositing samples	 22


 APPENDIX:  ANALYTICAL  METHODS FOR REGULATORY PROCEDURES

Section                                                                Page

  A.I.   Introduction	  A-l

         A.1.1      Objectives of Regulatory Procedures	  A-l
         A. 1.2      Sampling Issues	  A-2
         A. 1.3      Hypothesis Testing	  A-3

  A. 2.   Monitoring	  A-3

         A.2.1      Considerations in Monitoring Programs	  A-3
         A.2.2      Hypothesis Testing for Monitoring Programs	  A-4
         A.2.3      Effects of Sampling and Analytical Error	  A-5

  A.3.   Clean-up Verification	  A-8

         A.3.1      Consideration in Clean-up Verification	  A-8
         A.3.2      Hypothesis Testing for Clean-up Verification	  A-12
         A.3.3      Effects of Sampling and Analytical Error	  A-12
         A.3.4      What to Do When Clean-Up Is Not Verified	  A-18


                        LIST OF APPENDIX TABLES

 Table                                                                 Page

  Worksheet A-1: Hypothesis Testing for Monitoring PCB Levels	  A-6

  A-l    Cut-off  values  for monitoring	  A-7

  A-2    r-values for hypothesis testing	  A-8

  A-3    Chance of finding violations in monitoring with a 25 ppm standard	  A-10

  A-4    Chance of finding violations in monitoring with a 50 ppm standaid	  A-11

  A-5    Chance of finding violations in monitoring with a 100 ppm standard	  A-12
                                    -VI-

-------
                 LIST OF APPENDIX TABLES  (Continued)



Table                                                                  Page





 Worksheet A-2: Hypothesis Testing for Verifying Clean-Up of PCB's	  A-14




 A-6     Cut-off values for clean-up verification	  A-15




 A-7     Chance of requiring additional clean-up with a 25 ppm standard	  A-16




 A-8     Chance of requiring additional clean-up with a 50 ppm standard	  A-17




 A-9     Chance of requiring additional clean-up with a 100 ppm standard  	  A-18
                                  -vu-

-------
                               1.    INTRODUCTION
             Purpose of this Document. The purpose of this document is to provide basic
instructions for collecting and statistically analyzing samples of materials that are produced as a
result of shredding automobiles, refrigerators, washing machines, and other metal objects.
Shredders constitute an important component of this country's environmental management
program, annually recycling 6-9 million cars, 19 million appliances, and 10 million tons of scrap
metal.  Unfortunately, the by-products of these recycling operations may,  in some cases, contain
significant concentrations of polychlorinated biphenyl's (PCBs) or other toxic substances, notably
lead and cadmium. As a result, communities,  environmental agencies, and shredder operators
have expressed concern over the possibility of contamination in waste  products generated at
shredder sites and have indicated a need for guidance in assessing the presence of toxic substances
in these materials.

              Previous Studies.  Several States have done exploratory studies of shredder sites.
Analysis of approximately 200 samples of waste materials collected at shredder sites have revealed
concentrations of PCBs ranging from 0 to 1,242  parts per million (ppm).

              Based on concerns raised  by these studies, the U.S. Environmental Protection
Agency (USEPA) has gathered samples  of various waste materials at seven shredder sites
distributed across the United States.1  In this study, analysis of samples of PCBs revealed
concentrations  ranging as high as 870 ppm. The  same study found concentrations  of lead and
cadmium ranging as high as 43,000 ppm and 200 ppm, respectively. Information from these prior
studies, particularly  the one done by the  USEPA, has been used in developing the sampling
methods discussed in this document.

              Shredder Output Streams.  Shredders are very large machines that convert
autos, truck bodies and other light gauge metal objects into fist size or smaller pieces of scrap
metal.2  A typical shredder operation is depicted  schematically in Figure 1.  The actual "shredding"
1 PCB, Lead, and Cadmium Levels in Shredder Waste Materials: A Pilot Study. USEPA, Office of Toxic
Substances. EPA 560/5-90-008B. 1991.
2 The technical background for this section is based on material taken from PCB, Lead, and Cadmium Levels in
Shredder Waste Materials: A Pilot Study, ibid.; on Chapters 1 and 2 of Analytical Chemistry of PCBs, by Mitchell
D. Erickson, Butterworth Publishers, 1986; and on conversations with shredder operators and environmental
consultants specializing in scrap metal recycling.

-------
                 Input: Automobiles,
               Appliances, Metal Objects
               Hamniermill: Objects are
               reduced to fist sized pieces
                Magnetic conveyer belts
                separate ferrous  from
                nonferrous materials
 Ferrous Stream
Nonferrous Stream
  Air Cyclone/
 Water Separator
   Air Cyclone/
 Water Separator
Figure 1. Schematic illustration of shredder process
                          —2—

-------
is accomplished by a large hammer mill, after which the resulting output is sorted into three main
output streams:
             •      Ferrous metals,
             •      Nonferrous metals, and
                    Fluff.

             Fluff is extremely heterogeneous. While it consists largely of plastic and foam, it
may also contain pieces of metal, rubber, fabric, wire, and other materials. In general, it has a
fibrous, "fluffy" appearance, at least when viewed from a distance.  The initial separation into
ferrous and nonferrous materials is carried out using magnetic devices. After this step, metal and
fluff are separated using either air cyclone or water separation processes. In addition, nonferrous
metals are often subjected to some hand-sorting as well.  Both ferrous and nonferrous metals are
recycled, while fluff is typically deposited in landfills.

             It should be noted that this is a description of a "typical" shredder, but there are
many types of shredders and the instructions in this document may have to be adapted for special
circumstances at a given location.

             How PCBs Enter Output Streams.  PCBs enter output streams when materials
containing PCB-bearing  fluids are shredded.   PCB-bearing fluids  have been  used in the
construction of  capacitors, transformers, electric motors, air conditioners, and hydraulic devices.
PCBs have also been used as additives in pesticides, paints, sealants, and plastics.

             The materials processed at shredder sites may be roughly categorized as follows:

                    Motor vehicles, including  passenger cars, light trucks, vans and  small
                    school buses: In such vehicles, PCBs may be found in paint, hydraulic
                    fluids, oil capacitors, plastic materials, and in oily dust accumulated from
                    roads.
             •      Appliances,  including refrigerators, washers, dryers, dishwashers,
                    freezers, ranges, air-conditioners, microwaves, and hot water heaters:
                    These materials are generally called "white goods."  In white goods,  PCBs
                    may be found in capacitors and electric  motors.
             •      Other materials, such as scrap metals, or industrial or office equipment:
                    PCBs  mignt  be found in oil-filled capacitors,  plastics,  paints, and
                    adhesives.
                                          -3-

-------
             When objects containing PCB-bearing fluids are shredded, the fluids are dispersed
and may be absorbed by the fluff, or the fluids may coat metal and plastic objects.  Similarly, when
plastics or painted objects are shredded, PCBs in paniculate form may enter  the fluff output
stream. In any case, the concentration of PCBs in (or on) materials produced at shredder sites may
pose an unreasonable risk to health or the environment

             PCBs have been regulated by the Toxic  Substances Control Act (TSCA) since
1976. According to these regulations,  materials that contain PCBs in a concentration of 50 ppm or
more must be disposed of in a chemical waste landfill, boiler or incinerator approved under TSCA.
EPA has determined that fluff is regulated under TSCA, 40 C.F.R, Part 761. The U.S. Shredding
Industry produces approximately three million tons of fluff a year. If widespread contamination
were found and the materials were deposited in TSCA landfills, the demand for these landfills
could exceed their capacity due to the volume of fluff.

             Where to Look for PCBs and  Other Toxic Substances.  Very little is known
about the volume and distribution of PCBs at shredder sites.  It is generally suspected that PCBs
are much more likely to enter output streams when processing white goods than motor vehicles
because of the higher prevalence of electric motors in the  former.  Because of this, many operators
refuse to process white goods, while  others accept them only if the motors have been removed.
Those operators that do process white goods typically "mix" them with motor vehicles, usually at a
rate of about 10% or less white goods (by weight).

             When PCBs are present at a given site, it is generally expected that they would be
found in fluff because of its absorbent nature.  While metal output may  be coated with PCB-
bearing fluids, it seems unlikely that the coating  would contain enough PCBs to constitute a health
hazard. PCBs may be present in the soil at shredder  sites, particularly in locations where fluff
accumulates or is moved for storage.  However, it must be stressed that very little is known about
levels  of PCBs  at shredder sites and the  possible contamination of materials produced by
shredders.

             Even less is known about other toxic substances  that may be present at shredder
sites. Lead and cadmium may enter output streams from paint and metal plating on component
parts in motor vehicles. Unlike PCBs, lead and cadmium are not typically suspended in fluids, but
they might adhere to particles of fluff as materials are shredded.

-------
             Sampling  Objectives.  There are several possible objectives in sampling for
PCBs.  At the time of this writing, no one knows very much about the presence of PCBs at
shredder sites. Large concentrations of PCBs have been identified in some samples that have been
collected; some of these findings have been questioned, based on data collection procedures and/or
analytical methods. Thus, agencies may wish to collect data at shredder sites in order to study the
situation in their locality.  In such studies, the objective is simply to gather data and make a
preliminary assessment of possible contamination, as measured by the overall concentration of
PCBs, without any preconceived ideas about whether such contamination exists.

             Another objective is to monitor the output of one or more shredder sites. In this
situation, the monitoring agency - which may  be the shredder operator or an outside agency -
develops a program of regular sampling and analysis of materials to assure that shredder output
meets specified standards.

             In the event that a shredder site or output from a site is established as being
contaminated with PCBs - if large piles of stored fluff or the soil around the site are known to
contain high concentrations of PCBs, for example - then it may become necessary for the site to
undergo some form of clean-up or change in operating procedures.  Thus, a third objective of
sampling might be to collect data to verify that a site is free of PCBs.

             The sampling procedures described in this document are  intended  to produce
representative samples of fluff that will give reasonably accurate estimates of the overall
concentration of PCBs in the  material being sampled. The sampling methods are suitable for any
of the  objectives described above. The document primarily addresses  analytical methods for
exploratory  studies; an  appendix discusses analytical methods for monitoring and clean-up
verification.

             Contents of This Document.  The document consists of three main parts.  In
Chapter 2, we will discuss procedures for selecting samples of fluff and other media at shredder
sites.  Next, in Chapter 3, we will discuss subsampling and other issues in laboratory testing.
Finally, in Chapter 4, we will discuss statistical procedures for deriving conclusions after the data
have been analyzed at the laboratory. The methods discussed in Chapter 4 are intended for
exploratory studies undertaken to assess the extent of PCB contamination, if any, at one or more
shredder sites. Analytical methods for regulatory procedures are discussed in an appendix.
                                         -5-

-------
             This document is intended for users of all backgrounds and no special statistical
knowledge is required.  The statistical background and technical justification for the material
presented here is given in a companion volume.1

             Cautions about Using This Document. This document consists of directions
for collecting and analyzing samples of materials at shredder sites. The sampling plans, estimated
sample size requirements, and the accuracy of statistical tests that are discussed in this document
are based on data from samples collected at seven different shredder sites located throughout the
United States. Although it is not likely, the data that you encounter at your shredder (or the site
you are investigating) may differ substantially from the data used to develop the guidelines in this
document.  If this occurs, the sample sizes shown in tables in this document may yield results that
are somewhat more or less precise than you would expect based on the parameters discussed in
Section 4 and in the appendix.
 1 Sampling Guidance for Scrap Metal Shredders: Technical Background. USEPA, Office of Pollution Prevention
 and Toxics.  EPA/560/5-9 i-002.

-------
                         2.     SAMPLING  PROCEDURES
2.1          Basic Sampling Guidelines

             Overview. The purpose of the field sampling procedures described in this section
is to estimate the overall concentration of PCBs, rather than to identify "hot spots" with high
concentrations.  Thus the sampling methods described here are intended to produce representative
samples of fluff, since this material is generally considered to be the most likely to contain PCBs,
if they are present at all.

             Fluff is often stored in piles on the shredder site before being shipped to a landfill
for disposal.  We will differentiate between  stored fluff, which is stored in piles at the shredder
site, and fresh fluff, which is produced at the site while sampling is being done. In particular, we
will  describe different sampling procedures for stored and fresh fluff.  The former may consist of
very large piles which are difficult to access, while the latter is being continuously produced and is
generally easier to sample.

             In collecting samples, care should be taken to minimize the disruption of the normal
operations of the shredder.  This is important not only from the standpoint of maintaining good
relations with the shredder operator, but also  because the samples collected should, to the greatest
extent possible, reflect the normal output of  the shredder. If shredding procedures are altered in
order to collect samples, the data collected may not reflect the usual PCB content (if any) of the
shredder output streams.

             How  Large Should Samples Be?  The materials present in fluff  are very
heterogeneous, and samples must be relatively large in volume to get a good cross-section of the
types of materials present.  In most cases, we suggest taking  individual samples of about one
gallon in size.  Many of the  sampling procedures  we recommend require combining several
samples of which each is one-half to one gallon in size.  In any case, we recommend that the total
volume of fluff collected at a site be at least five gallons.1

             Duration of the Sampling Period. When sampling from the stream of fresh fluff
as it  is being produced, the duration of the sampling period is an important consideration.  Samples
1 This recommendation is based on techniques for sampling heterogeneous materials presented in a seminar titled
"Sampling Methodologies for Monitoring the Environment" by Pierre Gy and Francis Pitard Sampling Consultants.

                                          —7-

-------
may be collected only once during a visit, once each half-hour for several hours, or once each half-
hour for an entire day. The longer the duration of the sampling period, the greater the likelihood of
obtaining  a representative sample of shredder output, since it is more likely that the materials
shredded will be representative over a longer period. It is difficult to give fixed guidelines on how
long to collect samples, but, in general, we suggest collecting samples of fresh shredder output
each half-hour for a period of at least eight hours, or one working day.  In any case, the general
operating procedures followed at the shredder should be considered in deciding how long to make
the sampling period and how frequently to collect samples. For example, if an operator runs white
goods in the morning and automobiles in the afternoon, samples should be taken of each.

             When different types of materials are recycled, the PCB content of the samples may
vary considerably.  Thus, regardless of the duration of the sampling period and the number of
samples collected, the results of one day's sampling cannot be extrapolated to any other day unless
the materials that are recycled on the two days are similar.  Because of the variability in  the
materials  shredded, high or low concentrations of PCBs may be found at one visit but not on a
subsequent visit.  Because of this fact, it is important that the samples collected at a site are as
representative as possible of the usual activities of the shredding operation.

              Collecting Representative Samples.  The basic technique that we recommend
for collecting samples requires two steps.  First, a square, two-dimensional grid is superimposed
over the material that is  to be sampled, as shown in Figure 2.  Stretching strings across the material
is an efficient way of constructing the grid; the cells should be approximately equal in area.  Next,
samples should be taken from each cell in the grid and combined.  This type of sampling is called
grid sampling. It may be applied in sampling  either fresh or stored fluff. The purpose of grid
sampling is to obtain a sample that is  spread throughout the material that is being sampled. Larger
grids (e.g., four squares on each side) may be used, but a three-by-three grid is generally sufficient
for this purpose.

              When sampling material that is spread out in a grid, it is important to dig down into
the material to the bottom. Finer particles will settle down and samples that are simply  grabbed off
the top will not be representative.

              In order to collect more than one  grid sample, use replicated grid sampling. Using
this procedure, multiple samples are taken from each cell and combined in separate  buckets, as
illustrated in Figure 3. Each bucket is analyzed as an independent sample of material.
                                          -8-

-------
             Grid superimposed over
             material to be sampled
                                                               XXX
                                                               XXX
                                                               XXX
                                                            All samples
                                                             combined
                                                               in one
                                                               bucket
Take samples from the approximate
centers of squares in the grid.
                Figure 2.  Illustration of grid sampling
                                -9-

-------
                              One Bucket
X
X
X
X
X
X
X
X
X
                                                          All samples combined
                                                              in one bucket
                              Two Buckets
   *
X    '
 1   '
 ' X
'  X2
                        Bucket 1
                       (9 samples)
                                              Bucket 2
                                             (9 samples)
        X1 and X2 samples
       combined separately
        in buckets 1 and 2
  3   4
X   X
  31  4
                              Four Buckets
                        Bucket 1
                       (9 samples)
                                Bucket 2
                               (9 samples)
                                                        X( - X4  samples combined
                                                        separately in buckets 1-4
 Bucket 3
(9 samples)
 Bucket 4
(9 samples)
                    Figure 3.  Replicated grid sampling
                                     -10-

-------
             In some cases, grid sampling is not a practical option.  For example, when
sampling from large piles of fluff, it will be necessary to collect samples from various points in the
pile without formally creating a grid. Detailed descriptions of how to sample stored fluff will be
discussed below.

             Sampling Over Time. When samples are collected from freshly produced fluff,
samples must be collected at different times; for example, sampling might be done each half-hour
over a 4- or 8-hour period. Figure 4 illustrates the basic technique for sampling over time. Here a
separate grid sample is taken at each point in time, with each time period represented by a different
bucket  Each bucket may consist of 1 gallon or more, but only one bucket per time period should
be collected. If three samples are required, then samples should be collected at three different time
periods (e.g., every 2 hours for a 6-hour period). If more samples are required, then either more
time periods must be sampled (e.g., every hour for a 6-hour period)  or samples must be collected
for a longer duration (e.g., every 2 hours for a 12-hour period).

             How Many Samples Should Be Collected?  The number of samples that need
to be collected depends on the accuracy required. As we will see in more detail later, about 10-20
samples should be sufficient for most purposes. For example, in sampling over time, 16 samples
could be taken at half-hour intervals over the course of an 8-hour work day. These samples can be
combined, using the technique of compositing which will  be discussed later in Section 3.2, to
reduce laboratory costs. Of course, fewer samples can be taken but at the risk of greater error. In
Section 4, we will discuss the trade-offs between sample sizes and the reliability of conclusions.

             What Equipment Should be  Used?  Because of the size and heterogeneity of
materials that are produced at shredder sites, conventional core-sampling tools are usually of little
use.  Front-end loaders and backhoes may be useful for transporting and arranging materials,
particularly if large amounts of fluff are involved.  Similarly, trowels, rakes and shovels may be
useful for smaller amounts of fluff.  Because of the difficulty in manipulating fluff, it may be
necessary to pick it up by hand  and place "grab samples" manually in gallon containers.  If
available, a rotating gravity tumbler drum (RGTD) may be useful for mixing samples.
                                        -11-

-------
  Output from
9:30 to 10:00 a.m.
   Output from
10:00 to 10:30 a.m.
   Output from
10:30 to ll:00a.m.
      Bucket 1
                                Bucket 2
                                Bucket 3
                    Samples Composited into Three Buckets
                      Figure 4: Sampling over time
                                 -12-

-------
             Cleaning Equipment and Handling Samples. Whatever equipment is used, it
must be clean in order to avoid contaminating the samples that are collected.  Furthermore,
equipment should be cleaned regularly, preferably after each sample is taken. To clean shovels,
hoes, buckets, containers, and other equipment, soak them in dilute (20%) nitric acid and then
rinse them three times, first with deionized water, then acetone, and finally hexane. Alternatively,
steam cleaning can be used; if the steam condensate is free of PCBs, it can be disposed of easily.
By comparison, disposal of solvents is always expensive.

             If equipment is not cleaned, samples can  become cross-contaminated.  Cross-
contamination occurs when PCBs from a sample that is contaminated are transmitted to a second
sample which was not previously contaminated. This problem can occur when materials are not
handled carefully and one sample leaks into another, or when equipment is not cleaned and a
residue of PCBs builds up and is transmitted to multiple samples.

             Besides keeping equipment clean, it is important to handle samples carefully. All
samples should be clearly labelled, indicating  the time, date and location.  Samples should be
stored in clean, sturdy containers. If samples are handled manually, gloves should be changed
after collecting each sample.

             Clearly,  the cleaning of equipment can be cumbersome; moreover, it will be
impractical in most circumstances to clean large equipment, such as backhoes.  However, small
equipment and containers should be cleaned as often as possible. While the risk may be small, it is
in the best interests of both  the shredder and environmental agencies that samples be as free as
possible from cross-contamination. Cross-contamination can lead to erroneous conclusions about
the level of toxic substances in the media.  For example, stored fluff may be contaminated by fresh
output, leading to the erroneous belief that the stored material may not be deposited in a sanitary
landfill.  Cross-contamination is especially serious when  it occurs with samples from different
sites, since questions of  liability may be involved.
2.2          Sampling Fluff

              General Guidelines. As described earlier, fluff is generated as a waste product
which is separated from recyclable metals after the shredding operation.  First, ferrous and
nonferrous materials are separated using magnetic devices, and then fluff is separated from the
metals either by using cyclone blowers or by washing with water, most commonly the former.
                                         -13-

-------
Fluff may either pile up below the cyclone separator or it may be removed to storage piles using
conveyor belts.

              There are generally three sources of fluff at a shredder site.  First, fresh fluff is
continuously being produced during the shredder operation. Second, there may be piles of stored
fluff, although most shredder operators regularly ship fluff to avoid wasting storage space. Third,
some fluff, which we will call spillover, is likely to have piled up around conveyor belts and other
equipment. Although the basic sampling procedures are similar, we will give directions for
sampling each form of fluff separately.

              Fresh Fluff:  Front-End Loader  Assisted. We will describe two methods for
sampling fresh fluff, the first of which involves the use of a front-end loader.  This method is
preferred for reasons of safety, sampling consistency, and  minimal facility interruption.

              Briefly,  the front-end loader method involves (1) collecting the fluff in the front-end
loader bucket as it is produced, (2) spreading the collected fluff out on the ground, and (3) taking
samples from the fluff after it has been spread out on the ground. In order to use this method, you
will need a front-end  loader, which should have a safety cab and should  be used only by an
experienced operator. You will also need a clean space of ground on which to spread out the fluff.
In some cases, it may be necessary to arrange with the operator to start and stop the shredder at
appropriate intervals.

              First, the front-end loader bucket  should be positioned under the mouth of the
cyclone (or the end of the conveyor belt, depending on which is used) during shredding to collect
the fluff.  The shredder should run  until the bucket  is full,  typically about 3  minutes, or the
equivalent of about two automobiles.  (Note:  If large objects are being shredded, it is preferable to
process the entire object, rather than part of it.) After the shredder has stopped, move the front-end
loader to an open, clean area  for spreading the fluff. This area should be about 10 feet square, or
large enough that the contents of the  front-end loader can be spread evenly to a depth of about 1
foot.

               Second, have the front-end loader operator spread the collected fluff on the ground
in a square area to an even depth of about 1 foot, using the  back of the bucket.  Divide the square
into nine roughly equal subsections, as shown in Figure 2.  Take one-half gallon of material from
the approximate center of each subsection, using a shovel and digging down into the material;
combine the samples in the 5-gallon bucket. Smaller samples may be collected on a tarpaulin
                                          -14-

-------
placed under the cyclone or conveyor, moved to a clear area and then spread with a rake. For
small samples, four roughly equal subsections may be used, with a half-gallon being selected from
the center of each one.

              At  some sites,  the fluff stream is fed continuously into rolloff boxes which can
contain up to 20 cubic yards of material.  In order to collect samples of fluff at these sites, the
boxes must be pulled away from the output stream, which can then be collected using a front-end
loader as described above.

              Fresh Fluff Sampling Without a Front-End Loader. Arrange for the operator
to shut down the line after shredding material for about 3 minutes. Take five one-gallon samples
as follows.  First, take four one-gallon samples by systematically sampling at four equidistant
points around the perimeter of the pile, approximately 1 foot above the ground. Dig about 18
inches into the pile horizontally, or, depending on the size of the pile, far enough to obtain layers
of fluff deposited at different times. Take the fifth sample from the center of the pile, digging
down about a foot into the pile.

              Stored Fluff.  It is much  more difficult  to obtain representative samples from
stored piles of fluff, but such samples are potentially more useful  because they may be more
representative of the normal output of the shredder.  (We will assume that the stored pile to be
sampled  is large;  small piles can be raked into a square shape, divided into  nine roughly equal
subsections, and sampled as described above for fresh fluff.) In collecting samples from stored
piles of fluff, the objective is to obtain samples of the oldest fluff, the deepest fluff, and two
samples of surface fluff.  If a  large pile of new fluff has been stored  next to a smaller pile of old
fluff, then the deepest fluff may not be the oldest.  However, if the oldest fluff is also the deepest,
take a sample half-way between the bottom and the surface in place of the  deepest fluff. The
procedures described below, which are illustrated in Figure 5, will provide a total of 20 one-gallon
samples. To prevent cross-contamination between samples, collect one five-gallon bucket at a
time.

              First, take five one-gallon  samples of surface fluff from the edge of the pile, at
equal distances around the pile, one foot off the ground.   Dig straight into the surface, including
the actual surface material in the sample.
                                         -15-

-------
            Notch #2
               etc.
   Edge of pile
            Notch #1
                                               Deepest
                                                                               Surface
                                                               Midway between
                                                                top and bottom
                              Figure 5.  How to sample stored fluff
1. Take five one-gallon samples of fluff at equal distances around the edge of the pile.

2. Cut five notches at equal distances around the pile and take a one-gallon sample from the deepest
   fluff in each notch.

3. Take five one-gallon samples of the oldest fluff.

4. Take five one-gallon samples of fluff from the surface of the pile.
                                             -16-

-------
             Second, use heavy moving equipment (such as a front-end loader) to cut five
notches in the pile for the other samples, as shown in Figure 4. These notches should be located at
equal distances along the perimeter of the pile, if possible. From each notch, take a one-gallon
sample from the fluff that is deepest down in the pile. Some care may be required to get a sample
of the deepest fluff in  the notch, since fluff from the surface may fall down into the notch. One
approach would be to have the operator remove upper layers of the pile before cutting the notch; it
might also help to take the sample from the center of the notch, rather than the sides where material
is more likely to fall into the notch.  In making notches and collecting samples, remember that
safety is a paramount consideration. Do not cut notches deeper than five feet in height Proceed
with caution at all times.

              Third, collect five one-gallon samples of the oldest fluff. You will have to ask the
shredder operator which fluff is the oldest. It may be a particular area of the fluff pile, or it may be
the deepest layer. If it is not known which fluff is the oldest, then take a one-gallon sample from a
point mid-way between the bottom of the pile and the surface in each of the notches.

              Finally, collect five one-gallon samples of fluff from the surface of the pile at points
near the center of the pile.  The notches may provide easy access to points near the center of the
pile.

              As noted above, this procedure  will result in  20 samples.  After reviewing
Section 4, which discusses analyzing the samples, you may decide that more samples are needed.
The number of samples may be increased by taking more samples at each of the steps described
above.  For example, if six samples are taken from the perimeter, six notches are cut, etc., six
samples of the deepest fluff are taken, and so forth, there will be 24 samples.

              Spillover. During normal shredding operations, fluff will pile up along conveyor
belts and  cyclone separators. We will refer to this fluff as spillover. Spillover tends to consist of
smaller particles, sometimes called "fines".  Because these "fines" are suspected of being more
susceptible to PCB contamination, you may want to take some samples of this material.

              Inspect the area  along the conveyor belt for spillover. Take five one-gallon samples
of any spillover material along  the conveyor belt at approximately equal distances.  Mix these five
one-gallon samples into one five-gallon bucket. If desired, repeat this procedure to fill additional
buckets. In some cases, the pattern of spillover may not be regular enough to use this strategy. If
necessary, identify the areas where spillover exists and take a one-gallon sample (or more) from
                                          -17-

-------
each location to achieve one five-gallon sample (or more) that is representative of the spillover
material.
2.3          Quality Assurance

              The Necessity for Quality Assurance. There are many sources of error in
evaluating contamination by PCBs or other substances. First, since we are selecting samples of
material to analyze, there is sampling error, which is due to the fact that not all of the material is
being analyzed and thus there is variability in the results from one sample to another. (Please note
that sampling "error" is a statistical term which reflects the natural variation that exists from one
sample to another.  This term does not imply any "error" on the  part of those collecting the
samples!)  Second, there is analytical error, which results  from  the difficulty of accurately
identifying and quantifying the substances present in a given sample of material.  Third, there is the
possibility of errors through cross-contamination, which results from PCBs (or other substances)
being introduced into a sample during the collection process. For example, PCBs might be present
in the buckets used for data collection and then transferred  to the fluff during the process of
collecting samples.

              Below we describe two quality control procedures. The first, the use of field
blanks, will help to  detect the presence of cross-contamination.  The second,  the analysis of
duplicate samples, will help to quantify analytical error.

              More extensive treatment of quality control issues can be found in the  following
publications:

              OTS Guidance Document for the Preparation of Quality Assurance Project Plans.
              USEPA, Office of Toxic Substances.
              Test Methods for Evaluation Solid Waste. USEPA, Office of Solid Waste and
              Emergency Response.  SW-846, Third Edition. 1986
              Analytical Chemistry  of PCBs, Mitchell D. Erickson. Butterworth Publishers,
              Stoneham, Massachusetts.  1986.
              Field Blanks. Field blanks are materials that are known not to contain PCBs, but
 which are handled using the procedures specified for collecting fluff, soil or other materials which
 are suspected of being contaminated. When the field blanks are analyzed, they should not contain

                                          -18-

-------
any PCBs. Empty containers, such as buckets, should be taken to the site, opened for the duration
of the time that sampling is done, and then closed and taken to the laboratory, where wipe samples
can be taken and analyzed. This procedure will indicate whether containers were contaminated
either before data collection or through improper handling. The use of field blanks helps protect
the operator by indicating when samples are being collected improperly and possibly giving
incorrect findings.

             Duplicate Analyses.  As a general practice, at least 10% of the samples selected
should be analyzed in duplicate, meaning that the same sample (or parts of it) should be analyzed
twice. In particular, if one sample has an extremely high concentration of PCBs relative to other
samples, replicates should be analyzed for verification; Section 3 will discuss how replicates are
formed.  Preliminary studies suggest that laboratory or analytical error for the procedures described
in this manual are, on average,  about 30% of the estimated PCB level, ranging from 5% to 80%.
If the results for  replicates vary  by  more than this, it may be due to inadequate  laboratory
procedures.
                                         -19-

-------
                      3.     PREPARATION FOR ANALYSIS
3.1          Preparing Fluff Samples  for Laboratory Analysis

             Overview.  After samples are collected in the field, they must be prepared for
laboratory analysis.  Because of the extreme heterogeneity in some of these materials, one part of
the sample can give an estimate which is not representative of the whole.  In this section we will
discuss procedures for splitting the collected samples into several replicates so that each replicate is
representative of the original sample, containing the same components in approximately the same
proportions. One or more of these replicates can then be analyzed to test for PCB contamination.
The reason for creating such replicates is, first, to reduce the amount of material that is actually
subjected to laboratory analysis, and,  second, to create backup replicates for retesting if this
becomes necessary. Altogether, at least five gallons of material should be prepared for analysis,
with about 400-500 grams of this material actually undergoing analysis. In Section 3.2, we will
discuss compositing, a technique for combining samples to reduce laboratory costs.

             Step 1:  Weigh the Fluff Sample.  Determine the weight of the entire fluff
sample. Since 400-500 grams of fluff are required for each replicate, weighing will indicate what
fraction of each bucket of material will comprise a replicate. Generally, a five-gallon bucket of
material will produce about eight replicates.  However, if the  weight of your fluff sample is
substantially smaller than 3,200 grams or larger than 4,000 grams, then divide the weight of the
sample by 450 to determine the number of replicates.

             Step 2: Sort Out Large  Pieces of Material. Pour the contents of the bucket
onto a 9.5 mm screen above a laboratory tray or table with a nonabsorbent surface. Pieces that do
not pass through the screen should be cut into pieces or milled until they are small enough to pass
through the screen and then mixed into the sample.  Larger pieces of material (metal, atypical wire,
hard plastics) that cannot be cut with  shears should be segregated. Smaller pieces of wire or other
solid material that are distributed uniformly throughout the sample should remain with the sample.

             Step 3: Divide Material into  Replicates. Uniformly distribute the fluff which
remains over the tray or table. This material will vary in composition, and dense granular materials
(e.g., dirt, pulverized metal, plastics, glass,  ceramics,  etc.) will tend to settle below lighter
material, such as shredded fabric  and foam rubber.  Care  must be taken to ensure that these
components of the fluff are uniformly distributed throughout the tray.
                                         -20-

-------
             Using the information on the total weight of each sample, divi.de the fluff on the
table into approximately equal pans, with the number of parts being equal to the number of
replicates to be obtained. In most cases, you will divide the material on the table into eight roughly
equal parts to form eight replicates.

             Step 4: Cut Large Pieces and Distribute Among Replicates. In Step 2, large
pieces that could not be easily cut were removed and set aside. Now cut these pieces with either tin
snips or a hack saw, assuming that the materials can be cut using one of these tools, and distribute
the pieces of the material equally among the replicates. If both cutting methods fail, the material
should be analyzed separately, and any detected PCB  levels should be prorated based on the
number of replicates, the weight of the replicate, and the weight of the material. For example,
suppose that eight replicates are produced, each  weighing about 450 grams, and a large piece of
material, weighing about 50 grams, cannot be cut.  If the piece of material is analyzed and shown
to have a PCB level of 30 ppm, then the revised PCB level for any replicate that is analyzed should
be calculated as

                                  (30)(5Q) + (Replicate PCBs)(450)
              Revised PCB Level =
                                                +(450)

              Step 5: Place Replicates in Containers.  Place each replicate in a container.
 Seal, label and number the container so that both the replicate number and original bucket number
 are included (e.g., Replicate #2 of 4 from Bucket #12).
3.2          Compositing

              Because of the expense of analyzing samples at the laboratory, equal sized parts of
two or more different samples are sometimes mixed together and sent to the laboratory for analysis
as if the mixture were only one sample. Samples can also be composited after the preparatory
steps described in Section 3.1; this method is prefereable to compositing in the field, although it
may be less cost effective. We will refer to the mixed sample as a composite  sample (or simply a
composite) and to the parts that were mixed together as subsamples. This procedure is illustrated
in Figure 6. Because the subsamples have been mixed, the concentration of PCBs or other toxic
substances in the composite sample should be roughly equal to the  average of the concentrations
                                         -21-

-------
60

1
 in



1
   00


"Q/N
r* ^"*

 Q.


 O >>



•£ 3*



is-c
^ *-»


Tf
                                 -22-

-------
that would have been obtained by analyzing the subsamples individually, even  though the
concentrations in the subsamples may vary substantially due to the heterogeneous nature of fluff.
Assuming that laboratory errors are not large compared with sampling error - which is almost
always the case when analyzing samples  of fluff - compositing effectively reduces the cost of
laboratory analysis while maintaining about the same level of accuracy as if the samples had been
analyzed individually.

             When  forming composite samples, several general rules should be followed.
First, mix each sample thoroughly before compositing.  Second, divide each isample into three or
four parts, or subsamples. All the subsamples must be of roughly equal size.  One simple method
for dividing the sample is to spread the sample out on a clean area and split it into two, then four,
equal parts.  Another method is to take scoops of the material and put the first scoop in the first
subsample, the second scoop in the second subsample, the third in the third subsample, and so on,
repeating the process until the material is exhausted. Finally, take one subsample from each of the
samples and combine them to make up the composite sample.  Mix  the composite sample
thoroughly.

             If the samples are from different sites or different parts of a single shredder (e.g.,
stored and fresh fluff), then use only one subsample - not the entire sample - for compositing. If
large concentrations of toxic substances are found, it may be desirable to analyze part of each
sample separately.

             Throughout the next section we will discuss the effects of compositing on various
analytical procedures.   While compositing  is normally considered to involve two or more
subsamples, it is preferable for simplicity in presenting tables to speak of composite samples which
consist of one or more subsamples. For example, if four samples of fresh fluff are taken over a
period of 4 hours (as described in Section 2.2), these samples might be analyzed as one composite
of four subsamples,  two composites of two subsamples each, or as four "composites" of one
subsample each.
                                         -23-

-------
                     4.     EVALUATING SAMPLE  RESULTS
4.1          Possible  Sources  of Error

             In Section 3.2 we noted that there are several possible sources of error in assessing
contamination by PCBs or other toxic substances.  Specifically, we discussed errors due to
sampling, laboratory analysis, or cross-contamination when the samples are collected.  Cross-
contamination creates bias and can be avoided only by careful handling of materials.  However, the
first two types of errors can be taken into account by using the statistical methods described in this
section. For example, if the laboratory analysis of five samples of fluff at a given site shows an
average PCB concentration of 60 ppm, does this conclusively indicate that the entire output of fluff
from that site actually contains more that 50 ppm? Is it possible that the actual concentration is 45
ppm and the difference (i.e., 60 ppm instead of 45 ppm) is due to sampling error and/or laboratory
error?  In this section we discuss a statistical procedure, called a confidence interval, for answering
such questions.

             Because of the errors associated with the selection and analysis of samples, we
cannot be sure that the numerical value (e.g., an average PCB concentration of 60 ppm) resulting
from a series of laboratory tests is exactly accurate.  Instead we must use statistical analysis to
obtain an interval (e.g., 50 to 70 ppm) which we are relatively sure is accurate. This interval is
called a confidence interval and our degree of certainty is called the level  of confidence.  For
example, based on the results of our statistical calculations, we may be 95% confident that the
actual average concentration is somewhere between 50 and 70 ppm. In Section 4.2 we discuss the
calculations necessary for making statements like this one.
4.2          Confidence Intervals

              Overview. The objective of an exploratory study is to estimate the concentrations
of PCBs or other toxic substances present in the output streams, soil, or other material at a given
shredder site.  Because of the  sampling error and laboratory error, it is not possible to determine
exactly the concentration of toxic substances. However, by using the methods in this section, you
will be able to make statements such as, "As a result of our study, we are 95% certain that the
concentration of PCBs in this pile of stored fluff is between 40 and 100 ppm."  In this statement,
the interval "between 40 and 100 ppm" is called a confidence interval. Because of sampling and
                                         -24-

-------
measurement errors, we are never sure of the exact concentration of a given substance in the
material we are studying.  By calculating confidence intervals, we obtain a range that is likely to
contain the actual concentration.  In this manual, all confidence intervals are calculated to have a
95% chance of being correct - i.e., of including the actual PCB concentration — and are thus called
95% confidence intervals.

             Preliminary Calculations. The first step is to make two basic calculations, the
average and standard deviation of the samples. These calculations are illustrated in Worksheet 1.
In the example given in Worksheet 1, 6 samples are analyzed and found to have measured PCB
concentrations of 5, 15, 65, 11, 33, and 27 ppm, respectively. For these data, the average and
standard deviation are 26 and  21.72 ppm.

              Confidence Intervals for Concentrations.  To find estimates of the actual
concentration of PCBs or other substances, follow the calculations shown in Worksheet 2. For the
example data shown in Worksheets 1 and 2, the lower and upper limits are 3.21 and 48.79  ppm,
respectively, so that we are 95% certain that the estimated PCB level is between 3.21 ppm and
48.79 ppm.

              Interpretation of Estimated Concentrations.  What conclusions can be made
based on the estimates that you have made? There are several ways to answer this first question,
but  the overriding concern  should be whether estimated levels  of PCBs and/or other  toxic
substances are considered to be too  high. Suppose, for example, we regard 50 ppm to be an
acceptable level of PCBs in shredder output. There are three possible cases:
              •      Case 1: The upper  limit of the interval falls below 50 ppm.  In this case,
                    we are  95% certain that the level of PCBs is acceptable.
              •      Case 2: The lower  limit of the interval falls above 50 ppm.  In this case,
                    we are  95% certain that the level of PCBs is not acceptable.
              •      Case 3: The interval contains 50 ppm.  In this case we are unsure as to
                    whether the level of PCBs is acceptable.  If the interval is not too wide
                    (e.g., 45 to 51  ppm) then we might be willing to assume that the level of
                    PCBs is acceptable; otherwise, the study is inconclusive.

With regard to Case 3, it should  be noted that most of the time it can be avoided by specifying a
large enough sample size when planning the  study;  this problem will be discussed shortly.
Furthermore, whenever it is necessary to make an absolute judgment about the safety of shredder
                                         -25-

-------
WORKSHEET 1:  Calculation of Average and Standard Deviation
Example Data. Assume that 6 composite samples are analyzed and are estimated to have these
PCB levels:
PCBs (ppm)
5.0
15.0
65.0
11.0
33.0
27.0
Squared PCBs
25.0
225.0
4,225.0
121.0
1,089.0
729.0

Step 1: Find the sum (£):

            £x = 5+ 15 + ... +27 = 156.0.
Step 2: Find the sum of the squares:

             I x2 = 25 + 225 + ... + 729 = 6,414.0.
Step 3:  Find the average:
             A            £ x       156.0   O.n
            AveraSe = Sample Size =  ~T" 26'°-
Step 4: Find the Standard Deviation:
                             x2
               ,, .                Sample Size
               Vanance   =  Sample Size -1
                                 5

                         = 471.9.


      Standard Deviation    = V Variance = 21.72.
                                      -26-

-------
WORKSHEET 2:  Calculation of Confidence Intervals
Example Data. As in Worksheet 1, the example data consists of laboratory measurements

from 6 composite samples, showing the following PCB levels:
PCBs (ppm)
5.0
15.0
65.0
11.0
33.0
27.0
Squared PCBs
25.0
225.0
4,225.0
121.0
1,089.0
729.0

Step 1: Find the average and standard deviation.  Follow the directions in

Worksheet 1.  For the data shown above:
                Average of Samples = 26.0



                Standard Deviation =   21.72
Step 2: Estimation of Confidence Intervals. In Table 1, find the r-value for a sample

size of 6, which is 2.57. Now make the following calculations:
 and
       A       ro    i        i   Standard Deviation   _,_  _.._ 21.72      ,, 01
       Average of Samples-r-value—.          =— = 26.0-2.57—==—  =   3.21
                                  V Sample Size                  V6
       A       ro    i        i   Standard Deviation    „, _   ___ 21.72    Aa-,r\
       Average of Samples + r-value—         —— =  26.0 + 2.57 —=— =  48.79.
                                  VSample Size                 V6
Step 3: Interpretation of Confidence Intervals. We are 95% certain that the actual


PCB level is between 3.21 and 48.79.
                                      -27-

-------
output, then the hypothesis testing procedures described in the appendix should be used instead of
the exploratory procedures discussed here.

             In each of the preceding scenarios, we have used the expression "95% certain." As
we discussed earlier, there will always be some uncertainty as to the actual concentration of PCBs
because of sampling and laboratory error. When we say that we are 95% certain that the level of
PCBs is within a given range, we simply mean that there is a 5% chance that we are wrong. Put
another way, this means that if we checked PCB  levels  at 20 sites (or at the same site  at 20
different times)  using the  procedures described here, we could expect, on average, that our
estimate for one of the sites would be wrong.
4.3          Sample  Sizes

              Sample Sizes and Relative Error for PCB Levels.  Because of sampling and
laboratory measurement error, we can never be certain of the exact concentration of PCBs.
However, by increasing the number of samples analyzed, we can reduce the degree of error in our
estimates.  How many samples need to be taken? There is no universal answer to this question,
but based on data from preliminary studies, we can make rough estimates of the level of error that
can be expected from samples sizes ranging from 1 to 25.l

              When we select a sample and average the measured PCBs, there is always some
difference between our sample average and the true concentration of PCBs in the sampled material.
This difference represents error that is  due to both sampling and laboratory analysis.  The relative
error is the absolute difference between the sample and true concentrations divided by the true
value:

              ~  ,  .   „           Sample Average-True Concentration
              Relative Error =     	c	~—^	:	
                                           True Concentration

Since the sample average is subject to random fluctuations, the relative error will vary also, and we
will never know the relative error for any given sample. However, as the sample size increases,
 1 The estimates for standard errors, sample sizes and precision presented here are based on preliminary data from an
 EPA-supported study of 85 samples collected at seven shredder sites throughout the country and on a dataset of 200
 samples collected and analyzed by various state and local agencies.

                                          -29-

-------
 Table 1:  t-values for confidence intervals
Number of
composite
samples
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
30
50
75
100
>100
t-values
12.71
4.30
3.18
2.77
2.57
2.45
2.36
2.31
2.26
2.23
2.20
2.18
2.16
2.15
2.13
2.12
2.11
2.10
2.09
2.09
2.08
2.07
2.07
2.06
2.05
2.01
1.99
1.98
1.96
"The values shown in the table are
 taken from Student's t distribution.
 This distribution is often used as a
 measure of uncertainty due to
 sampling and other sources of error
                    -28-

-------
the relative errors decrease and, although the relative error may change from one sample to another,
we can give a value, the maximum relative error, that it will generally not exceed.

             Table 2 shows the maximum relative error for estimating PCS levels with sample
sizes of 1 to 25.  Unfortunately, even to  get 50% maximum relative error may require a large
number of samples. For example, if 10% white goods are processed (with 90% automobiles or
other materials), approximately 25 samples are required to obtain 50% maximum relative error
when no compositing is used.  Notice that when compositing is used, the number of samples that
must be analyzed to achieve a desired maximum relative error is reduced.  For example, 64%
maximum relative error can be expected when 16 samples are analyzed without compositing. If 18
samples are composited into 9 groups of 2 samples each, however, then 68% maximum relative
error can be obtained by analyzing the 9 composited samples.  There is a slight increase in
maximum relative error (since 68% is greater than 64%), but the laboratory costs are reduced
almost by half (i.e., 9 samples analyzed instead  of 16). Finally, notice that to obtain maximum
relative error of less than 25% requires very large sample sizes, even when compositing is used.

             In discussing sampling over time in Section 2,  we recommended taking samples
every half-hour for at least 8 hours, which would result in 16 samples.  From Table 2, we see that
the resulting maximum relative error would be about 64%, if no compositing is used. This will be
adequate when the level of PCBs found is low (e.g., 10 to 20 ppm), but may be unacceptable if a
high level of PCBs is found.  If the 16 samples are composited into 8 composite samples  of 2
subsamples each, the maximum relative error would be about 70% (i.e.,  slightly higher than that
shown for  9 composites of 2 subsamples each).  If the 16 samples  are  composited into 4
composites of 4 subsamples each, the maximum relative error increases  to 106%.  Again, this is
probably acceptable when the level of PCBs is low, but will not be acceptable when the PCB level
is, say, 20 or 30 ppm.   The sampling procedures described in  Section 2 for stored fluff will
produce 20 samples; the maximum relative error for 20 samples would be similar to those for 16
samples, although slightly lower.

             The key factor in deciding how many samples to take is the maximum relative error
desired. In deciding the maximum relative error, the concentration of PCBs must also be taken
into account.  Suppose,  for example, that the actual PCB  concentration is 10 ppm and that we
estimate the level of PCBs as being between 0 and 20 ppm. Then the maximum relative error is
100%, but since the estimated PCB concentration is well below the 50 ppm standard, this level of
error is acceptable. However, if the actual PCB concentration is 50 ppm and we estimate that the
level of PCBs is  between 0 and 100 ppm, the maximum relative error is  again 100%, but it  is
                                        -30-

-------
Table 2: Relative error for estimating PCB levels with sample sizes of 2 to 25
Total
samples
collected
Number of
composites
analyzed
Subsamples
in each
composite
Maximum relative error*
2
4
9
16
25
4
8
18
32
50
8
16
36
64
100
16
32
72
128
200
2
4
9 1
16
25
2
4
9 2
16
25
2
4
9 4
16
25
2
4
9 8
16
25
1084%
192%
93%
64%
50%
793%
140%
68%
47%
36%
597%
106%
51%
35%
27%
468%
83%
40%
28%
21%
*A relative error of 50% means that with 95% certainty, the estimated average
 concentration will be within 50% of the actual average concentration. A
 relative concentration of more than 100% (e.g., 150%) has the same interpretation
 (e.g., the estimated concentration will be between 0% and 1.5 times the actual
 concentration).
                                 -31-

-------
clearly not acceptable.  In exploratory studies, high relative errors can generally be tolerated, since
more data can be collected to investigate the situation more closely if high levels of PCBs are
suspected.

              Sample Sizes and Relative  Error for Lead  and Cadmium.  In general, the
samples sizes required for estimating PCB levels should be more than adequate for estimating
levels of lead and cadmium.  Analysis of preliminary data indicates that both sampling and
measurement errors are smaller for these substances than for PCBs. Comparable data for other
toxic substances is not available.
4.4          Analytical  Methods for Other Objectives

              Exploratory studies are only one possible objective of sampling for PCBs at
shredder sites. Another objective would be monitoring shredder output to make sure that PCB
levels do no exceed a given level. In practice, monitoring programs are often put in place by
shredder operators to verify to landfill operators that fluff from the site meets TSCA landfill
regulations. A third objective would be "clean-up" verification, which might be required if a site -
or the fluff produced at a site - were found to be extensively contaminated with PCBs. In both
cases, the statistical method of hypothesis testing would be used in place of confidence intervals.
These topics are discussed in an appendix.
4.5          Additional Reading
              For more details on statistical procedures for use in environmental sciences, see

              Statistical Methods for Environmental Pollution Monitoring, Richard O. Gilbert.
              Van Nostrand Reinhold Company Inc.  1987.
                                         -32-   .

-------
                                 APPENDIX

                       ANALYTICAL METHODS FOR
                       REGULATORY PROCEDURES
A.I.        Introduction

A. 1.1       Objectives  of Regulatory Procedures

             As discussed in the Section 1, there are several possible objectives in
sampling for PCB's.  Analytical methods for exploratory studies were discussed in Section
4 of the Sampling Guidance. The two objectives of regulatory functions are monitoring
and  clean-up verification.  This appendix discusses statistical methods for these
applications.

             When monitoring the output of a shredder site, the monitoring agency -
which may be the shredder operator or an outside agency - develops a program of regular
sampling and analysis of materials to assure that shredder output meets specified standards.
In this situation, the output is assumed not to be contaminated until the samples collected
for the monitoring program demonstrate otherwise.

             In the event that a shredder site or output from a site is established as being
contaminated with PCB's - if large piles of stored fluff or the soil around the site are
known to contain high concentrations of PCB's, for example - then it may become
necessary for the  site to undergo some form of clean-up or change in operating procedures.
In this case,  the site  (or output from it) is assumed to be contaminated until the samples
collected  during the clean-up verification demonstrate otherwise.

             The statistical methods for these two applications appear to be very similar.
In each case, the average PCB concentration is found and compared with a known value to
make conclusions about the PCB level. Although the procedures differ slightly in the
methods of calculation, the important difference is in the decision-making process indicated
by the italics shown above.  While the procedures discussed in Sections A.2 and A.3 may
appear redundant, purpose of the analysis and the conclusions that would be reached are
different.
                                     A-l

-------
A. 1.2       Sampling Issues

             A number of sampling issues arise in planning monitoring and clean-up
verification programs. These issues are mainly related to the frequency and duration of
visits to the shredder site to collect samples.  This is more of an issue for monitoring
programs, where regular visits are more likely to be required.

             Should samples be collected once a week?  Once a month? Four times a
year? In deciding how often to collect samples, it must be remembered that the material
output from a shredder is the direct product of the input to the shredder. The primary
objective in sampling is to obtain a representative sample of the material that is output
during the normal operation of the shredder. It is possible for the shredder operator to run
only "clean" materials  - for example, materials that have had all electric motors, air
conditioning units, etc., removed - while the samples are being collected.  If this is done,
the samples may not reflect the materials that are normally output at the shredder.

             Ultimately, the question of "how often" is really less important than whether
the samples collected are representative of the normal output of the shredder.  Obviously,
samples taken four times a year may not be representative of the output being produced
during the rest of the year.  However, sampling even once a week may not be sufficient if
the samples selected are not representative.

             When  monitoring programs are in place, sampling usually takes place at
regular intervals, ranging anywhere from four times a year to once a week.  Within this
context, samples may be collected once a visit, once each half-hour for several  hours, or
once each  half-hour for an entire day.  As part of  either a monitoring or a  clean-up
program, we suggest collecting samples of fresh shredder output each half-hour for  a
period of 8 hours, or one work  day. As noted in the Sampling Guidance, the longer the
duration of the  sampling period, the greater the likelihood of obtaining a representative
sample of shredder output. Sampling for an entire working day is likely  to provide good
representation of the shredder's normal operations, at least for that day, and  also will
provide a minimum number of samples for statistical analysis.
                                      A-2

-------
A. 1.3       Hypothesis  Testing

             As we have noted, there are several possible sources of error in assessing
contamination by PCB's or other toxic substances.  For exploratory studies, we used
confidence intervals as a statistical procedure for analyzing data in the presence of error.
For monitoring and clean-up programs, hypothesis tests are the primary analytical tool.

             In hypothesis testing, an assumption is made - for example, that the normal
fluff output of a given shredder site has a PCB concentration that is 50 ppm or less - and
then evaluated in relation to the results of a laboratory test For example, suppose that
laboratory tests indicate that the average concentration in samples collected is 60 ppm. We
know that because of sampling and measurement errors, the actual concentration is not
exactly 60 ppm.  In an hypothesis test, we do a set of calculations which provide a
numerical cut-off against which our sample value is compared. This cut-off depends on the
number of samples analyzed and some  other considerations.  For example, suppose that the
cut-off is 75 ppm.  Comparing the sample estimate of 60 to the  cut-off value of  75, we
would conclude that the laboratory results are within the range of sampling and laboratory
error and that we do not have sufficient evidence to conclude that the output of the shredder
is more than 50.
A.2.         Monitoring

A.2.1        Considerations  in  Monitoring Programs

              As we discussed earlier, the objective of a monitoring program is to make
sure that the output of a shredding operation meets some specified standard.  Frequently
this standard is taken to be 50 ppm, since this is the requirement for TSCA landfills, but
other standards might be considered as well.  In this manual, we will use three possible
standards - 25, 50 and 100 ppm - as illustrations.  Monitoring programs may  also vary
with respect to the frequency and duration of sampling. Samples of output materials may
be taken weekly, monthly, or quarterly, with samples collecting over several  hours or an
entire day.  In most cases, the sample sizes discussed for monitoring are intended for a
single visit.
                                      A-3

-------
             There are two major difficulties in monitoring shredder sites. First, because
of the time delay in having samples analyzed, the actual shredder output that is sampled will
probably be in a landfill by the time the analysis is done to determine whether it is
contaminated  or  not.  Second, the amount of PCB's can be loosely controlled by
processing different materials, since, for example, automobiles appear to be less likely to
produce PCB contaminated output than white goods.  Thus, shredder operators being
monitored by outside agencies could deliberately process materials with low PCB levels
during the monitoring period.  If the materials processed during the monitoring period are
not representative of the normal output of the shredder, then the results of the monitoring
program will not be valid.

              Clearly,  monitoring programs, which depend on statistical principles and
random inspections, cannot detect  all violations.  The best strategy  for keeping
contaminated output out of landfills is to develop monitoring programs that are likely to
detect most violations,  so that appropriate enforcement actions can be taken. One of the
key steps in developing an effective monitoring  program is to collect representative
samples. We suggest three steps. First, regulatory agencies can make unannounced visits
to the  shredder site at  randomly chosen times to  help  assure obtaining representative
samples.  Similarly, shredder operators can collect samples at irregular intervals to help
assure representative sampling.  Second, the longer the duration of the data collection
period, the more likely  that shredder input will be representative; we recommend that the
monitoring period last  8  hours or for the normal duration of operating hours. Finally,
samples of stored fluff  and spillover should be collected, in addition to fresh fluff, since
these materials are likely to reflect the output during normal operation even when fresh fluff
may not.
 A.2.2        Hypothesis Testing for  Monitoring Programs

              When monitoring the output of a shredder site, it is first assumed that the
 output streams are not contaminated. Samples are collected and chemically analyzed at
 intervals to monitor the shredder output, and, based on a statistical analysis of these
 samples, the monitoring agency determines whether this assumption - i.e., that the
 shredder output is in compliance with safety standards - is reasonable. The process used
 to make this determination is called a hypothesis test. The basic steps are simple: the
 average and standard deviation are calculated, a cut-off value is determined and the average

-------
is compared to the cut-off value. If the average is larger than the cut-off value, then the
output is declared in violation, otherwise it is assumed  to be in compliance.  In the
following sections we will discuss how to determine the cut-off value and the sample sizes
necessary for making hypothesis tests.

             As we discussed earlier, the presence of sampling error and analytical error
make it difficult to determine whether shredder output is in compliance with regulations.
The fact that chemically analyzed samples are above the safety standard is not sufficient
evidence that the entire output from which the samples were taken is in violation. A more
careful evaluation must be done to account for sampling and analytical error.  The
procedure that must be followed is illustrated in an example in Worksheet A-l.

             The first step is to  find the average and  standard deviation  using the
procedures given in Worksheet 1 in Section 4. Next, the cut-off value must be determined.
This value can be found by following  the calculations in Worksheet  A-l. Finally, to
evaluate whether or not shredder output violates the relevant standard, simply compare the
average of the analyzed samples to the cut-off value and follow these rules:

             •       If the average is larger than the cut-off, conclude that the output
                     violates the standard
             •       If the average is smaller than the cut-off, assume that the output is in
                     compliance with the standard.
A.2.3        Effects  of Sampling and Analytical Error

              Like all  decisions that are based on statistical methods, hypothesis testing
procedures are subject to error. For example, in a pile of fluff that is relatively free of
PCB's, we may pick a sample simply by chance that has an unusually dense concentration
of PCB's, leading us to conclude that the entire pile of fluff is contaminated.  In this case
we would incorrectly conclude that the output was in violation. On the other hand, in a pile
of fluff that is heavily contaminated, we might happen to pick a sample that has a relatively
low level of PCB's, leading us to incorrectly conclude that the output is in compliance.
These two errors have many names in the statistical literature, but they are most commonly
called "Type 1" and "Type 2" errors, respectively.
                                      A-5

-------
Worksheet A-l:   Hypothesis Testing  for  Monitoring PCB  Levels
Example Data. Assume that 4 composite samples are analyzed and have these PCB
levels:
PCB's (ppm)
70.0
121.0
48.0
51.0
Squared PCB's
4,900.0
14,641.0
2,304.0
2,601.0

Step 1: Find the average and standard deviation. Use the directions in Worksheet
1. For the example data given above:

                 Average of Samples  =  72.50

                 Standard Deviation   =  33.77
Step 2: Determine the Cut-Off Value.  Make the following calculations:

             Short-Cut Method.  In Table A-l, select the appropriate safety standard
             and then find the cut-off which corresponds to the standard deviation and
             sample size that are closest to the yours. For the example data, the standard
             deviation and sample size are 33.77 (which is close to 35) arid 4. Assuming
             the safety standard is 50, the cut-off is 91.1.

       •     Exact Method.  This method is slightly  more complicated.  First, in
             Table A-2, find the r-value for a sample size of 4, which is 2.35. Now
             make the following calculation:

             _,  f^cf\r  i    c   j  j      i   Standard Deviation
             Cut-Off Value = Standard + t-value	         —— .
                                               •v Sample Size

             If the standard is 50 ppm, then


             Cut-Off Value = 50 + 2.35?^ =  89.7.
Step 3: Interpretation. Since the average, 72.5, is smaller than the cut-off, 91.1 (using
Method 1, or 89.7, using Method 2) we do not have sufficient evidence to conclude that
the output exceeds the 50 ppm safety standard.
                                     A-6

-------











*
OO
monitorir
t-off values for
U
IE


•o
^.
c
"5
E
«
03
&
0
O.
U
<«.
o
1
a
Z




V)


\o
••*


»\


^T

•
Standard
Deviation
Safety
Standard
oo o ~ P- CN en m
— r-^ CN o ON vd O
en en ^ in m r- -^

oo cn ON oo oo vo •*!•
cn o vo r- oo o •*
CO ^^ ^^ ^^^ ^O ^^ CO

• CN cn •** \o >— j_
0 wn 0 in g ° °
in
CM
00 0 —
*O ^D ^D

oomON
oo m -^
in \o r^-

•* r- o
CN ~-> —
VO t*- OO

in <— i oo
en -H oo
f"^ ON CD
CN tS -^
ON vO cn
cn o t~-
O m O
CN cn in

r** CN cn in
F— oo O en

00 00 VO Tt
-• — -^ -^ CN CN cn
cs CN -. vo CN en m
ON vo en rf vo ON m
oo in CN cn TJ- vo •— '
»— < CN en rj- in t — CN
S$S£§§g
8
0 4)
•a -C
3 **
— 0?
" 7;
c ^
3s
IE
*§•
o o
^ 4J
s-s
» «
•s-s
c u
0 1
3 3
-— c«
ctf cn
> W
^IM *^
C U
O e«
°'>^
= C 
g»l w
Cd C/3 OO
.2 S :|
as > C
JS '5b^
Q. 0 >»
i*1?
s S M
•O « C
u ^2 o
£.2 S
13 > £
§|S
1) C* cj
•s § i
•s s-g
8>1^
E ^ H
U J=
> M -6
« o 5
^•s-i
•5 «-• c
<*- S 2
- -5 «
A-7

-------
 Table A-2: t-values for hypothesis tests*
Number of
composite
samples
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
30
50
75
100
>100
t-values
6.31
2.90
2.35
2.13
2.02
1.94
1.89
1.86
1.83
1.81
1.80
1.78
1.77
1.76
1.75
1.75
1.74
1.73
1.73
1.73
1.72
1.72
1.71
1.71
1.70
1.68
1.67
1.66
1.65
"The values shown in the table are taken
 from Student's t distribution. This
 distribution is often used as a measure
 of uncertainty due to sampling and
 other sources of error.
                     A-8

-------
             Using the procedure described in Worksheet A-l, you will have a 5%
chance of making a Type 1 error - that is, of concluding that output is in violation when in
fact it is not.  The chance of this type of error is 5% regardless of the sample size.  The
chance of a Type 2 error - the chance of missing violations when they actually exist - does
depend on the sample size.  Because characteristics of fluff vary from place to place, it is
difficult to determine the exact probability of making a Type 2 error, but based on
preliminary studies we have made some approximate calculations that are shown in
Tables A-3 through A-5. These tables give the chance of correctly identifying violations
(i.e., not making a Type 2 error) for a range of sample sizes and hypothetical PCB levels
for safety standards of 25, 50, and 100 ppm.

             For example, in Worksheet A-l, the hypothesis test based on four samples
concluded that the output met the 50 ppm safety standard. In Table A-4 (which covers the
50 ppm standard) we see that with 4 composite samples, assuming each consists of 1
subsample, the chance of detecting a violation of even 125 ppm is only 11%. Thus, we
should not feel too confident that the material is actually in compliance with the standard.
As might  be expected, the larger the  sample size the greater the chance of detecting
violations.  This is true if the  sample size is increased by  analyzing more composite
samples or by compositing more subsamples together. Thus, when 9 composites of one
subsample each are analyzed, the chance of detecting a violation of 125 ppm is 44%,
meaning that 44% of the time a violation of 125 would be detected using procedures like
this, while 56% of the time a PCB level of 125 would remain  undetected. Notice that the
situation improves substantially if 9 composites are used with 4 subsamples each, in which
case the chance of detecting a violation of 125 ppm increases to 88%.
A.3.        Clean-up  Verification

A. 3.1       Considerations in Clean-up Verification

             In exploratory studies, there is little  if any  prior knowledge about
contamination by PCB's  or other substances at a site. In monitoring programs, it is
assumed that shredder output streams are in compliance with PCB standards unless the data
indicate otherwise. However, when a statistical evaluation is undertaken to verify a site
                                     A-9

-------
T3

CQ
O.
a.


(N
 bO


'C
 o
 O
•o
.2
_o
'>
 bo
 c

•3

c;
 D
 o

 CQ


U
 ca
H
      til
       08 V
      'Zg'O

       fc "S N
      _2 O >n


       |sl

      I 8 •
                -  $3
                Say
                £  sl
                H  a  s
                    8—«  NO ON O    Q Tf NO O O
                    •— •<* NO OO    O «S NO OO_ ON   O f> OO ON O    O in ON O O
                 ooooo    ooooo   o'oo'o^    d d d -^ «-^




*

      _   ._    o oo en NO m    o Tj- m ON en   oinmino\    Q o o ON o
      O   5    O O fj «n r-;    O •-; W} t-; ON   O     es
 W    3         o ^ i/-j t/^ \Q    e~> \f) fN| ON >n   O OO Tt" f^* NO    t^ c\| OO ^ C^
 C    ^j   ^O    f^ f^*i ^H f»«^ fO    ^^ r J ^| ^n l/%   c^J ^7  CO IO t^*    f^ ^—M ^^ J^fc ON

_g    <   m    ddddd    ddddd   ddddd    oddo'd

U





                                       *-« ^

           ^    do'odd    ddddd   ooddd    ddddd
                                                                                                  o
                                                                                                  at

                                                                                                  bO
                                                  CS
                                                                                   00
                                                                                                  o
                                                                                                  o
I
CO

o
                                                                                                  03

                                                                                                  U

                                                                                                  |


                                                                                                  CO
ations
                                                                                                  O




                                                                                                  1
                                                      A-10

-------





••rt
1
T3
«
en
E
o.
o,
0
«n
«
JS
£
00
c
"C
0
'c
o
s
c

V3
c
o
•a
CO
13
U}
c
•5
c
IS
U-c
O
u
u
c
(O
-C
U
Tf
<
O
3
cct
H


«
o
»••
5
^3
*>
M
e
"3
V
•**
dj
9*
ro
<_
O
1
03
JS
U



























_o
"3
U
JJ
V
cj

O
U
CQ
u
a.

"3
^^
u
•<









SR
V
^L
E
es
|
C/3
«g
O
Ji
•MB
3



_
2
"5
C™"

V)
r4
i-H
g
**




IT)
00



o
t*>




o
NO




w
*J •*•
il
si
U

— 'O
• •• Q^
S ^
B«
i s
g «


8*
a-y
E|
38

8 — rror- o — OO-HOO or~ooo\o ovor-QQ
— •^•r^oo O«SNOO\O\ omooosO oo\ OOOCT^Q
Ocnmr- o^-«ot"~o\ OtNt^ovON O^O\OvO
ddddd ddddd ddddd dddd—



8V^i l^"\ ff\ ^*1 ^^ ^™i ^^\ T^ ff\ ^^ t*^ fy\ lf\ Vrt ^^ 1*^. t^fc V^^ ^"^
>o |*j ™* ^J v^ W' w* ^j r*i "^^ r^ w^ «« * >fc^ s^ r^ r^ H*' ^^
ocs-^vo O — «n vo oo o — «n oo c\ O(Nt~-o\o
OOOOO OOOOO OOOOO OOOO —



8^-mmvo o«or^^>n Qoo^tt^vo Or^oori-o
O'-<
08
bfi
"B
u:
•3
O
fl\
i
u
c
«4-l

u
u
c
03
U
*
w>
C3
U
3
CA
V3
A
«3
_O
'2
CO

3
o
"«
^
§
|

A-ll

-------




"g
1
I
£
1
o.
8
t— •
.g
'I
00
c
1
c
o
s

c
en
§
•O
PS
"o
">
00
c

u
c
u
«n
t
1>
1
H


i
5
*£
at
'w
9*
•o
^m
0
I

U
























|
es
u
4j
I
ca
u
IX

S
"5









eS
"a.
es

3
^
•5
4,
i
3
z


"«
rj~


0
tf>

g
^


JO
^



0
"^


«q
*H




4>
£ «
II
e g
- s

jg«
X u
1^
li
O es
u

83
11
£ =
Sg

g—'Tfor- o— nr--ON O(sr~-ONON OTJ-ONONO
o'oo'o'o oo'o'o'o o'oo'oo o'o'oo*^


88^J$S 8-^SS 8-SSSS; 8$S£8
OOOOO OOOOO OOOOO OOOO—



OO'—mrJ- OOCSTtvO O'-<'^fvOoo O'-<>nooON
OOOOO OOOOO OOOOO OOOOO


8o2£8 8S2?;?; 8SS^?p 8S?;5-!2
OOOOO OOOOO OOOOO OOOOO





— fN TJ- OO





cs^os^S C4^0N^^J r^^-o^S^ r^osi£8



rsrj-ON^10 TtooS^0 00^^38 ^^^^18
•— < 
-------
clean-up, it must be assumed that the site (or the output stored on a site) is contaminated
until the data demonstrate that an effective clean-up has been carried out. Except for this
important distinction, the procedures for clean-up verification are nearly identical to those
described in Section A.2.
A.3.2        Hypothesis Testing for Clean-up Verification

              The procedure for determining cut-off values in clean-up evaluation is
illustrated in an example in Worksheet A-2. As before, the first step is to find the average
and standard deviation using the procedures given in Worksheet 1. Next, the cut-off value
is determined, either by following the calculations in Worksheet A-2 or from Table A-6.
Finally, to evaluate whether or not the output attains the safety standard, simply compare
the average of the analyzed samples to the cut-off value as follows:

              •     If the average is smaller than the cut-off, conclude that the site has
                    attained the safety standard; and
              •     If the average is larger than the cut-off, assume that the site is still in
                    violation and requires further clean-up.                    ^
A.3.3        Effects of Sampling and Analytical  Error

              Because of sampling and analytical error, these procedures are subject to
Type 1 and Type 2 errors, just like the methods described in Section 2. Here the possible
errors are (1) concluding that the site has attained the safety standard when the actual
concentration of PCB's still exceeds it, and (2) concluding that additional  clean-up is
required when in fact the site has attained the safety standard.

              For the methods described above, the chance of incorrectly concluding that
the site has attained the safety standard is at most 5%.  (It is exactly 5% when the actual
level of PCB's meets the standard and it decreases sharply as the level of PCB's increases.)
Tables A-7 through A—9 show the chance of requiring additional clean-up for standards of
25, 50, and 100 ppm when the concentration of PCB's at the site actually meet the
standard. This probability becomes larger when either the level of PCB's approaches the
standard, or when the sample size is small. It should be noted that because clean-up will
                                      A-13

-------
Worksheet  A-2:   Hypothesis Testing  for  Verifying Clean-Up of PCB's
Example Data. Assume that 4 composite soil samples from the cleaned site are analyzed
and have the following PCB levels:
PCB's (ppm)
11.0
5.0
52.0
10.0
Squared PCB's
121.0
25.0
2,704.0
100.0

Step 1: Find the average and standard deviation.  Use the directions in Worksheet
1.  For the example data given above:

                 Average of Samples   =  19.50

                 Standard Deviation    =  21.83
Step 2: Determine the Cut-Off Value. Make the following calculations:

       •     Short-Cut Method. In Table A-6, select the appropriate standard and
             find the cut-off which corresponds to the standard deviation and sample size
             which  are closest to yours.  Assume  the standard is 50 ppm.  For the
             example data, the standard deviation and sample size are 21.83 (which is
             close to 20) and 4, indicating a cut-off of 26.5.

       •     Exact  Method.  This  method is  slightly more complicated.  First,  in
             Table A-2, find the t-value for a sample size of 4, which is 2.35.  Now
             make the following calculation:

             /-  nccir i     o   j   j      i   Standard Deviation
             Cut-Off Value = Standard - r-value—,          =— .
                                              V Sample Size

             For the example data,

                                     91 o a
             Cut-Off Value =  50-2.35^^ = 24.3.
Step 3: Interpretation.  Since the average, 19.5, is smaller than the cut-off, 26.5 (using
Method 1, or 24.3, using Method 2), we can conclude that the site meets the 50 ppm
standard.
                                     A-14

-------















.0
*•«
o
IS
•c
u
a.
3
i
C
R)
JU
O
Ut
3

£
*O
^
U
NO
I
15
03




N
13
08
8
"a
E
s!
cv
1
a

u
u.
o
_2
E
3

















8



NO





0-V




"*




«M


•o c
l- O
ea ~
c -2
S w
C/3 "O

L.
ea
•a
I
V2

^O ^?N ^^ (^^ ^^ x*» «^»s ^O ^^ ^^ ^^ ^^^ ^^ f^N ^^ 1^^ ^^ ^^ OO ^^* ^^^
^H QS oo vO CO f^,* /^J ^O ro ON fS ON ^^ f^ ^f ^^ c^ ^f >o oo ^
CS f-* ^-< »~* ^"^ ^^ ^ ^ CO m CS C*l ON ^N OO f** ^5 Tf ^^


vo -^ m i— « ^ ^^ vomON^^oocs -^t^^csco^
OOOVO^ospX ' wS'-^VC5o6cntNl* 1 COONOO't-^VO^' f
f\| ^H «— * ^* ^^ ^f *^ CO f^l 0s! ^^ ^^ OO P""* ^D ^i CO




oor*-^^^ oovo^ooo^^ r*-»nO>no^
oo wS CM — ^ —4 ' ' co r** *™* O\ CN .^ ' O ^ ON co oo ^ 1
S iJ i-! ^ co ^JncnX**-.^ ONOOvomco^-



CO «fc .—. CO *O OO "^ NO CO ON
rn Z- • 1 1 1 1 06 NO T!-' 1 1 1 1 ov-i«nov> ooooo«nlQ "^^owiSSS
« <— " r^  NO — ' 
Ji
*^
i
.SJ
^3
.s
X
en
83
•o
^^
*
A-15

-------











•o
i
en
E
o.
o.
in

1
E
«•
z

_
1



o
***




«n



0
«H




v>



"«


 ^J ^^ OO **^ ^^ C^i C1^ CO IO ^? C^ *™^ I"""" OO
»— « O O O O '-^OOOO i— • O O O O »— ' O O O O




g^*SB g?So i 8^0 , i 8SS i i
T-C>C5C5ed -^C3OC5 -«OO — OC5


S — -« Q t- Q r-
m O | | OO||| O | | | | ON | | | |
~ O O « C5 — I O



fS VO
°°. 1 1 1 1 ~ 1 1 1 1 1 1 ! 1 1 1 1 1 1 1
O O



1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1



•— «N Tt OO




^^^VOW, ^^.^vo^ ^^^vOUO ^^^VO^
-^n
S<

« fe
H "
^D 2?
*^I^ ^3
wO C
^ C
ON &<
c3 *O
0 ^
S.52

52 «. -
•So
J^ C
3 -a
"e3 ^
c
U U
en u
|l


A-16

-------









1
-a
3
«3
C
o.
a.
o
IT)
CO
JZ
•^
'>
^
ON
3
1
.H
_o
'•3
"O
C8
00
C
'C
cr
£
u—
O
O
1
u

oo
<£
c 	 .


*
D.
3
•
e
•rj
2!
O
E
ox
c
"C
*3
CP
La
*s
i
CO
JS
^^
^^






















^o
•«*
eg
L.
•*^
e
§
CO
U
0,
*vm
a
ts









"EL
£
S
J5
3
C/3

*s
Number

_
2
"o
H
0
^



o
f*>


o
»s




12






o


n
e E
• ••• JjS
u

£ ^
composi
analyze

8-g
"S ^^
c" S
So
5
gSSSSS 8S5SS SSJPES S8SR2
— cSo'OGJ — OOC5O ^OCSCJC) — O O O O



OOO^tSO Of^OO| | O O O | |
— • C3 C> C5 O —CJCJO —  — ' c> — d



o rs vo r- r-
-*' o" d d d






°°. I I I I I I I I I I I i i i i i i i i
o

— • 
g.2
3 on
i«
CO (J
1 o
- o
3 .2
"3 g
C
4> 4>
c/a o
e §
H o
A-17

-------









tandard
C/9
E
o.
ex
8
S5o^ 8cP?JK^ 8c?Scn2
— oooo ~oooo — o o o o — • o o o o


O'Oootsr- OTfTtm Q-'tr>- Q^'*
^'CJCJOO ^OOO «C>CJ -^OO


S^ "- o r— o r~
C^ ^5 I t ^3 ^5 1 1 t ^|^ 1 1 1 I ^^ lilt
^ C5 O —" C5 ^ C5




o ts oo r» r—
-! 0 C5 C> C5



(N \O
°°. 1 1 1 1 *^. 1 1 1 1 I 1 1 1 1 1 1 1 1 1
0 0


i-* 
*- < r^ «-^ 
CO ^
0 -
U
*o *-

wi 22
C ctt
C *^N
'="=
0*

O —
U en
U C3
§ "°
•C ^
•C «-'
— rt
§ [jj>
s o
^3

C^ O«
*o ^*
°" S
C3 O
E-2
3 en
ig
C« Q*
| o
"2 o
3 .2
O C
"3 Q
u g
?? 0
U g
S S
*
A-18

-------
remove PCB's from the contaminated area, the homogeneity of samples taken after clean-
up may be greater, that is, the standard deviations after clean-up may be smaller than the
standard deviations before clean-up. In this case, the chance of requiring additional clean-
up would be decreased from the values shown in Tables A-7 through A-9.

              Notice that the probability of being required to  do additional clean-up is
related to both the PCB level remaining after clean-up - and thus to the intensity of the
clean-up effort - and to the amount of data collected for verification. For example, suppose
that the standard is 50 ppm. If the clean-up effort is less rigorous, resulting in residual
PCB levels of about 30 ppm, say, then it will require more data to verify the clean-up than
if the clean-up had been more intensive and the residual PCB level were only 20 ppm.  This
point has implications for allocating funds between the clean-up and verification efforts.

              Clean-Up Verification for  Lead and Cadmium.  Because of smaller
sampling and measurement errors, it is easier to detect whether lead and/or cadmium have
been cleaned up with  the amount of data required for detecting clean-up of PCB's.
 A.3.4       What to Do When Clean-Up Is Not  Verified

              When the sample results indicate that the site has not been cleaned up
 thoroughly, it is very important to realize that it is not sufficient to simply clean and re-
 inspect the parts of the site that are in the sample. The reason for this is that the samples
 collected are representative of the entire site; if the collected samples have not been
 thoroughly  cleaned up, then it must be assumed that the rest of the site has not been
 satisfactorily cleaned up, either.

              Therefore, where clean-up does not pass verification, the entire site must be
 cleaned again!  Then, after the site has been cleaned, all the verification steps must be
 repeated using a second, independent collection of samples.
                                      A-19

-------
U.S. Environmental Protection Agency
Region 5, Library (PL-12J)
77 West Jackson Byutevard, Utft
Chicago, IL  60604-3590

-------
50272-101
         DOCUMENTATION
           PAGE
1. REPORT NO.
 EPA 747-R-93-009
3. Recipient's Accession No.
4. Title and Subtitle
  Sampling Guidance for Scrap Metal Shredders
  Field Manual
                                             5. Report Date
                                               August 1993
7. Author**)
  James Bethel, Westat, Inc.
                                             8. Performing Organization Rept No.
9. Performing Organization Name and Address
  Westat, Inc.
  1650 Research Blvd.
  Rockville,MD  20850
                                             10. Profect/Task/Work Unit No.
                                             11. Contract (C) or Grant (G) No.

                                               68-02-4293
12. Sponsoring Organization Name and Address
  U.S. Environmental Protection Agency
  Office of Prevention, Pesticides and Toxic Substances
  Washington, D.C. 20460
                                             13. Type of Report & Period Covered
                                               Technical Report	
                                             14.
15. Supplementary Notes
16. Abstract (Limit: 200 words)

        The purpose of this document  is to provide  basic  instructions for collecting  and statistically
 analyzing samples of materials that are produced as a result of shredding automobiles and other metal
 objects, since the by-products of these recycling operations may contain concentrations of polychlorinated
 biphenyl's (PCBs). Shredders are large machines that convert light metal objects into fist size or smaller
 pieces of scrap metal. PCBs enter the shredder output when materials containing PCB-bearing fluids are
 shredded. Large concentrations of PCBs have been identified in some samples that have been collected at
 some recycling sites.  Thus agencies may wish to collect data at shredder sites in order to study the  situation
 in their locality. The sampling procedures described in this document are intended to produce representative
 samples of fluff that  will give reasonably accurate estimates of the overall concentration of PCBs in the
 material  being sampled.  The document  discusses sample  selection, laboratory testing, and statistical
 procedures for analyzing the data.
17.  Document Analysis   a. Descriptors

  Environmental contaminants, scrap metal recycling


   b. Identifiers/Open-Ended Terms

  PCB, sampling, statistical analysis



   c. COSATI Field/Group
18. Availability Statement
                        19. Security Class (This Report)
                          Unclassified
                                                     20. Security Class (This Page)
                                                       Unclassified
           21. No. of Pages
                                                                                                   58
                                                        22. Price
                                               .tractions on Reverse
                                                         OPTIONAL FORM 272 (4-77)
                                                         (Formerly NTIS-35)
                                                         Department of Commerce

-------