SEPA
               UniiM Sut
-------
                                  cs environments) Protection AQGncy
                                  Washington. DC 20460
                 oswER Directive Initiation Request
                                                          1. Directive Number

                                                            91*33.00-2
                                  2. Originator Information
Name of Contact Person
N/A
Mail Code
N/A
Office
osw
Telephone Code
N/A
        Test Method Equivalency Petitions: - A Guidance Manual
      4. Summary of Directive (include brief statement of purpose)
        The purpose of this manual is to  provide guidance to parties vho wish to submit a
        test method petition.  It explains, in detail, the information a test method equiva-
        lency petition must include.   The term "test method equivalency petition" is used to
        denote all petitions that propose methods to replace or supplement any of the test
        methods described in the EPA manual "Test Methods for Evaluating Solid Waste"(SW-8^6)
      5. Keywords
ey*oras
Guidance/Test Method/Equivalency
      6a. Does This Directive Supersede Previous Directive(s)?
      b. Does It Supplement Previous Directive(s)?
                                            No
                                            No
                                            Yes    What directive (number, title)
                                            Yes    What directive (number, title)
      7. Draft Level
          A - Signed by AA/DAA
                       B - Signed by Office Director
C - For Review & Comment
D - In Development
8. Document to be distributed to States by Headquarters?


Yes
X

No
This Request Meets OSWER Directives System Format Standards.
3. Signature of Lead Office Directives Coordinator
10. Name and Title of Approving Official
N/A
Date
Date
     EPA Form 1315-17 (Rev. 5-87) Previous editions are obsolete.
   OSWER           OSWER               OSWER               O
VE     DIRECTIVE         DIRECTIVE         DIRECTIVE

-------
 TEST METHOD EQUIVALENCY PETITIONS:

         A GUIDANCE MANUAL
       OFFICE OF SOLID WASTE
U.S. ENVIRONMENTAL PROTECTION AGENCY
      WASHINGTON, D.C.  20460

                1986

-------
                                                       OSWER  Policy Directive
                                                              #9433.00-2
                                   CONTENTS
   1.  Introduction 	   1
   2.  Regulatory Requirements  	   2
   3.  The Petition Process and EPA Evaluation  Criteria  	   3
   4.  Statistical Procedures for Evaluating Test Method Equivalency   .   7
   5.  How to Prepare a Petition	11
   6.  Checklist of Petition Requirements	21
References  «	25
Appendices.
   A.  40 CFR 260, Subpart C—Rulemaking Petitions	A-l
   B.  Statistical Procedures to Evaluate Test  Method Equivalency .  .  .  B-l
   C.  Section Ten—Quality Control/Quality Assurance 	  C-l
   D.  Example Test Method Equivalency Petition 	  0-1
                                      ii

-------
                                                        OSWER Policy Directive
                                                               #9433.00-2
                                  SECTION 1

                                INTRODUCTION
     The  EPA  manual, "Test  Methods for  Evaluating  Solid  Waste"  (SW-846)
provides an up-to-date, unified source of information and methods on sampling,
analysis, and quality assurance for compliance  with the Resource Conservation
and  Recovery  Act (RCRA)  regulations.  It is  a collection  of sampling  and
analysis  methods and  procedures  approved  for evaluating  the properties of
wastes and environmental media, and for monitoring  the efficacy of treatment.
SW-846 is incorporated by reference in the RCRA regulations.

     The RCRA regulations (40 CFR 260.20) establish procedures by which
persons may petition EPA to approve the use of alternative or equivalent
testing procedures when conducting testing under RCRA.  Throughout this
guidance manual, the term "test method equivalency petition" is used to denote
all petitions that propose methods to replace or supplement any of the test
methods described in SW-846.

     Any person or organization may submit a petition to request  approval for
a  proposed  test  method.   Petitioning  entails  the  submission  to EPA  of
comprehensive information describing  the method, data from tests  designed to
evaluate equivalency with existing methods, and a statistical analysis  of the
equivalency test data.  EPA  evaluates  test method equivalency  petitions for
completeness, applicability,  and technical quality.  In addition,  the Agency
reserves  the right  to  conduct an  independent statistical  analysis  of the
equivalency  data. Methods  that  pass  the  evaluation  and  do  not  receive
substantial negative public comment will be approved, and may be  published in
SW-846 for public use (e.g., 1f 1t can be applied to multiple sites).

     The purpose of this manual 1s to provide guidance to parties who  wish to
submit a test method petition.  It explains, in detail, the information a test
method equivalency petition must include.   The following  chapters and appen-
dices provide:

         An explanation  of the  RCRA regulations that  require a  test method
         equivalency petition

         A discussion  of the  process and procedures  by which  a petition is
         submitted to and reviewed by EPA

         A  description  of  basic  statistical  procedures  to  be  used  for
         evaluating test method equivalency

         A description of how to prepare a petition, with a suggested petition
         format

         A checklist to help ensure the completeness of the petition

         A discussion of proper experimental designs and  statistical analyses
         for more complex petitions

         An example of a test method equivalency petition.

     By following the guidance in this manual, a petitioner should be  able to
develop a  petition that satisfies  the regulations  and that  EPA   can review
expedltiously.

-------
                                                         OSWER Policy Directive
                                                               #9433.00-2

                                  SECTION 2

                           REGULATORY REQUIREMENTS


     Title 40 of the Code of  Federal Regulations (CFR), Part 260,  Subpart C,
defines the procedures and~Tnformation required for rulemaking petitions.   In
particular, Sections 260.20  and 260.21  specify information that  is required
for test method equivalency petitions.*

     Section 260.20 of the regulation contains information that applies to all
types of petitions including test method equivalency petitions.   It describes
general information the petition must include and outlines  the decisionmaking
procedures  EPA follows  to approve or  deny petitions.  Section 260.21 estab-
lishes  specific  information  needs for  test method  equivalency  petitions;
i.e., what data and information EPA must have to  determine if a proposed test
method is equal or superior to a corresponding SW-846 test method.

     By providing  the  necessary information,  the petitioner  can  meet  the
objectives of the regulations, which are:

          To ensure that  appropriate, accurate, and precise test  methods are
          used

          To ensure  the  comparability  of  all  hazardous  waste  test  data
          gathered in support of EPA's regulatory program.
Appendix A contains the text of 40 CFR 260.20 and 260.21,

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
                                  SECTION 3

               THE PETITION PROCESS AND EPA EVALUATION CRITERIA
3.1  DESCRIPTION OF THE PETITIONING PROCESS

     This section will acquaint  the petitioner  with the complete  process by
the which a test method equivalency  petition is submitted to and  reviewed by
EPA.   The  discussion  is organized  around a  flowchart  (Figure  3-1)  that
illustrates the  sequential steps in the process.   The time  periods shown in
this figure  are approximate;  they  provide  only  an estimate  of  the  time
required for each step in the petition process.


2.1.1  Pre-petition Contact with the EPA

     Prior to submitting a petition, the petitioner is advised to  contact the
EPA  Methods Section  staff within the  Office of  Solid Waste (OSW)  at (202)
382-4761.  Such contact can be used to identify whether the proposed method is
actually different from an existing test method.   For example, a modification
of measurement  technique or a  modification to  the equipment may  already be
within the scope of a current method and may not constitute a  method variance
requiring a petition.   Alternatively, the  use of different measurement tech-
niques or new types of equipment frequently requires petitioning for approval.
Familiarity with the SW-846 document and a phone call or correspondence to the
EPA  Methods Section staff can normally  resolve the question of how  much the
proposed method varies from the approved method.

     Unless there  is a  great  deal of  existing data  that  demonstrate  the
equivalency or capabilities of a  proposed test method, a specific  program of
equivalency testing must  be undertaken by the petitioner.   During this stage
of pre-petitlon contact, the  petitioner is  urged to develop  an experimental
design that will satisfy the  data quality objectives (DQOs) specified  by the
EPA Program Office.  (See Section 4.)


3.1.2  Submission of the Petition

     Any applicant  may  request  approval  for an  alternative  test  method.
Requests should be submitted in triplicate  and by certified mail to:   Chief,
Methods Section, Office of Solid Waste (WH-562B), U.S. EPA, 401 M-  Street, SW,
Washington, DC 20460.   The EPA  Methods Section  Chief will then  acknowledge
receipt of the petition by sending a letter to the petitioner.


3.1.3  EPA Evaluation and Preliminary Recommendations

     The initial technical review of  petitions is  conducted by EPA  staff in
OSW  and  several EPA  laboratories.   This  review  does  not  include actual
laboratory testing but is limited to a critical analysis of the  reported test
results and the associated statistical analyses. If sufficient  information or
data are not available for an  acceptable review, EPA returns the  petition to
the applicant with specific requests for additional information.   Assuming no

-------
            PETITION SUBMISSION

                   1 week
PETITION EVALUATION

      12 weeks
ADMINISTRATIVE REVIEW

       12 weeks
TENTATIVE DECISION
& PUBLIC COMMENT

    12-24 weeks
FINAL DECISION

Initial
contact
with EPA

t

»u

Establish
DQOs and
experimental
design


Submit
to EPA

»_


EPA
issues
letter of
receipt





OSW staff
& EPA labs
evaluate
petition



Request
additional
information

•fc


EPA staff submits
recommended
tentative decision
•




EPA
» Arlmin - IBM.
Aumin.
review
A
I
1
V
Request
additional
nformation


_. Tentative
decision
i
P


Notify
petitioner
of tentative
decision


Tentative
decision
published
in Federal
Register.
Request
public
comment
1
1
L 	 ]

Public
ment
[Conduct I
J informal 1
*1 public I"
(hearing \\

f
\
1
1
1
_J



Final
decision
i
r
»^

Notify
petitioner
of final
decision




Final
decision
published
in Federal
Register
\
t
Issue
update to
include
approved
methods in
SW-846
NOTE: Dashed lines indicate optional procedures.
                                                                                                                                                o
                                                                                                                                                I/)
                                             Figure 3-1. Petition procedures flowchart.
                                                                                                                                             OJ —•
                                                                                                                                             CJ O
                                                                                                                                             • *<
                                                                                                                                             CD
                                                                                                                                             O O
                                                                                                                                              I  -••
                                                                                                                                             ro -j
                                                                                                                                                CD
                                                                                                                                                n

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2

additional testing and data submission are required of the applicant,  the EPA
technical evaluation should take about 12 weeks.

     Following technical review of the petition, the EPA Methods Section Chief
recommends  approval or denial  of the  proposed method through  a memorandum.
This memorandum  contains  a complete  review of  the  decisionmaking  process
including a technical rationale for the proposed action.


3.1.4  EPA Administrative Review

     The  recommendations of the OSW  staff are reviewed  by the methods work-
group and then circulated to other EPA Program Offices (e.g., Office of Water,
Office of Toxic  Substances) for  compliance with  national  policies.  If more
information  is needed, the Agency may request additional information from the
petitioner.  The  Agency then  makes a tentative decision to grant or deny the
petition, notifies the  petitioner accordingly, and  prepares a  notice of the
tentative decision for publication in the Federal Register.


3.1.5  Notice of Tentative Decision and Public Comment

     The notice of the tentative decision published in the Federal Register is
accompanied by  a request  for public comment.   Comments may  be submitted in
writing, or commenters  may request  a public hearing.   Regulations governing
EPA  decisionmaking on petitions  state that  "the Administrator  may,  at his
discretion, hold an informal public  hearing to consider oral comments  on the
tentative  decision" (40  CFR Part  260.20(d)).   Although an  informal public
hearing must satisfy all fairness criteria in public participation, it differs
from a formal public hearing in the following respects:

          Legal  standing:   a judge  does not  preside at an  informal public
          hearing.

          Scheduling:   informal public hearings usually do not schedule  wit-
          nesses in advance.

          Transcript:   EPA  does  not prepare  a complete  written  or  taped
          transcript of informal public hearings.

The petitioner is expected to participate in any public hearing  regarding the
proposed test method.


3.1.6  Publication of Final Decision and Incorporation into SW-846

     After evaluating all  public comments and EPA staff  recommendations, the
Agency makes  a final  decision  and publishes  (in the  Federal  Register)  a
regulatory amendment  or denial of the petition.   If the proposed test method
is approved, and if it is  likely to have widespread applications,  the method
may be incorporated in "Test  Methods for  Evaluating Solid Waste:   Physical/
Chemical Methods," SW-846.   Updates (procedures  for newly  approved methods)
and revisions (changes in existing methods) to SW-846 are issued  as necessary
by EPA to all persons or organizations holding subscriptions to SW-846 through
the Government Printing Office.

-------
                                                        OSWER Policy nirective
                                                              #9433.00-2

3.2  DESCRIPTION OF ERA'S TECHNICAL EVALUATION CRITERIA

     When a petition is submitted  to EPA,  the Agency must  determine whether
the proposed test method is at least equivalent to an existing method  in that
it yields comparable results.   In this case, equivalency is defined as "equal
to or superior to the corresponding SW-846 method in terms of its sensitivity,
accuracy, and precision  (i.e., reproducibility)" (40 CFR 260.21).   To deter-
mine equivalency, EPA  evaluates test  method equivalency petitions  using the
following three criteria:   (1) applicability of the test method, (2) adequacy
of documentation, and  (3) statistical comparability of proposed  and approved
test methods.

     Test method applicability  refers to the appropriateness of  the proposed
for test method for the type of waste the method is designed to detect.  Thus,
example, a  method designed to  detect a  constituent normally found  in waste
sludge  should be applicable  to various  types of sludge.   In addition,  the
petition should be clear regarding  how broad an approval is  desired, because
the method can only be approved for the universe of wastes to which  test data
 fl"e applicable or validly extrapolatable. Adequacy of documentation  refers to
the completeness with which  the applicant  has fulfilled the  information and
lata requirements outlined in Sections 5 and 6 of this guidance manual.

     The  final  criterion, statistical  comparability, is  used  to  evaluate
equivalency test  data in terms  of specific  data quality objectives  such as
precision, bias,  and method  sensitivity.   Specific  numerical  criteria for
determining statistical equivalency will be developed in coordination with the
 JPA  Program Office, with  each proposed  method considered on  a case-by-case
basis.   Section  4  provides more  detailed information  on  the  statistical
analysis aspects of equivalency petitioning and guides the  petitioner through
the development of an appropriate, statistically sound testing program.

-------
                                                        OSWER  Policy Directive
                                                              #9433.00-2

                                  SECTION 4

        STATISTICAL PROCEDURES FOR EVALUATING TEST METHOD  EQUIVALENCY


4.1  INTRODUCTION

     This section and Appendix B address statistical  issues  in the  petitioning
process.

     The Agency recognizes  that no single specification of  data requirements
(type and amount) for demonstrating test method equivalency  is appropriate  for
all test methods.   As a result, the petitioner has various  options regarding
information submitted in  support of a proposed method.    For  instance, a test
method could, under certain conditions,  be  approved on  the  basis   of  existing
data  on the  proposed method  only.   Thus, the  time and  cost of generating
special  test data could be eliminated.   Under other conditions, a controlled
experiment involving both the proposed  method and the approved method may be
necessary to assess the adequacy of a particular method.


4.2  DATA QUALITY OBJECTIVES

     In a  recently prepared  guidelines document  for   validating  measurement
methods (2), EPA states that "Specifications of the data quality needed  for a
particular data collection activity are called data quality  objectives (DQOs).
DQOs are definitive, quantitative or qualitative statements  developed   by data
users  about the  accuracy, precision,  representativeness,  comparability  and
completeness of measurement data needed to support their  specific  decisions."
(This  document (2)  and  others  (3,4)  provide guidance  in   validation   and
equivalency.)

     It is the responsibility of  the EPA/OSW program office to specify DQOs.
These objectives are not fixed numbers but  may vary depending on the   type of
measurement under consideration.  Once DQOs  are specified, the petitioner must
then  select (or work with the  OSW staff  to develop) an  experimental design
that will  enable the petitioner  to collect  adequate test  data  to determine
whether the proposed method complies with the DQOs.

     While  it is  the  Agency's intent  to  allow  appropriate flexibility in
determining  DQOs,  the  Agency  expects  uniformity in  certain   aspects   of
formulating the problem.   Key data  quality objectives  will involve the  bias
and precision  of the proposed test methods.   However,  frequently, DQOs will
also  include detection  limits  or limits  of reliable  measurement   and   the
significance level and power of hypothesis tests about bias  and precision.

     A  literal Interpretation of  40 CFR  260.21 is misleading;  the  proposed
test method does not always have to be equal or superior (i.e., more accurate)
to an  EPA-approved test method - regardless of  application - in order to be
approved.   Existing approved  test methods   are sometimes more accurate than
needed for specific Agency uses.   In these cases, the DQOs  should  reflect  the
Agency's measurement needs  rather than the capabilities of  existing  approved
methods.  Thus, a  proposed  test method  does not  necessarily have  to be at

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2

least as good as the approved method to be useful  (and acceptable)  for certain^
testing purposes.

     In  developing acceptance criteria  based upon  DQOs,  it is  important  to
keep in mind that, for example,  precision can be  improved  by  taking repeated
measurements  of a quantity and estimating  it as  the average of  the measure-
ments.   For instance, the average of four measurements is  twice as precise  as
a single measurement.  This is important to remember if the proposed method  is
somewhat less precise (based on  single estimates)  than EPA criteria  or than
offered  by  the  existing  approved  method.    Were  the  method  simple and
inexpensive  enough  that  multiple  independent measurements  could  be  made
without increasing  the cost or  time of  the  overall  analyses,  the resulting
precision from average values might  be quite  acceptable for the  desired use.
This principle  must  be applied  cautiously since  repeat   measurements  must
indeed be  truly  independent and  reflect appropriate  sources  of variation.
Replicates often fail to meet these requirements.

     In general, data  quality objectives  (or acceptance criteria  based upon
them)  can be classified  as either  absolute  or  comparative.   Absolute DQOs
reflect the Agency's intrinsic needs in the given  situation and do not involve
a direct comparison of the proposed method  to some approved test method.   On
the other hand, comparative DQOs typically state that the proposed test method
must be equivalent (or superior) to an approved test method.

     The statistical  methods  used to  assess compliance  with  absolute and
comparative DQOs differ.  In the absolute case, the precision (a2new) and
(bnew) of the proposed test  method are compared witn numeric  criteria
tor  precision and  bias (oz0  and  b0) established  at the  beginning  of the
petitioning  process.   This comparison  is carried  out by  testing  the null
hypothesis (Hj) that the precision offered by  the  proposed  method is  no worse
than that established as a  program requirement and the null  hypothesis (H2)
that the bias of the new method is  no greater than what has  been established
as a progran requirement:


                               HI:  a2   < a2
                                     new      o
                               H2=  |!>newl 1 bo


In the comparative case, the new test method's precision and bias are compared
with  the  approved method's  precision  and  bias  (o20]d   and  b0]d).   This
comparison uses the null hypothesis  (HS) that there is no   difference between
the precision of the proposed  method and that demonstrated with  the approved
method  it  is  proposed to  replace.   This  comparison also  uses  the null
hypothesis (fy) that there is  no difference  between the bias  encountered  in
the use of  the proposed method than that  is  expected in use of  the approved
method:

                              H3:  a2   = a2
                                    new    old
                              H4'  bnew = bold

-------
                                                        OSWF.R Policy Directive
                                                              #9433.00-2

     Note that  the way the  null hypotheses  above are  formulated,  the null
hypotheses  must be  rejected  to disapprove  the new  method.   EPA  is  more
concerned about the errors  of falsely accepting these hypotheses  (i.e., Type
II  error) than  about the  errors  of falsely  rejecting them  (i.e.,  Type I
error).  In other words, the Agency is primarily concerned that the supporting
test  data  provide  sufficient  assurance  that  each  of  the  data  quality
requirements are  met or exceeded.   This assurance  is related to the "power"
(power is defined as 1 minus the Type II error) of the resulting test  data to
detect with known probability  the likelihood  that data quality  would suffer
with  the proposed  method.   For  instance,  F.PA  may  desire an  80  percent
probability of rejecting ti^ if the proposed  (i.e., new) test method has  a 50
percent  (or  worse)  loss in  precision.   In  this  example, the  50 percent
precision loss  is the change in data  quality to  be concerned about  and the
power (likelihood of correctly identifying it) is 80 percent.

     The  petitioner should  also  note that  if the  comparative  approach is
elected, then the approved  test method  provides the standard.   The proposed
method fails the equivalency test if  the difference between its bias  and the
bias of the approved method is found to be significant.   Where it is believed
that the proposed method is more  accurate than the approved method,  or where
resources  necessary to  apply  the  approved  method are  limited,  then  the
absolute approach may be selected.

     As noted earlier, both the significance level and the power of hypothesis
tests about bias and precison  are themselves data quality objectives.   These
measures  describe  the quality  of  the  data  submitted in  support  of  the
petition.   Studies should  have adequate power to detect  negative changes in
data quality that are of concern to EPA.   Test power requirements for various
data quality parameters should  be determined on a case-by-case  basis through
discussions with the OSW program staff.

     It  may not be necessary to  design a  special experiment to  support the
petition.  The petitioner may have extensive experience with the proposed test
method  for  relevant  waste  samples.   For  example,  data  from  a properly
conducted quality control (QC) program may represent an acceptable alternative
to  a special experiment.   Specifically, spiked  waste sample data spanning a
sufficiently  long time period  may provide  the information needed  to assess
method  precision and bias.   The need  for comparability  of the petitioner's
samples with those analyzed under  SW-846 is  a critical issue  in determining
whether the petition may be based solely on QC data.

     Both the  absolute  and  comparative  cases  can  be  further  classified
according to  the need  for  multiple  waste  sites in  the  equivalency  test
program.  The extent to which the  proposed test  method is used  (e.g., local
versus regional application)  is a  determining factor for  requiring multiple
sites.   The petitioner should determine, in coordination with  the OSW staff,
which of the following categories best describes the nature of the equivalency
study method:

     a.  Absolute - single site/local significance

     b.  Comparative - single site/local significance

     c.  Absolute - multiple sites/regional or national significance

     d.  Comparative - multiple sites/regional or national significance.

-------
                                                        OSMER Policy Directive
                                                              #9433.00-2

     The first step in choosing an experimental  design is to define the target
population to  which  statistical inferences  are desired.  The  design should
incorporate relevant sources of variation and exclude irrelevant sources.   One
consequence of this basic principle is that if validation or  equivalency  data
involve only one laboratory (or only  one site), then approval  may  be limited
to use  of the  method at  that  laboratory (or  at that  site).   Equivalency
studies for  multilaboratory or  multisite applications  will  usually require
data from multiple laboratories or multiple sites.

     Appendix B, Section  1 provides  guidance for  developing  and conducting
test programs and  associated statistical  analyses for  petitioners  who  only
plan to use categories (a) and (b) above.

     Appendix B,  Section  2  discusses  issues  related  to  petitions where
multiple  sites  and multiple  laboratories  will  use  a  test  method.   The
development and analysis of  test programs  for these cases  are significantly
more complex than for the single site  case; for this reason, in  multisite or
multilaboratory categories, it is strongly recommended that the services  of a
professional statistician  be  available  to  ensure  an  adequate  and cost-
effective test program.

     Specific experimental designs are proposed (in Appendix R, Sections 1 and
2) for each of the four cases.   These designs have been evaluated by EPA, and
are considered minimal.   Deviations from  these designs would normally be  in
the direction of more data,  e.g., more sites, more samples/days  of analysis,
and/or  more  concentrations. While  replication is  expected  in  equivalency
testing, the petitioner is not encouraged to run more than two replicates.
                                      10

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
                                   SECTION 5

                          HOW TO PREPARE A PETITION
5.1  PURPOSE OF A PETITION

     The purpose of a test  method equivalency  petition is to  demonstrate to
EPA  that a  proposed test  method  (whether it  be a  sampling  method, waste
characterization  method, or  any other  method  believed comparable  to those
found in SW-846) is equal or superior to the corresponding SW-846 method.   To
accomplish this, the  petitioner must provide EPA with  sufficient information
to determine equivalency by following  the approach outlined in 40  CFR 260.20
and 260.21.   To identify  what information  must be provided,  the petitioner
should ask himself the following questions:

          What is the objective of the  petition, i.e., can I provide  a clear
          statement of scope and applicability?

          What  information must be  known about  the proposed test  method to
          understand how to use it?

          What test method  performance information can be provided  to demon-
          strate that it is equal to or better than an SW-846 method?


5.2  CONTENT OF A PETITION

     Although  the regulations  do  not  require  a specific  format  for  the
petition, for ease of review, the necessary information should be organized in
a logical  manner.   Figure  5-1 is a  suggested format  for a  petition.   By
following the suggested format and addressing the items in  sufficient detail,
the petitioner can contribute to a more expeditious and uniform  evaluation of
the test method petition.


5.2.1  Name and Address of Petitioner

     The petition should begin with the following administrative information:

     •    Name of the firm submitting test method petition.

          Address (i.e., street, city, State, and zip code).

          Names, titles,  and  telephone numbers  of persons  to  contact  for
          additional petition information.
                                      11

-------
                                                  OSWER Pol-icy Direct ive
                                                        #9433.00-2
                TEST METHOD EQUIVALENCY PETITION
1. Name and Address of the Petitioner
2. Certification of Accuracy and Responsibility

3. Description of Proposed Action
     Description of Test Method
     Description of Applicable Samples or Matrices
     Assessment of Limiting or Interfering Factors
     Test Method Quality Control Procedures
4. Statement of Need and Justification for the Proposed Test Method
     Need
     Justification
             Figure 5-1.  Suggested petition format.
                                12

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2

5.2.2  Certification of Accuracy and Responsibility

     Each petition must include the following certification statement:

     I  certify under penalty of law  that I  have personally examined  and am
     familiar with the  information submitted  in this  demonstration  and all
     attached  documents and that  based on  my inquiry  of  those individuals
     immediately responsible for obtaining the information, I believe that the
     submitted information  is true, accurate, and complete.   I am aware that
     there are significant penalties for submitting false information, includ-
     ing the possibility of fine and imprisonment.

                                              Signed,
                                              Title

                                              Date
5.2.3  Description of Proposed Action

     This  section  of  a petition  contains descriptive information regarding
(1) the test method itself, (2) the applicability of the test method,  (3) any
factors  that limit or  interfere  with  the  performance  of the  method, and
(4) the  necessary  quality  control  procedures  that  should  accompany  the
method's use.   In general, this section should contain all of the information
necessary for EPA to understand the proposed method.

     5.2.3.1  Test Method Description

     The description  of the proposed  test method  should be  presented  in a
fashion that will help EPA evaluate the petition effectively and  allow direct
comparison with the current test method description.   Therefore, descriptions
of a  proposed test  method  should  follow  the established  format  for  the
comparable  SW-846 test method.   The following  sections reflect  the current
format  and information  needs  for analytical  test methods  and  RCRA  waste
characteristic  test methods  found in SW-846.   In general,  the test  method
description should contain all of the information.described in  these sections
that is needed for any qualified person to perform the test method.

     Scope and Application,   This  portion  of  the  test method  description
identifies what  themethod  accomplishes;  i.e., it  identifies  the  sample
property  or  constituent(s)  that  the  method  measures.   It  also  briefly
describes the types of samples that one can analyze with  acceptable accuracy.
The petitioner should mention (1) steps taken to prepare the sample before the
method  is applied  and  (2) procedural  changes required  for  special  waste
samples. Finally,  the petitioner  should identify any  special qualifications
required for personnel that will perform the test method.
                                      13

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2

     Summary.   Petitioners should  provide a one-or two-paragraph summary  of
the test method description.  If detection limit(s) are applicable to the test
method, they should be provided  for the  properties or constituents  that the
method measures.

     Interferences.   Caused  by  unexpected  chemical   reactions or  physical
properties such as heat, light, or  the physical  state  of a  sample, interfer-
ences can lead one to believe  that sample concentrations are higher  or lower
than  actual.   Therefore,  it  is  important  to  identify  and  correct  all
interferences that may occur when the proposed test method is conducted.   For
example,  in  atomic  absorption  spectroscopy, light  scattering  and  sample
viscosity may cause interferences.  With gas chromatography,  sample processing
hardware, glassware,  low-purity solvents,  or reagents could  cause artifacts
and/or elevated baselines in chromatographs.   For ignitability  test methods,
ambient pressure or drafts could significantly affect flash point values.

     Materials and Apparatus.   The petitioner must provide a detailed list of
materials  required to conduct  the test  method  procedures  accurately.   The
apparatus must be described thoroughly, with product names and  model numbers,
equipment size,  construction materials,  and other information  important for
the  success of  the method identified.   When an  apparatus  must be specially
built  for  the  test,  the  petitioner  must  include   detailed  construction
instructions.

     Reagents.  A "reagent" is "a substance, chemical,  or solution used in the
laboratory to detect,  measure, or otherwise examine other  substances, chemi-
cals or solutions"(5).   All reagents  used in any aspect of   the test  method
(including sample  preservation  and preparation)  must be  identified  in the
description.  Important details about reagents include:

          Proper grade(s) to use such as American Chemical Society (ACS)

          Concentrations of reagents and how to prepare them

          Maximum shelf life allowed for accurate analysis.

     Sample Collection, Preservation, and Handling.   Petitioners   for   test
method equivalency must acknowledge  that they use a sampling  plan addressing
considerations in  Section  1 of  SW-846. (This  section  concerns statistical
techniques for obtaining accurate and precise samples.) This  ensures  that the
sampling program does not interfere with the test method's performance.

     Procedures.   The test procedures are the heart of  every test method and
must be described  in a succinct, stepwise fashion.   Excluding petitions  for
new sampling methods,  descriptions must  include all procedures  after sample
collection.   If the method requires sample preparation steps,  the petitioner
must Include these in this section.   The following topics are considered test
method procedures:

        Sample preparation, e.g., extracting a constituent from a solid sample
        into a liquid media.
                                      14

-------
                                                        OSWER Policy nirective
                                                              #9433.00-2

        Test method apparatus.

        Entering the sample into the testing device.

        Operating conditions for test method apparatus.

        Calibration procedures

        -  number  of standard  calibration  concentration  levels  to use per
           parameter
        -  preparation method for blanks and calibration samples
        -  frequency for preparing fresh blanks and calibration samples
        -  tabulation of results from calibration standards
        -  frequency of verifying instrument calibration
        -  acceptable qualitative minimum recovery yields
           amount of variation  allowed from the  calibration  standard before
           recaiibration is required.

        Analytical test procedures (where applicable)

        -  retention times
        -  sensitivities
        -  acceptable recovery yields
        -  method  for handling unexpected events such as peak areas  that are
           greater than the linear range of the analytical system
        -  method for verifying the absence of interference
        -  method for coping with interference.

        Calculating results

        -  units of measure
        -  complete equations
        -  data validation, if applicable.

        Proper  management of  remaining  sample  and  contaminated laboratory
        equipment and materials.

     Quality Control (QC) Procedures.   The information  required in this sec-
tion Tsdescribedfn"TestMethod  Quality  Control  Procedures"  (Section
5.2.3.4).

     References.   Each  test  method  description  shall include  a  list  of
literature or other references (1) pertinent to the development and content of
the method, and  (2) demonstrating the procedure's applicability.   A complete
reference list  allows the  person  performing the  test method  to  refer  to
original documents if questions arise.


     5.2.3.2  Description of Applicable Samples or Matrices

     A test method equivalency  petition should  describe each sample  type or
matrix  for which  the  proposed method  is suitable.  Information  about  the
                                      15

-------
                                                        OSWFR Policy Directive
                                                              #9433.00-2

sample's chemical and  physical character provides both the  person performing
the  test  method  and  petition  reviewer  insight  into  the  test  method's
applicability.

     Where the method is applicable to a  general type of matrix such  as oily
sludge,  information about the  percent water,  solids contents, or  any other
physical property that  affects the test method's performance  facilitates the
petition evaluation.

     If the test method is  designed for a specific waste  stream(s), informa-
tion  about the  variability of  the  waste  stream  as it  might  offset  the
applicability of the method is required.

     In general, the  petitioner should  provide enough information  about the
sample(s) or matrix for the  petition reviewer  to evaluate the  test method's
applicability.


     5.2.3.3  Assessment of.Limiting or Interfering Factors

     This section  of  the  petition  includes more  than  the  "interference"
statement in the method description.  It is an assessment of how interferences
and limitations affect the performance of the proposed method.

     The petitioner must (1) identify every limitation and  interference known
to occur with the test method and  (2) describe the effect each  limitation or
interference has on the proposed  test method's results and present  test data
that  demonstrate such  impacts.   Where these  limitations and  interferences
depend on specific samples or matrices, a description for each interference or
limitation should be provided.   Where measures of waste constituents or waste
characteristics typically  vary, the description must include  demonstrated or
predicted effects which the limitations or interferences have on  the accuracy
of such variations.   The petitioner  should describe  any corrective measures
the analyst can take to prevent or minimize the limitation or interference.

     A common  test  method  limitation  is  the  "quantitation  limit"  which
represents  the minimum waste  constituent concentration  or value of  a waste
property that  a  test method  can measure  accurately.   When preparing  this
portion of the petition, the petitioner should identify the  proposed method's
known quantitation limit  (where applicable) and discuss its  implications for
the types of samples intended for testing (i.e., how often the method  will be
able to  detect the  constituent  or property  of interest  given  a  sample's
typical characteristics and composition).   Other limitations encountered with
test methods include the type of matrices the method apparatus can  accept and
the  species of  a constituent that  are detectable,  such as  the  variety of
halogenated compounds.

     "Light scattering" when particulate matter is present in the sample is an
examp <£ of  test method interference.   In atomic  absorption spectroscopy, it
alters the sample's true absorption of light for a specific metal.  Therefore,
a petitioner should describe how significantly this interference  changes true
absorption.   Other examples  of interference are high sample  viscosity, test
method apparatus construction materials, and low-purity reagents.
                                      16

-------
                                                        OSWER Policy nirective
                                                              #9433.00-2
      5.2.3.4  Test Method QC Procedures
     Quality control procedures  "define the frequency and methods  of checks,
audits, and  reviews  necessary to  identify problems  and  dictate corrective
action,  thus verifying  product quality" (SW-846).    Quality control   ensures
that test method results  are representative  of the sample's  true character.
Procedures  such  as  good  recordkeeping,  equipment  maintenance,  personnel
training, and test method practices contribute to a  successful QC program.

     Each petition must describe  clearly the  QC procedures followed   for  the
proposed  test  method because  they  will  become  part of  the  test  method
description if approved for SW-846.   The petitioner  also should describe  how
the  selected QC  procedures  enhance the  method's  overall  performance.   QC
procedures that should be addressed in the petition  include:

          Training and  evaluating personnel  that operate test  method equip-
          ment.

     .    Maintaining, inspecting, and servicing test method equipment.

          Test method procedures (where applicable)

           -  preparing  calibration curves for a blank and a specific  number
              of standard solutions

           -  diluting  samples if the  constituent   concentration  exceeds or
              falls  on the  plateau of the  calibration curve for the highest
              concentration standard solution

           -  running a given number of blanks for each sample batch to assess
              contamination

           -  checking  standard  solutions  and a  duplicate  sample  after a
              certain increment of samples is tested

           -  incorporating  spiked or  standard  reference  samples  into  the
              regimen periodically to ensure that test procedures are followed
              and test method equipment is operated  correctly.

          Recordkeeping procedures

           -  labeling  laboratory  samples  and conducting a sample chain-of-
              custody program within the laboratory

           -  using  statistical  procedures to check  accuracy, precision  and
              bias

           -  continuously  reviewing analytical  results to identify problems

           -  documenting training and equipment performance

           -  maintaining all records properly.
                                      17

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2

The statistical check for accuracy, precision, and bias listed  above requires
additional discussion to  understand fully the information requirements  for a
petition.

      "Accuracy" measures how well a specific sample data point agrees with the
true  value of interest, and it is assessed based on test result  precision and
bias.    "Precision"  represents  how well  repeated measurements  of  the same
constituent or property agree with one  another, and it is assessed  using the
standard deviation of  a series  of controlled  measurements.   "Bias" is  the
constant difference between the average  of data  points produced by  the test
method and  the true value  of interest;  thus, bias  measures  any systematic
error in  a test method and is  calculated as the difference between  the true
constituent value  and  the average  of laboratory  runs.   A petition  should
address these three statistical concepts  along with  steps to determine  if a
database is complete,  representative, and  comparable to other  databases for
the same test method.

     Section  10 of  SW-846, "Quality Control/Quality  Assurance," illustrates
how  EPA defines  and uses  QC  procedures to  ensure that  test  methods  are
performed properly.  Section 10 includes guidance on:

          Using QA/QC procedures to ensure achievement of program goals.

          Developing  a sampling program  that can  be measured  for  how well
          samples represent the true value of interest.

          Developing  a test  program  that  provides  data at  the  level  of
          accuracy and precision that will  be required  by users of  the data
          for decisionmaking under RCRA.

          Assessing the quality of the data  that result from use of  the test
          method, e.g., accuracy and precision.

Appendix C contains the full text of SW-846, Second edition, Section 10.   The
petitioner  should, however, refer to the most current edition of SW-846 when
preparing a petition.


5.2.4  Statement of Need and Justification for the Proposed Test Method

     This final section of a petition contains two parts.   In the first part,
the  petitioner should briefly state why  the proposed method is heeded.   The
second part (the justification) should contain the test data that will be used
to establish equivalency between  the proposed method and the  existing SW-846
method.   Thus, while  the preceding  section describes  the  test method  and
associated procedures and provides the information necessary to understand the
method,  this section supplies the data  that will enable EPA to  evaluate the
method's performance equivalency.

     5.2.4.1  Need

     The petitioner shall describe  briefly the benefits of the  proposed test
method compared to the existing method.   For example, (1) the proposed method
may be  less expensive to run than  the approved methods, or (2)  the proposed
method may employ a new, proprietary technique.   Numerous other reasons could
exist.


                                      18

-------
                                                        OSWER Policy nirective
                                                              #9433.00-2

     5.2.4.2  Justification—The Presentation of Test Data

     Petitioners who request approval  of a proposed test method  are required
to  provide performance  data  and  statistical  analyses  that  are  used  to
demonstrate  the equivalency of  the proposed  method to  the  approved SW-846
procedure.   Along with  the test data, the petitioner  should provide  all  QC
data to substantiate the validity of the test data.  Petition approval is then
based on EPA's evaluation  of the  petitioner's data and  statistical analyses
from representative samples of hazardous wastes, ground water, or whatever the
appropriate matrix may be.

     Where the proposed test method measures waste characteristics or constit-
uents, EPA requires that  petitioners propose  an experimental design  for the
Agency's review and approval before testing begins.   (See Figure 3-1.)   This
design must yield the comparability  data that reflect the results  of testing
samples spanning the range of the method's applicability. This design includes
both the sampling and testing program to  be conducted for the petition.   The
number of samples and their locations  may vary with each petition  along with
the number of sample replicates and the timing for conducting tests.   Section
4 and Appendix B discuss  the development  of simple and  complex experimental
designs.

     Sampling Procedures.   Except when  new sampling techniques are proposed,
petitioners may  obtain the  samples in any  manner so  long as the  sample is
representative of the material that is to be tested.   If the ambient level  of
the  constituent or property of interest  is below the detection limit  of the
SW-846 approved test procedure, the  samples should  be spiked to  levels near
the detection limits of the SW-846 approved test procedure and the recommended
maximum property or concentration level.

     Documentation of Test Results.   All  test  results should  be documented
with the following information:

         Name,  address,  and telephone  number of  each  laboratory  facility
         performing the testing, if different from the sampling laboratory

         Date the test method was performed

         Sample number

         Parameter or constituent measured

         Test method

         Test results.

Explanations or  additional information  on inconsistencies  or  deviations in
test  results  should  be  furnished as  necessary.   A  suggested  format for
displaying the test result information is  shown as Table 3-1A in  the Example
Test Method Equivalency Petition (Appendix D).
                                      19

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2

     Statistical Testing.  The petitioner performs the statistical analyses of
all testresults and submits these analyses  along with  test data to  EPA as
part of the petition. The data  analyses entail a series of  statistical tests
that  are used to determine if  results from  the proposed method  satisfy the
data quality  objectives  established  at  the beginning  of  the  petitioning
process. Section  4 and  Appendix  B  of  this manual  provide  more  detailed
information about statistical evaluation procedures.

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2

                                  SECTION fi

                      CHECKLIST OF PETITION REQUIREMENTS


     Table 6-1 is a checklist of items developed to help petitioners prepare a
complete petition; the checklist  is designed so the petitioner  can determine
whether an item is or is not properly addressed.   By addressing the checklist
items appropriately, the petitioner  can reduce  the likelihood that  EPA will
request additional information about the method.

     The  checklist is divided  into the  four major  categories  discussed in
Section 5, "How to Prepare a Petition":

          Name and Address

          Certification of Accuracy and Responsibility

          Description of Proposed Action

          Statement of Need and Justification for the Proposed Test Method.

These four  categories  also correspond  to the  example  equivalency petition
presented in Appendix D of this manual.  Applicable RCRA regulations are cited
within the checklist.

-------
                                                       OSMFR  Policy Directive
                                                              #9433.On-?
          TABLE 6-1.   A CHECKLIST OF  ITEMS  TO  INCLUDE  IN A  PETITION
  I.  NAME AND ADDRESS  (40 CFR  260.20  (b)(l))

     a.  Are the petitioning party's  name  and address  identified?  	yes 	no

     b.  Is a specific contact listed along with  his or  her
         title and telephone number?                              	yes 	no


 II.  CERTIFICATION OF ACCURACY  AND RESPONSIBILITY

     a.  Does the petition include the  appropriate  certification
         statement?                                              	yes 	no


III.  DESCRIPTION OF PROPOSED ACTION   (40  CFR 260.20 (b)(3))

     a.  Does the petition include a  test  method  description  that
         follows SW-846 format?  (40 CFR 261.21  (b)(l))?           _yes 	no

       i.  Does the "Scope and Application" section —

           .  address sample properties or constituents  measured
           .  describe types of  samples to be tested
           .  briefly describe sample preparation before
              the test method is applied
              specify any procedural  changes for  special  samples
           .  specify qualifications  for participating personnel?	yes 	no

      ii.  Does the "Summary of  Method" section ~

           .  summarize the test method description
           .  include any detection limits?                       	yes 	no

     iii.  Does the "Interference" section identify all  potential
            interferences that may occur when the test method
            is conducted?  For example  —                         	yes 	no

              light scattering
           .  nonspecific absorption
           .  high concentrations of  other metals suppressing absorbance
           .  viscosity variations that may alter aspiration  rates
           .  artifacts from sample processing  glassware, hardware, or
              low-purity reagents
           .  additional sample  cleanup for desired sensitivity
           .  contaminated glassware
           .  plastic apparatus
           .  ambient pressure
           .  sample nonhomogeneity
           .  operator bias

 ~~                                                            (Continued)

-------
                                                        OSWER  Policy  Directive
                                                              #9433.00-2

    TABLE 6-1.  A CHECKLIST OF ITEMS TO INCLUDE  IN  A  PETITION  (Continued)
Ill.a.iv.  Does the "Apparatus and Materials"  section  —

            .   provide a detailed list of each item used  for  the method
            .   thoroughly describe apparatus  as  needed
            .   provide detailed construction  guidelines for apparatus
               that must be built?                                 	yes 	no

      v.  Does the "Reagents" section identify reagents according to —

            .   test method
            .   sample preservation and preparation
            .   proper grades
               reagent concentrations and preparation  method
            .   maximum reagent shelf life allowed?                  	yes	no

     vi.  Does the "Sample Collection, Preservation, and
          Handling" section

            .   address the considerations in  Section 1 of
               SW-846                                               ^yes _no

    vii.  Does the test method "Procedures" section —

            .   describe procedures in a succinct, stepwise fashion
            .   begin immediately after sample  collection
            .   address the following topics as appropriate:
                -  sample preparation
                -  entering the sample into the testing device
                -  operating conditions of the  test  method apparatus
                -  calibration procedures
                -  test method procedures
                -  calculation of results
                -  management of leftover samples and
                  contaminated equipment and materials?             	yes	no

   viii.  QC procedures (see Item Ill.d)

     ix.  Does the test method description include  a list of
          references —

            .   pertinent to the development and  content of the  method
            .   demonstrating the procedure's  applicability?         	yes	no


                                                                 (Continued)
                                      23

-------
                                                        OSWER  Policy  Directive
                                                              #9433.00-2

    TABLE 6-1.  A CHECKLIST OF ITEMS TO INCLUDE IN A PETITION  (Continued)


Ill.b.  Does the petition include a section  on  "Description  of
        Applicable Samples or Matrices" (40  CFR 261.21  (b)(2))
        that —

        .  for a general  waste matrix,  such  as  sludge,  describes
           the physical  properties affecting test  method
           performance
        .  for specific  waste streams,  provides RCRA waste numbers
           and a brief background on the waste  generation source
           in general, provides enough  information about the
           waste for EPA to evaluate the method's  applicability?    _ yes _ no

    c.  Does the petition include a section  on  "Assessment of  Limiting
        or Interfering Factors" (40 CFR 260.21  (b)(4))  that  addresses--

        .  identification of interferences and  limitations
        .  effect of interferences and  limitations on test method
           performance (effects on specific  wastes where appropriate)
        .  corrective measures available                           _ yes _ no

    d.  Does the petition include a section  on  "Test Method  Quality
        Control Procedures" (40 CFR 260.21(b)(5))  that  addresses —

        .  training and  evaluation of laboratory personnel
        .  maintenance of equipment
        .  test method practices
        .  recordkeeping and statistical procedures                _ yes _ no


IV.  STATEMENT OF NEED AND JUSTIFICATION FOR THE PROPOSED TEST METHOD
    (40 CFR 260.20 (b)(4))

    a.  Does the petitioner briefly describe why the proposed  test
        method should be approved?                                 _ yes _ no

    b.  Does the petition include an experimental  design that
        satisfies the data quality objectives established by the
        EPA?                  .                                        es    no
    c.   Does the petition include the test  data (absolute or
        comparative) required as a result of prepetition            _ yes _ no
        negotiations with EPA?

    d.   Are test data displayed clearly with full  documentation
        for each sample?                                           _ yes _ no

    e.   Does the petition present complete  statistical  analyses
        of test data along with QC data collected  during  testing?  _ yes _ no
                                      24

-------
                                                   OSWER Policy Directive
                                                         #9433.00-2
                             REFERENCES
1.  Test Methods for Evaluating Solid Waste:   Physical/Chemical  Methods.
    SW-846,  3rd Edition, U.S. Environmental Protection  Agency,  Washing-
    ton, D.C.   1986.   Available from  Superintendent of Documents, U.S.
    Government  Printing   Office,  Washington,  DC   20402  as  document
    955-001-00000.

2.  Guidelines  for  Selection and  Validation of  US  EPA's  Measurement
    Methods.   Draft  document  prepared by  Office of  Acid  Deposition,
    Environmental Monitoring  and Quality  Assurance,  U.S. Environmental
    Protection Agency, Washington, DC 20460.  January 1986.

3.  Validation of Testing/Measurement Methods.  Prepared by U.S.  Environ-
    mental Protection Agency, Office of Research and Development   for the
    Office of Solid  Waste, Washington, DC  20460.  EPA  600/X-83-060.

4.  Harmonization of Biological Testing Methodology:  A Performance-based
    Approach.  Aquatic Toxicology and Hazard Assessment:   Eighth Sympos-
    .ium.  ASTM STP891.   R.C. Banner and D.J. Hamsen, editors.   American
    Society for  Testing and  Materials.   Philadelphia, PA.  1985.   pp.
   . 288-301.

5.  MacGraw-Hill  Dictionary  of  Scientific  and  Technical  Terms.  2nd
    Edition.MacGraw-Hill Book Company, New York City, NY.1978.
                                 25

-------
                                       OSWER Policy Directive
                                             #9433.00-2
                 APPENDIX A

40 CFR 260, SUBPART C - RULEMAKING PETITIONS

-------
                                                        OSWER  Policy  Directive
                                                              #9433.00-2
                                  APPENDIX A

                40 CFR 260, SUBPART C - RULEMAKING PETITIONS
260.20 General
     (a) Any  person may petition  the Administrator  to modify or  revoke any
provision in Parts 260  through 265 of this chapter.   This  section  sets  forth
general requirements which  apply to all  such petitions.  Section 260.21  sets
forth  additional requirements for  petitions to  add  a testing  or  analytical
method  to  Part 261,  264, or  265.   Section  260.22  sets forth  additional
requirements for petitions to exclude  a  waste  at a particular  facility  from
261.3 of this  chapter or the lists of  hazardous  wastes in  Subpart  D   of  Part
2611.

     (b)  Each petition must  be submitted  to the Administrator  by certified
mail and must include:

     (1)  The petitioner's name and address;

     (2)  A statement of the petitioner's interest in  the proposed action;

     (3) A description  of the proposed action, including  (where appropriate)
suggested regulatory language; and

     (4) A statement of the  need and  justification for the  proposed action,
including any supporting tests, studies,  or other  information.

     (c) The Administrator will make a  tentative  decision to grant  or deny a
petition and  will publish notice  of such  tentative  decision, either in the
form  of an  advanced notice of  proposed rulemaking,   a proposed rule,   or a
tentative  determination to deny  the petition,  in the  Federal  Register for
written public comment.

     (d) Upon the written request of any  interested person,   the Administrator
may, at  his discretion,  hold  an informal  public hearing   to  consider oral
comments on the tentative decision.   A person requesting a  hearing  must  state
the issues to be raised and explain why written comments would not  suffice to
communicate the person's views.   The Administrator may in any case  decide  on
his own motion to hold an informal public hearing.

     (e) After evaluating all  public comments  the Administrator will make a
final decision by publishing in the Federal Register a regulatory amendment or
a denial of the petition.
1 Note:  Section  260.22 (Delisting  Procedures)  is not  applicable  to  this
  guidance manual.
                                     A-l

-------
                                                        OSWER  Policy Directive
                                                              #9433.00-2

260.21  Petitions for equivalent testing or analytical  methods

     (a) Any person seeking to add a testing or analytical  method  to Part 261,
264, or 265 of this chapter may petition for a regulatory  amendment under this
section and  260.21).   To be  successful, the  person must demonstrate  to  the
satisfaction of  the Administrator that  the proposed   method  is   equal  to or
superior  to the corresponding method prescribed  in Part  261, 264, or  265 of
this  chapter, in terms  of its  sensitivity, accuracy, and   precision (i.e.,
reproducibility).

     (b)  Each petition  must Include, 1n addition to the  information required
by 260.20 (b):

     (1)  A full description  of the proposed method, including all procedural
steps and equipment used in the method;

     (2)  A description of the types of wastes or waste matrices for which  the
proposed method may be used;

     (3)  Comparative results  obtained  from using the proposed  method with
those obtained from using the relevant or corresponding methods prescribed in
Part 261, 264, or 265 of this chapter;

     (4)  An assessment of any  factors which may interfere with,  or  limit  the
use of, the proposed method; and

     (5)  A description of the  quality control procedures necessary  to ensure
the sensitivity, accuracy, and precision of the proposed method.

     (c)  After receiving a  petition for  an equivalent method,   the Adminis-
trator may request any additional information on the proposed  method  which he
may reasonably require to evaluate the method.

     (d)  If the Administrator amends the  regulations  to  permit use  of  a  new
testing method,  the method  will  be incorporated  in  "Test   Methods  for  the
Evaluation of Solid Waste:   Physical/Chemical Methods," SW-846, U.S. Environ-
mental Protection Agency, Office of Solid Waste, Washington, D.C.  20460.
                                     A-2

-------
                                 OSWER Policy Directive
                                       #9433.00-2
           APPENDIX B



   STATISTICAL PROCEDURES TO

EVALUATE TEST METHOD EQUIVALENCY

-------
                                                        OSWER  Policy  Directive
                                                              #9433.00-2
                                  APPENDIX B
          STATISTICAL PROCEDURES TO EVALUATE TEST METHOD EQUIVALENCY
     This  appendix is intended to assist  both EPA and the petitioner in  the
selection and evaluation  of experimental  designs and methods  of statistical
analysis for equivalency studies.   Section 1 establishes the  basic approaches
to equivalency testing  and addresses simple (single-site) cases.   Sections  2
and 3 deal  with more complex (multi-site) cases.   Although this   appendix is
written mainly for statisticians, much  of the  material is elementary;   it is
included here to fix  ideas, terminology,  and approaches recommended  by EPA.
Alternative  designs  and  approaches  are  of  course  possible  and  may  be
negotiated between the Agency and the petitioner.

Bl.  EXPERIMENTAL DESIGNS AND METHODS OF DATA ANALYSIS FOR SINGLE  SITE CASES
     This section presents statistical methods to determine the equivalency of
test methods proposed for simple (single-site) cases.    To assist  the readers,
a list of symbols used in Section Bl is provided at the end of Appendix B.
Bl.l  Experimental Designs for Single-Site Cases
     Ideally, the choice  of experimental design and statistical  analysis  for
equivalency  testing involves a professional statistician.  As noted earlier,
EPA encourages the petitioner to use the services of a statistician  for  those
cases involving multiple  waste site  designs and analyses.   For  single   site
cases,  the  following  sections "offer some  reasonably  simple  experimental
designs and statistical analyses where professional statistical assistance  may
not be available.
     The  designs for the  absolute and  comparative cases are  quite similar.
Both involve a data matrix consisting of two columns (corresponding  to levels
of spiking  concentrations for  the  absolute case  and test  methods  for the
comparative  case) and ten rows (corresponding  to the number of days).   Each
                                     B-l

-------
                                                        OSWER  Policy  Directive
                                                              #9433.00-2
cell of the data  matrix contains two observations  or  replicate  measurements.
A schematic of these designs  is given in Figure Bl-1.    Each  design   uses  ten
randomly selected waste  samples from a single site.    For  the absolute  case,
each  sample is split into four  subsamples,  two subsamples are spiked at  the
low concentration and two are spiked at the high concentration.  Spiking  waste
samples yields information on precision and bias.   The spiking concentrations
should be  near the  regulated level of  the waste   component  and also in  the
operation  range of  the  method  where  the percent  recovery is  relatively
constant.   In  particular,  the spiking  concentrations should  be   above  the
detection limit,  with  recommended spiking  concentrations of T/2   and  3T/2,
where T is the regulatory threshold.
     For the comparative case, each sample is split into four   subsamples,  two
subsamples (randomly selected) are  analyzed by  the proposed  test  method  and
the remaining two are  analyzed by the approved method.  All  four  subsamples
are to be  analyzed on a single day.   These designs  are regarded as minimal.
If additional resources are available, increasing the number  of  concentration
levels, the number of days and/or the number of samples is  recommended.
B1.2  Preliminary Analysis
      This section gives  procedures for (1)  identifying and  handling unusual
or "outlying" measurements, (2) testing whether error variances are  equal  (as
assumed when applying the analysis of variance procedure),  and (3) determining
what data  transformation  will "help  stabilize  the  variance  when  unequal
variances are found.   Material in  this section applies to both   the absolute
and comparative single-site cases.
      81.2.1   Screening for Outliers.   Whenever  the  numerical value  of an
observation  (as opposed to the rank,  for instance)  is used  in   an analysis,
outliers or unusual observations can seriously affect the results.  Therefore,
                                     8-2

-------
                                                  OSWER Policy Directive
                                                        #9433.UU-2
      a)  Absolute

           Spiked
        Concentration:
        Low       High
Day 1
(Sample 1)
Day 2
(Sample 2)
Day 3
(Sample 3)
Day 4
(Sample 4)
Day 5
(Sample 5)
Day 6
(Sample 6)
Day 7
(Sample 7)
Day 8
(Sample 8)
Day 9
(Sample 9)
Day 10
(Sample 10)
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
b)  Comparative
      Test Method:
 Proposed	Approved
                                Day 1
                                (Sample 1)
                                Day 2
                                (Sample 2)
                                Day 3
                                (Sample 3)
                                Day 4
                                (Sample 4)
                                Day 5
                                (Sample 5)
                                Day 6
                                (Sample 6)
                                Day 7
                                (Sample 7)
                                Day 8
                                (Sample 8)
                                Day 9
                                (Sample 9)
                                Day 10
                                (Sample 10)
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
X X
Figure Bl-1.  Layout of experimental designs for single site case.
                               B-3

-------
                                                        OSWER Policy  Directive
                                                              #9433.00-2
prior to running the planned  analysis, one should carefully  examine   the  data
and identify the  "suspect" values  for further  study.    If  it   is determined
(based  on supportable  evidence)   that an  outlying observation   is   a  gross
deviation from prescribed procedures (e.g., recording or  calculation errors),
the value in question should be rejected  unless  it can  be restored   through  a
correction  to  the identified  error.    In  those  cases  where   there is   not
sufficient evidence to make this determination, a statistical  procedure should
be employed in identifying "outliers".    Without  some objective procedure,  one
may find experimenters  using analytical results  to determine whether unusual
values are to be retained or discarded.  Clearly, this should be  avoided.
     It is suggested that each column of  the designs shown in Figure Bl-1 be
screened for outliers, i.e., each  concentration  in the  absolute  case or  each
method in  the comparative  case.    The discussion  here will  be in  terms  of
concentration  for  the sake  of  specificity,  but  can  be   adapted to   the
comparative   case  by   substituting  method  for  concentration.    For   each
concentration, calculate the  overall mean  (Y),  the  between-day mean square
error  (MSB), and  the  within-day mean  square error  (MSW).   Procedures  for
estimating MSB and  MSW are given in Section  B1.3; they may  also be   found in
most statistical texts under discussions of one-way analysis  of variance.   The
total variance of a single observation from Figure Bl-1  is estimated  by
                                 =  %(MSW + MSB)
and the corresponding standard deviation is
                                                  MSB)
It is also  suggested  that any observation  more than 4.0 standard deviations
from  the  grand average, i.e., any  value outside T ± 4.0 s   ,  be considered

                                     B-4

-------
                                                        OSWER  Policy  Directive
                                                              #9433.00-2
suspect.  (An  alternative  is to use the  standard deviation  derived from the
twenty observations for a given  concentration,  ignoring  days,  as  an estimate
of s   .)
    TOT
     The  petitioner is free to suggest  other procedures that may  be  used to
identify  outliers such as Thompson's t-test  or the Dixon ratio test.    These
and other approaches for handling outlying observations are described in  (1).
     In order to preserve the balance of an experimental  design, observations
rejected as outliers should  be replaced  by representative values.    A simple
approach is to substitute the overall mean, excluding the  suspect  observation
for  the value  of the outlier.   Other procedures  may be  employed.  In all
cases,  the raw  data should  be  reported along  with an  indication  of what
observations were considered outliers,  how they  were detected, and   how they
were treated.
      Bl.2.2   Equality of Replicate Variance.   One  of  the  assumptions for
applying the analysis of variance  procedure is   that the error   variances (or
replication  variances)  are equal  under all   conditions.  The experimental
designs shown in Figure  Bl-1 include  duplicate measurements  for  each condi-
tion, i.e.,  two observations  within each cell.   There are  two reasons  for
utilizing replicate measurements.   First, it   enables the assumption of  equal
variances to  be tested,  and,  second, it  allows the  concentration  by  day
interaction effect and the method by day interaction effect to be evaluated.
     It is suggested that the initial statistical analyses use recovery of the
spiked amount (i.e.,  difference in  measured  concentrations before  and  after
spiking  divided by  spiked  concentration) as  the response  variable in the
absolute case and the logarithm  of the  observed measurement  as the response
variable in the comparative case.   If waste samples contain concentrations in
the working range of the test methods, one often finds the  standard  deviation
                                     B-5

-------
                                                        OSWER  Policy  Directive
                                                              #9433.00-2
of concentration measurements to be proportional  to concentration.    When  this
is true, a benefit of analyzing recoveries in the absolute case and  logarithms
of measurements  in the comparative  case is  that variances will  be approxi-
mately equal.
     To test for equality of replicate variances, calculate (for the  itn   day
and jth concentration or  method) the  within-cell  average (V"ij)   and standard
deviation  (S-jj) (of  recoveries  in  the  absolute  case; of  logarithms   of
measurements in the comparative case) for each cell.    Plot log(S-jj)   vs.   log
(T-jj) and use ordinary linear regression

                          log (Sij) = a log(Tij)  + b

to test  for dependence  of  variation  on  level of   concentration.    If   the
regression is  significant, i.e.,  coefficient "a" is  significantly  different
from zero, the variances  are considered unequal  and  some  data transformation
should be employed in an effort to stabilize the  variances before applying the
analysis of variance procedure.   One approach is to  use the power transforma-
tion, yl"a, on  the raw data and test  again for  equality of variances.    Note
that "a" is the coefficient  in the above regression  equation.   When a=l,  the
log transformation is appropriate (2).
      B1.3  Statistical Analysis for Single-Site  Cases With Absolute  Data
            Quality Objective?
      This section provides details  of the statistical test used to evaluate
whether a proposed test method is  at least equivalent (in terms  of  precision
and  bias) to the absolute data  quality objectives  specified by the  EPA for
single-site cases.   It is assumed that preliminary analyses,  i.e.,   tests for
outliers  and variance described  in Section  B1.2, have already  been carried
out.   These objectives are formulated in terms of the null hypotheses (Hj and
H2) stated in Section 4.2.   All analyses  discussed  here are  to be   performed
                                     B-6

-------
                                                        OSWER  Policy Directive
                                                              #9433.00-2
separately for each level  of concentration.   If any test leads to  rejection  of
either  HI or  H2, then the  proposed method  is considered  unacceptable.    In
addition, it  is assumed that  two replicate  measurements are taken   for  each
sample and that any modificaton on  the minimal experimental design is in the
direction of additional concentrations and/or days.   The formulas provided  in
the  analysis of variance (ANOVA)  table  have been simplified  for  the   case  of
two replicates per sample.   All  the analyses described  below  are  dependent  on
values of either the  F, t, or chi-square distributions.  In  situations where
this design has been modified to include additional  days/samples, appropriate
percentiles can be found in Table Bl-1 for up  to 20 days (the number   of  data
points, N, equals 40).
     Once  the data have been collected,  it should  be  presented  in   a manner
similar to the  configuration displayed in Table Bl-2.    Here  the  first column
shows the day (i) that the sample was tested, and the second and third columns
contain recoveries (Yji and Y2i)  obtained from the two replicate measurements.
The  remaining columns are  simple manipulations  of the recoveries  which are
used in  the calculations for the ANOVA  table shown  in Table Bl-2(b).   Note
that  totals are calculated for some  of the  columns in Table Bl-2(a); these
totals  are referred  to by  a  capital letter  for easy reference  in later
formulas.   The computations for the ANOVA table should   be  straightforward  if
the  data are arranged  and manipulated  as described  here.    However, it  is
suggested that the petitioner work through the example provided at the end  of
this section before beginning to analyze his own data.
     After the ANOVA table  has been  completed, the petitioner should screen
for outliers  as described in Section Bl.2.1.   Note that the between-day and
                                     B-7

-------
DO


CO
TABLE Bl-1. PERCENTILES OF THE t v2
Percenti les of
Degrees of
Freedom
9
10

11
12
13
14
15

16
17
18
19
20

21
22
23
24
25

26
27
28
29
30

35
40







the t and v2
A*

t.975
2.26
2.23

2.20
2.18
2.16
2.14
2.13

2.12
2.11
2.10
2.09
2.09

2.08
2.07
2.07
2.06
2.06

2.06
2.05
2.05
2.05
2.04

2.03
2.02







• 	 	
distributions
2
X .95
16.92
18.31

19.68
21.03
22.36
23.69
25.00

26.30
27.59
28.87
30.14
31.41

32.67
33.92
35.17
36.42
37.65

38.89
40.11
41.34
42.54
43.77

49.80
55.76







Pei
Numerator
Degrees of
Freedom
1


9


10


11


12


13


14


15


16


17


18


19


AND F DTSTRIBUT
-"cerTtTles of thp"
Denominator
Degrees of
Freedom
29
30
40
9
10
20
10
11
22
11
12
24
12
13
26
13
14
28
14
15
30
15
16
32
16
17
34
17
18
36
18
19
38
19
20
40
IONS
F-distrib°


F.90
2.89
2.88
2.84
2.44
2.35
1.96
2.32
2.25
1.90
2.23
2.17
1.85
2.15
2.10
1.81
2.08
2.04
1.77
2.02
1.99
1.74
1.97
1.94
1.71
1.93
1.90
1.68
1.89
1.86
1.66
1.85
1.83
1.63
1.82
1.80
1.61

'ution


F.95
4.18
4.17
4.08
3.18
3.02
2.39
2.98
2.85
2.30
2.82
2.72
2.22
2.69
2.60
2.15
2.58
2.51
2.09
2.48
2.42
2.04
2.40
2.35
1.99
2.33
2.29
1.95
2.27
2.23
1.92
2.22
2.18
1.88
2.17
2.14
1.85


'

F.975
5.59
5.57
5.42
4.03
3.78
2.84
3.72
3.53
2.70
3.47
3.32
2.59
3.28
3.15
2.49
3.12
3.01
2.41
2.98
2.89
2.34
2.86
2.79
2.28
2.76
2.70
2.22
2.67
2.62
2.17
2.60
2.55
2.13
2.53
2.48
2.09
                                                                                                                                                                                              co _i.
                                                                                                                                                                                              co n
                                                                                                                                                                                              o c
                                                                                                                                                                                                  O
                                                                                                                                                                                                 <
                                                                                                                                                                                                 n>

-------
                                                       USHER Policy Directive

                                                             #9433.00-2
         Table Bl-2.  LAYOUT FOR SINGLE SITE CASE WITH ABSOLUTE DQOs
Day

i
1
2
3
*
D
a. Data Configuration for One Concentration
Recoveries

Yli
Yll
Yl2
Yl3
YID
9
Y2i
Y21
Y22
Y23
Y*2D
Y
li
Y2
11
Y2
12
Y2
13
Y2
ID
2
Y2
2i
Y2
21
Y2
22
Y2
23
Y2
20

T-i = YU + Y21
Tl
T2
T3
*
TO
9
T2
i
T2
T2
2
T2
3
i2
D
Totals
    °2      D    2    °
G= E  Y2    H= E  Y2  J= i  T.
                                                                    K=  I T
                            1 = 1
                                                  1 = 1
         b.  Analysis of Variance (ANOVA) Table for One Concentration
Source
Between Days
Within Days
Degrees
of
Freedom
D-l
N-D
Sum of
Squares
SSB = (K/2)-(J2/N)
SSW = SST-SSB
Mean
Square
Error
MSB = SSB/(D-1)
MSW = SSW/ (N-D)
F-statistic
F = MSB/MSW
Total
                 N-l
   SST = G + H - (JVN)
N = 2D


Y = J/N
                 (total  number of observations)


                 (overall  mean recovery)
s = SST/(N-1)    (sample variance)
                                     B-9

-------
                                                        OSWER  Policy Directive
                                                              #9433.UO-2
within-day  mean  square errors (MSB and MSW)  have already  been  calculated  for
the ANOVA table,  and the  formula for the  overall  mean  (T)  is  given  at  the
bottom  of Table  Bl-2.   As mentioned  before,  any   outlier should either  be
corrected (if an error was  discovered)  or replaced  by  a  representative  value
in  order to  preserve  the  balance  of the  desiyn.    In  either  case,   all
calculations performed up to this point  will  have  to be repeated  with   the  new
value substituted.
     Before  the system precison and bias  can be   tested,  it  is  necessary to
determine if  there is a  significant day  effect  (i.e., the   average  recovery
varies systematically  over the  days).    This is   done using  the F-statistic
calculated in the ANOVA table.    If this value is  greater than or equal to  the
critical F-value, then it is concluded that there  is a  significant day  effect.
The critical F-value for this test  is fQ-it^-Qt.go  where (D-l)  and  (N-D)  are
the numerator and denominator degrees of freedom,  respectively,  and .90 is  the
level  of confidence for the test.   For the minimal  design, there are  10 days
(D) with 2 replicates per day,  yielding  a total  sample  size N  =  2D = 20.    The
critical F-value is Fg^g.^O = 2.35.  Therefore,  under the minimal design, if
the F-value from  the ANOVA  table is greater  than  or   equal  to  2.35,   it is
concluded that a significant day  or sample effect is present.   The result of
this  test determines  the approach to  be used  in  the following analyses of
system bias and precision.
Case I—Day Effect Not Significant
     If the day effect is not significant (i.e., F-statistic < 2.35),  bias  and
precision can  be tested using  the overall   mean Y and  the sample variance s2
(from  the bottom of Table Bl-2).   A 95  percent  confidence interval  (CI)  for
the true mean recovery, p, can  be computed using the following formula:
                                     B-10

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2
          Pr(Y - tM-1,.975    fN     £  v £ Y + tN.lf .975 ?N      =.95
where ^,1^975  is  the  97.5th  percentile  of  the t-distribution  with  N-l
degrees of freedom.   When N=20 (minimal design), ^9^.975 = 2.09 and the above
CI simplifies to
                     JZ        _        J2T
          Pr(Y - 2.09  T20   £ u£ Y + 2.09  f 20    = .95
          Pr(7 - .467s £ u £ Y +  .467s) = .95
Substituting values  for Y  and  s will yield lower and upper bounds of y for the
proposed  method.   In   the absolute case, the  EPA will specify a maximum bias
(b0) for  the  proposed method.  Since, in  the  absence of bias, recovery is
defined  to be 1,  the   Agency  will accept a  proposed  method whose true mean
recovery is somewhere  between  (l-b0, l+b0).  Therefore,  if any portion of the
95 percent  CI for y  overlaps the interval (l-b0, l+b0), then the bias of the
proposed  method is  considered to be within  acceptable limits.  Similarly, a
lower  bound can  be given for the true  variance, a2, of the proposed method
(still assuming no day effect)
where Y*       is the 95th  percentile of  the chi-square distribution with N-l
       n~ * f • -?*)
degrees of freedom.   For the minimal  design v2      = 30.14.  The Agency will
                                    2        is , «SD
also  specify  a maximum variance,  a0, for a  proposed  method.  If the lower
                                                     2
bound  of the true variance is  less  than or equal to OQ, then the precision of
the proposed  method  is considered to be within acceptable limits.
                                    B-ll

-------
                                                        OSklER Policy Directive
                                                              #9432.00-2

Case 2—Day Effect Significant

     If the  day effect is significant (i.e., F-statistic _> 2.35),  methods  for

testing bias  and precision are slightly different.  In calculating the CI  for

the true mean recovery, p, s2 should be replaced by MSB (from the ANOVA table)

and the degrees of freedom of the  t-statistic is  changed from (N-l)  to (D-l)

yielding:
                     ..	         _          JMSB
      pr(Y-tD-l,.975 t  N    1 nl Y+*D-1,.975>  N     = .95


For the minimal design,  tg ,975 = 2.26.  As noted for the case where there was

no  day effect, if this  interval overlaps the interval  (l-b0, l+b0), then the

hypothesis  Hj:  |bnew|  £ b0  is  not  rejected, and  the  proposed  method is

considered  acceptable.   In  order to  test the precison  when there  is  a day

effect, the following calculations must be made:


      g = s2    = V2(MSB + MSW) ,


                     4g2
      n =
             MSB2          MSU2
             D-l     +     N-D
The value  n is an approximation for the new  degrees of freedom and should  be

rounded up to the next highest  integer (e.g., if n = 11.21,  then use n =  12).

A 95  percent  CI for  the lower bound of the true  variance  of  the  proposed

method is

              fng       -i
           X       < a2  = .95
            n,.95  -   J


                                                            2
As before, if the lower bound is less than or equal to the a  specified by the
                                                            o
EPA, then the hypothesis Hg:   a2   _< a2  is not rejected, and the precision  of
                               new    o

                                     B-12

-------
                                                        OSWER  Policy Directive
                                                              #9433.00-2
the  proposed method  is considered  acceptable.    Note  that   the  degrees   of
freedom for the chi-square statistic is  n rather  than  (N-l), as in  the  case  of
no day effect.   This percentile will  have to be  looked  up  in Table Bl-1 even
for the  minimal  design,  since the degrees  of freedom  is  a  function  of the
data.
Numerical Example 1
     Hypothetical recovery data from waste samples  spiked at two concentration
levels are analyzed to  illustrate the calculations involved in  applying some
of the analytical procedures just described.
                                     B-13

-------
Example  1.  Analysis of Recoveries From Low  and High
             Single Site Case With Absolute DQOs
                                   OSWER Policy Directive
                                         #9433.00-2

                                    Concentrations For The
A.
Day
i

1
2
3
4
5
6
7
8
9
10
Low Concentration
Recoveries
Y

1.17
1.17
0.86
1.00
0.70
1.43
0.98
1.66
0.96
1.17
Y2i

1.24
1.66
0.40
0.39
0.48
0.41
1.60
0.74
1.08
1.36
Y2
11
1.37
1.37
0.74
1.00
0.49
2.04
0.96
2.76
0.92
1.37
Y2
21
1.54
2.76
0.16
0.15
0.23
0.17
2.56
0.55
1.17
1.85
Ti * Yl1 + Y2i

2.41
2.83
1.26
1.39
1.18
1.84
2.58
2.40
2.04
2.53
T2
1
5.81
8.01
1.59
1.93
1.39
3.39
6.66
5.76
4.16
6.40
Totals
   G = 13.02  H = 11.14    J  =  20.46
                            K  =  45.10
0 = 10
N = 20 = 20
SSB = (K/2) - (JVN) = (45.10/2) - (20.46/201 = 1.62
SST = G + H - (J2/N) = 13.02 + 11.14 - (20.46720)  = 3.23
SSW = SST - SSB = 3.23 - 1.62 = 1.61
MSB = SSB/(D-1) = 1.62/9 = 0.18
MSW = SSW/(N-D) = 1.61/10 = 0.16
£ = MSB/MSW = 0.18/0.16 = 1.13
Y = J/N = 20.46/20 = 1.02
s2 = SST/(N-1) = 3.23/19 = 0.17
                                 ANOVA Table
Source
Between Days
Within Days
Degrees
of Freedom
9
10
Sum of
Squares
1.62
1.61
Mean
Square Error
0.18
0.16
F-statistic
1.13
Total
19
3.23
                                     B-14
                                                                     continued

-------
                                                      OSWER Policy Directive
                                                            #9433.00-2
Example 1 (continued)

1) Screening for Outliers
   T = 1.02
   s2  = (MSB + MSW)/2 = (0.18 + 0.16)/2  =  0.17
    TOT
   s   = Js2= 0.41
    TOT  f  TOT
   Any value outside the interval
                  (T - 4.0 s    , T =  4.0  s    ) =  (-0.62, 2.66)
                           TOT          TOT
   is  considered suspect.   Since the  recoveries  in this example range  from
   0.39 to 1.66, it is concluded that this  data set  is free of outliers.
2) Testing Day Effect
   Using  a 10 percent significance level   (a= 0.10), the critical  value for
   testing the day effect is

             FD-l,N-D,l-a = F9,10,.90 = 2.35      (from Table Bl-1)
   Since the  statistic F  =  1.13  (from  the ANOVA  table) is   less  than the
   critical  value 2.35, it is concluded  that the day effect is  not signifi-
   cant.
3) Testing the System  Bias
   Since there is no day effect, a 95 percent CI for the true  mean recovery,
   M, is given by
           Pr[T- tN-1.975       l ul ^+  tH-l.97S  \/NJ *  -95,
   where  tj^.j  975 is the  97.5th percentile   of  the  t-distribution  with N-l
   degrees of freedom.  From Table Bl-1,  ^9^975  =  2.09.
                                                                    continued
                                      B-15

-------
                                                       OSWER  Policy Directive
                                                             #9433.00-2
Example 1 (continued)
       Pr[1.02 - (2.09)   V°-17/20 1 u£ !-02 + (2-09)   VO.17/20  =  .95

       Pr[0.83 <_ u <_ 1.21] = .95

   For this example, suppose the EPA had  specified  a maximum  bias b0  = 0.10.

   Therefore, a proposed method whose true mean recovery  was somewhere between

   (l-b0,  l+b0) = (U.90,1.10)   would be  considered accurate  by   the Ayency's

   standards.   Since  the  interval for  the true  mean   recovery,  p,  of the

   proposed  method  (0.83,1.21)  overlaps  the interval   (0.90,1.10),  it   is

   concluded that  the bias  of  the proposed  method is   within   the approved

   limits for measuring low concentrations.1

4) Testing the System Precision

   Since there is no day effect, a lower  bound for  the true variance,  a  ,  of

   the proposed method is given by:


         Pr[(N-l)s2/vJ    nc 1 a2] = .95 ,
                     N-l .95

   where y2       is the  95th  percentile  of the chi-square distribution with
         Vlf.95
   N-l degrees of freedom.  From Table 4-1, y2      =30.14.
*«..
                                                 95
         Pr[(19)(0.17)/30.14 £ a] = .95

         PrCO.ll < a2} = .95
         test for bias  has the potential  to penalize labs  with  good  precision
   and reward labs with poor precision.   Therefore,  when establishing DQOs  for
   each petition, OSW sets precision levels accordingly to  avoid this problem.
                                                                    continued
                                     B-16

-------
                                                      OSWER  Policy Directive
                                                             #9433.00-2
Example 1 (continued)
                                                                           o
   For this example, suppose the EPA had specified  a  maximum  variance of   a  =


   0.25.  Since  the lower  bound of the  true variance  (0.11)  is  less  than or


   equal to a2 = 0.25, it is  concluded  that  the  precision   of the  proposed
             o

   method meets with the Agency's standards for measuring  low concentrations.
                                                                     continued
                                      B-17

-------
Example 1 (continued)
                                                       OSWER  Policy  Directive
                                                             #9433.OU-2
B.
Day
i
1
2
3
4
5
6
7
8
9
10
Hiyh Concentration
Recoveries
Yll
1.31
1.47
0.86
1.02
0.23
0.34
1.00
1.61
0.93
0.74
V21
1.21
1.06
1.31
0.15
0.42
1.04
1.69
1.26
0.93
0.46
Y2
li
1.72
2.16
0.74
1.04
0.05
0.12
1.00
2.59
0.86
0.55
C
2i
1.46
1.12
1.72
0.02
0.18
1.08
2.86
1.59
0.86
0.21
Ti = Yii + Y21
2.52
2.53
2.17
1.17
0.65
1.38
2.69
2.87
1.86
1.20
T2
6.35
6.40
4.71
1.37
0.42
1.90
7.24
8.24
3.46
1.44
Totals
G = 10.83  H = 11.10
                            =  19.04
K = 41.53
D = 10
N = 2D = 20
SSB = (K/2) - (J7N) = (41.53/2) - (19.047201 = 2.64
SST = G + H - (JVN) = 10.83 + 11.10 - (19.04720)  = 3.80
SSW = SST - SSB = 3.80 - 2.64 = 1.16
MSB = SSB/(D-1) = 2.64/9 = 0.29
MSW = SSW/(N-D) = 1.16/10 = 0.12
F. = MSB/MSW = 0.29/0.12 = 2.42
Y = J/N = 19.04/20 = 0.95
s2 = SST/(N-1) = 3.80/19 = 0.20
Total
                                 ANOVA Table
Source
Between Days
Within Days
Degrees
of Freedom
9
10
Sum of
Squares
2.64
1.16
Mean
Square Error
0.29
0.12
F-statistic
2.42
19
             3.80
                                                               continued
                                     B-18

-------
                                                      OSWER Policy Directive
                                                            #9433.00-2
Example 1 (continued)

1) Screening for Outliers
   Y = 0.95
      P= (MSB + MSW)/2 = (0.29 + 0.12)/2  =  0.21

                 = 0.46
           TOT
   (Y-4.0s   ,T=4.0s    ) = (-0.89, 2.79)
             TOT          TOT
   Since  all recoveries  fall  within  this  interval, it is concluded that this
   data set is free of outliers.
2) Testing Day Effect
   Since  the  statistic  F  =  2.42  (from the ANOVA table) is  greater than or
   equal to  the  critical   value  Fgjo.^O = 2.35,  it is  concluded  that a
   significant day effect is present.
3) Testing the System Bias
   Since  there  is a  day effect, a 95 percent CI for the true mean recovery,
   u, is given by
         Pr[T- tD_1(.975  ^/MSB/N  <. u 1 7 + tD-if.975      /N] - .95
   where ^.1^.975  is  the  97.5th  percentile of  the t-distribution with D-l
   degrees of freedom.   From Table Bl-1,  tg^.gys = 2.26.
       Pr[0.95 - 2.26^0.29/20 £ y <. 0.95 •«• 2.26 ^0.29/20 = .95
       Pr[0.68 < u < 1.22]  =  .95
                                                                    continued
                                     B-19

-------
                                                       OSWER  Policy  Directive
                                                             #9433.00-2
Example 1 (continued)
   Since  this  interval   overlaps the limits set  by the EPA for mean  recovery
   (l-b0, l+b0) = (.90,1.10), it  is concluded  that the  bias  of the  proposed
   method is considered acceptable by the Agency's standards.
4) Testing the System Precision
   Since  there is a  day effect, the  following equations must be calculated:

       y = (MSB + MSW)/2  = (.29 + .12)/2  = 0.21
       n = 	4g	  =  	4(.21)        =   16.36
                                        .29*79)  + (,12*/
                   2
                   MMSW-VN-D;     (.29*79; + (.12^/10)

.*. n = 17 (rounding n to next highest integer).
A 95  percent CI  for the  lower  bound of  the  true  variance,  a2,  of the
proposed method is
       Pr[ng/x2  QC 1  a2] = .95 ,
              n,.95

   where v2     is  the 95th  percent!le of the chi-square distribution  with  n
          n,.95
   degrees of freedom.  From Table Bl-1, y2       = 27.59.
                                         17, .975

       Pr[(17)(0.21)/27.59 1 a2] = .95

       PrL"U.13 < a2J = .95
                                                                     continued
                                     B-20

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2
Example 1 (continued)
   Since this lower  bound  is  less  than  or equal   to the  maximum variance


   specified by the EPA, o2 = 0.25, the precision of the proposed method meets

   with the Agency's standards for measuring high concentrations.



Conclusion:  Two hypotheses were tested for the proposed method:



                               HI:
                               u     2.2
                               H?:   a    < a ,
                                     new ~  o

                2
where  b0  and a  are  data  quality  objectives  specified  by the  EPA.   The
                o

recovery data  obtained from this method were analyzed  separately for low and


high  concentrations (parts A  and B, respectively).   In both  cases, the  data


supported  the  hypotheses tested;   therefore, the  proposed  method  would be


accepted by the Agency as an approved procedure for measuring solid wastes.
                                     B-21

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
      B1.4  Statistical Analyses for Single-Site Cases with Comparative Data
            Quality Objectives
            This section describes statistical  procedures for  testing equiva-
lency of the proposed and approved methods, with respect to precison and  bias.
It is assumed that preliminary analyses  to test for outliers and  equality of
replicate variances (described in Section B1.2)  have already been carried out.
Under this approach, the approved method is considered the standard (i.e., its
measurements are assumed to be "correct"), and  the proposed method is compared
with it.   The test  hypotheses, H$ and 84, specified  in Section 4.2, reflect
similar precision and bias for the two methods.   If either of these hypotheses
is rejected, then the proposed  method fails the equivalency test.   For this
reason, the  comparative  approach  is  not recommended  when  the  petitioner
suspects that  his proposed method  is more  accurate than (and  therefore not
equal to) the approved method.  As noted in Section 4.2, if there is reason to
believe  that the  proposed method is  less biased  or more  precise  than the
approved method, then the petitioner should request using absolute rather than
comparative  data  quality  objectives.   It is  assumed  that  two replicate
measurements are  taken for  each  sample and  that any  modification  on  the
minimal design is in the direction of additional days/sample.   Table Bl-1 can
be used to obtain appropriate  percentiles for  the statistical  tests  when the
design  has been modified as such.   An example is provided at the end of this
section to clarify the analysis discussed here.
     The first  step in this analysis is  to complete an ANOVA table  for each
method.  The layout of the data and the necessary formulas are the same as for
the absolute case, with method substituted for  concentration (see Table Bl-2).
It is not necessary to calculate the F-statistics for this analysis.  Once the
ANOVA tables have been completed,  the petitioner may use the  between-day and
                                     B-22

-------
                                                        OSWER Policy  Directive
                                                              #9433.00-2
within-day  mean square  errors (MSB and  MSW)  to  check for outliers  in  each
method  as described  in Secton Bl.2.1.    Note  that  once an  outlier has  been
found and treated, then all results affected by it must be recalculated.
     The  precision of  the proposed  method can be  compared to  that of the
approved method by obtaining a 95 percent confidence interval for the ratio of
the  true  replicate variances, a2 (proposed method) and a2 (approved method),
                                 p                        a
                     MSWn            a2            MSWn
          Pr  - - -
                     -D, N-D,.975)   a      MSWa(FN-D,N-D,.025)
                                      d
where MSWp and MSWa  are the within-day  mean square  errors  (from  the ANOVA
tables) for the proposed and approved methods, respectively, and f:N-D,N-D,.975
and  FN-D,N-D,.025  are tne 97.5tn and 2.5tn percentiles of the  F-distribution
with (N-D)  degrees of freedom  in both the  numerator and  denominator.  Note
that
          FN-D,N-D,.025 = FN-D,N-D,.975 •
For the minimal design with 10 days (N=20), Fio,io,.975 s 3.72 and
F10,10,.025 = 0»27, and the above CI simplifies to
                  MSWD       a2       MSWn
          Pr  - - — < ~C-<  - - -   = .95
                                 _
               MSWa(3.72)    a      MSWa(0.27)
                              a
                    MSWn   a2        MSWp
          Pr  (0.27 - -!-£-<. - ) = -95
                    MSWa   a      MSWa(0.27)
                            a
Substituting  MSW for the  two methods  yields upper  and lower bounds for the
ratio  of the true  variances.  If  this interval  contains the value 1, corre-
sponding to a  = a , then  equivalency of the  method precisions is supported.
             P    a
In  contrast, an interval  that does not  contain the value 1 implies that the
                                     B-23

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
proposed  and  approved methods differ  with  respect to   precision,  and the
proposed method would not be approved.
     If the method  precisions are  considered  equal, then the petitioner can
proceed to compare their bias.  This  is done by performing a two-way analysis
of variance with  method and days as the factors.  A method by day interaction
is also  considered in this analysis.  The  calculations for the two-way ANOVA
(shown  in Table Bl-3) are  easily obtained from those of the  one-way  ANOVAs
already performed  for each method.  Note  that the subscripts "a" and "p" are
used in Table Bl-3 to indicate totals  from the one-way ANOVA calculations for
the approved and proposed methods, respectively.  Once the two-way ANOVA table
has  been  completed,  the  method  by  day interaction  is tested  using  the
F-statistic
                              MSMD
                              MSE
If  this  value  1s greater than  or equal to  the critical value Fo-i,2D,.95»
then  a  significant  interaction  is  concluded.   For  the  minimal  design,
^0-1,20,.95 = F9,20,.95 = 2.39.  A  method  by  day interaction indicates that
the  difference  between the two methods is not the same for all days.  There-
fore, if the  method  by  day  interaction 1s  statistically  significant, the
proposed method will not be accepted, as there is sufficient evidence that the
methods  differ for  some of the days.  Plots  of the data may help to examine
the sources of interaction more closely.
     If the method  by day interaction  is not significant, its sum of squares
(SSMD) and degrees of freedom (D-l) are added to those for the error to form a
new mean square error:
                            * -  SSE * SSMD    SSE + SSMD
                                 2D + (D-l) "    3D - 1
                                     B-24

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2
           Table 81-3.  TWO-WAY ANALYSIS OF VARIANCE (ANOVA)  TABLE
Source
Method
Day
Method x Day
Error
Degrees of
Freedom
1
(D-l)
(D-l)
2D
Sum of
Squares
SSM
SSD
SSMD
SSE
Mean
Square
Error
MSM
MSD
MSMD
MSE
        Total               4D-1                 SST


C  = (Ja + Jp)2/4D

SST = (Ga + Ha +Gp + Hp) - C

SSM = [(J2 + J2)/2D] - C
         a    p
        D             2
SSU = C z  (T1a + T1p)2]/4 - C
       1=1

SS (replicates) = [(Ka + Kp)/2] - C

SSMD = SS (replicates) - SSD - SSM

SSE = SST - SSD - SSM - SSMD

MSM = SSM/1 = SSM

MSD = SSD/(D-1)

MSMD = SSMD/(D-l)

MSE = SSE/2D


NOTE:  The  totals G, H, J, K, and T-j  are  calculated  in  the one-way  ANOVA
tables  for each method  (see Table Bl-2).  The  subscripts "a" and "p" denote
the approved and proposed test methods.
                                     B-25

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
The main effect for method can then be tested with the F-statistic

                           _ MSM
                             MSE*

If  this  value  is  greater  than  or  equal, the  critical  value Fl,3D-l,.95
(= 4.18  for minimal  design), then  a significant  method effect is concluded
(i.e., the  methods  do not  yield  equivalent measurements).  Therefore,   the
proposed method will not be accepted.
Numerical Example 2
     Hypothetical  measurements  from waste  samples  tested   by  two  methods
(proposed  and approved)  are analyzed  to illustrate  some of  the analytical
procedures just described.
                                     B-26

-------
  Example 2.
                                                       OSWER Policy Directive
                                                             #9433.00-2

              Analysis  of Measurements  from The Proposed And Aproved Methods
              for the Single Site Case with Comparative DQOs
A.  Proposed Method:  one-way ANOVA

      Logarithms
Day   of Measurements
1 YH Y2i
1
2
3
4
5
6
7
8
9
10
2.16
2.15
2.02
2.03
1.91
2.19
2.08
2.30
2.06
2.10
2.17
2.30
1.84
1.78
1.81
1.82
2.27
2.01
2.10
2.16
Y2
li
4.62
4.62
4.08
4.12
3.65
4.80
4.33
5.29
4.24
4.41
Y2
2i
4.71
5.29
3.39
3.17
3.28
3.31
5.15
4.04
4.41
4.67
T1 " YH + Y2i
4.32
4.45
3.86
3.81
3.72
4.01
4.35
4.31
4.16
4.26
T2
i
18.66
19.80
14.90
14.52
13.84
16.08
18.92
18.58
17.31
18.15
Totals
Gp= 44.16  Hp= 41.42   Jp= 41.25
                                                               Kp= 170.76
D = 10
N = 20 = 20
SSB
SST
SSW
MSB
MSW
                                          2
    = (K/2) - (J/N) = (170.76/2) - (41.25/2Q) = 0.302
    = G + H - (JVM) = 44.16 + 41.42 - (41.25/20) = 0.502
    = SST - SSB = 0.502 - 0.302 = 0.200
    = SSB/(D-1) = 0.302/9 = 0.034
    = SSW/(N-D) = 0.200/10 = 0.020
Y = J/N = 41.25/20 = 2.063
s2 = SST/(N-1) = 0.502/19 = 0.026
                                 ANOVA Table
Source
Between Days
Within Days
Degrees
of Freedom
9
10
Sum of
Squares
0.302
0.200
Mean
Square Error
0.034
0.020
     Total
                        19
                  0.502
                                                                     continued
                                     B-27

-------
Example 2 (continued)
                                                       OSWER Policy Directive
                                                             #9433.00-2
B.  Approved Method:   one-way ANQVA

      Logarithms
Day   of Measurements
i Yli Y2i
1
2
3
4
5
6
7
8
9
10
2.32
2.36
2.18
2.18
1.94
2.01
2.24
2.39
2.21
2.11
2.30
2.27
2.29
1.92
2.00
2.21
2.40
2.32
2.21
2.03
Y2
li
5.38
5.57
4.75
4.75
3.76
4.04
5.02
5.71
4.88
4.45
Y2
21
5.29
5.15
5.24
3.69
4.00
4.88
5.76
5.38
4.88
4.12
T1 = YH + Y2i
4.62
4.63
4.47
4.10
3.94
4.22
4.64
4.71
4.42
4.14
T?
21.34
21.44
19.98
16.81
15.52
17.81
21.53
22.18
19.54
17.14
Totals
6a = 48.31  Ha = 48.39   Ja = 43.89
                          K  = 193.29
D = 10
N = 2D = 20
SSB = (K/2) - J2/N) = (193.29/2)  - (43.89Z/20l = 0.328
SST = G + H - (JVN) = 48.31 + 48.39 - (43.89/20)  = 0.383
SSW = SST - SSB = 0.383 - 0.328 = 0.055
MSB = SSB/(D-1) = 0.328/9 = 0.036
MSW = SSW/(N-D) = 0.055/10 = 0.006
Y = J/N = 43.89/20 = 2.195
s2 = SST/(N-1) = 0.383/19 = 0.020
                                 ANOVA Table
Source
Between Days
Within Days
Degrees
of Freedom
9
10
Sum of
Squares
0.328
0.055
Mean
Square Error
0.036
0.006
     Total
  19
0.383
                                                                     continued
                                     B-28

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2

Example 2 (continued)


C.  Two-Way ANOVA

C = (Ja + Jp)2/4D = (43.89 + 41.25)2/(4)(10)  = 181.22

SST = Ga + Ha + Gp + Hp - C = 48.31 + 48.39 + 44.16 + 41.42 - 181.22 = 1.060

SSM = [(J2 + J2)/2D] - C = [(43.892  + 41.252 )/(2)(10)] - 181.22 = 0.175
         a    p
        D
SSD = [ I  (T1a + T1p)2]/4 - C = C(727.09)/4] - 181.22 = 0.553


SS (replicates) = [(Ka + Kp)/2] - C = [(193.29 + 170.76)/2] - 181.22 = 0.805

SSMD = SS (replicates) - SSD - SSM = 0.805 -  0.553 - 0.175 = 0.077

SSE = SST - SSD - SSM - SSMD = 1.060 - 0.553 - 0.175 - 0.077 = 0.255

MSM = SSM/1 = 0.175

MSD = SSD/(D-1) = 0.553/9 = U.061

MSMD = SSMD/(D-1) = 0.077/9 = 0.009

MSE = SSE/2D = 0.255/(2)(10) = 0.013
                                 ANOVA Table
Degrees
Source of Freedom
Method
Day
Method x Day
Error
1
9
9
20
Sum of
Squares
0.175
0.553
0.077
0.255
Mean
Square Error
0.175
0.061
0.009
0.013
      Total
39
1.060
                                                                     continued
                                     B-29

-------
                                                      OSWER Policy Directive
                                                            #9433.00-2
Example 2 (continued)


1.  Screening for Outliers

    Proposed Method:     Y = 2.063
                         TOT
                             = (MSB +  MSW)/2  =  (0.034  +  0.020)/2  = 0.027
                        (Y+.4.0 ^s2  )  =  (2.063^4.0  \To27)  = 1.406,2.720)



    Approved Method:     Y = 2.195


                        s2   = (MSB + MSW)/2  =  (0.036  +  0.006)/2 = 0.021
                         TOT


                        (Ti4.0 J^~)  =  (2.195^4.0  \/^021)  = 1.615,2.775)


    Since  the  observations  for  both  methods   fall   within   the   specified


    intervals, it is  concluded that this  data set  is free  of outliers.


2.  Testing Equality  of the Method Replicate  Precisions

                                                                   2
    A 95% CI  for the  ratio  of the  true   replicate  variances a  (proposed
             *)
method) and a  (approved method) is given by:
             a
Pr (
           MSWr
                             2
                            a
                          <  P  <
MSWr
              ) = .95
     MSWa(FN-D,N-D,.975)    o2     MSWa(FN_D>N.Dj.025T
                             a
where FN-D,N-D,.975 is tne 97.5th  percentile of the F-distribution  with  (N-D)

degrees of  freedom in both the numerator and denominator,  and FN_D , N-D,. 025  =

     1 _ .
FN-D,N-D,.975

     From Table Bl-1, Fg

                                = 3.72 and Fg>9>.o25 = 3.72  = 0.27
Pr (     U.02Q      <   P  <     0.020      )   = .95
     (0.006)(3.72)    ~^    (0.006H0.27)
Pr (0.90 1   P  1 12.35) = .95
                                                                     continued
                                     B-30

-------
                                                      OSWER Policy Directive
                                                            #9433.00-2
Example 2 (continued)
Since the  above interval  includes   the value  1,   corresponding to  a2 -  a2, it
                                                                   P    a
is concluded that  the  method  replicate   precisions   are   not   significantly
different.
3.  Testing the Method x Day Interaction
     Usiny a 5  percent significance  level  (a =  O.Ob), the  critical value for
testing the interaction term is

            FD-l,2D,l-a = F9,20,.y5 =  2.39        (from  Table Bl-1)
Since F =  MSMD s 0.009 s 0.692  is less than the  critical value  2.39, it is
           MSF~   (Ton
concluded that the interaction term is not  significant.
4.  Testing the Method Main Effect
     Since the interaction term was not significant, its  sum of squares  (SSMD)
and degrees of freedom  (D-l)  can be added  to   those for  the error,  and a new
mean square error (MSE*) can be calculated.

                MSE* = SSE + SSMD =  0.255  + 0.077  =   0.011
                       2D + (D-l)    (2)(10) + (9)
The  critical  value  for  testing   the main   method   effect  is  fl,3Q-lil-a =
Fl 29  95 = 4-la (from Table Bl-1).  Since  F = MSMD  = 0.175  =  15.91 is greater
  '  *'                                        MST*"  0.011
than or equal to the critical  value 4.18, it is concluded that the  two methods
are not equivalent, and, therefore, the proposed   method  would not  be accepted
by the EPA.

                                                                    continued
                                     8-31

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2
Example 2 (continued)
Conclusion:  Although  the method  replicate precisions were  determined to be

similar, the  average measurement for the proposed  method did not concur with

that for the approved method, as indicated by the method main effect.   There-

fore, the proposed method would not be accepted by the EPA.
                                     B-32

-------
                                                       OSWER  Policy  Directive
                                                             #9433.00-2
82.  EXPERIMENTAL DESIGNS AND METHODS OF DATA ANALYSIS FOR MULTISITE CASES
     If cost  permitted, extensive experiments involving many  sites,  samples,
and  laboratories  could  be  done  to  estimate  bias,    individual   variance
components and total measurement  system variance for the range  of conditions
over which  the method would be applied.   Unfortunately such experiments  tend
to be lengthy and economically unfeasible.   A more compact,  specially  designed
experiment may assure that relevant sources of test method variance are built
in (in the absolute case) or  controlled for (in the comparative  case), while
reducing the study  to a more manageable period  of time,  such as one  or two
weeks.   The designs  suggested in this document are  intended to  incorporate
realistic total system  variance rather than to resolve   individual components
thereof.
B2.1  Experimental Design Terminology
     Experiments are designed to investigate the effects of certain factors on
variables  of interest, dependent  variables, or  response variables (y).    In
this document y_ usually denotes either (1)  waste sample  recovery (measurement/
concentration) in  the absolute case,  or (2)  logarithm of  measurement  of a
given analyte in the comparative case.
     A factor  is a  variable, such  as  a  waste  site or  method  of  chemical
analysis.   A factor  is divided  into categories  for  the experiment  called
levels of the factor, e.g., 6C/MS, variations 1 and 2.
     The influence of factors on outcome or response are called effects.
     The mathematical description of an experiment to investigate  the effects
of ji factors on the variable of interest is called an n-way classification.
     A fixed effect  such  as a  quantitative analysis  method  (treatment) is
assumed  to be constant.   Random effects include  laboratory or analyst where
the lab or analyst is regarded as a randomly selected member of  a population.
                                     B-33

-------
                                                       OSWER  Policy  Directive
                                                             #9433.00-2
Further discussion of the difference can be found in Winer (3).   Combinations
of fixed and random effects are called mixed effects.
     The experimental situations for equivalency testing fall   largely outside
fixed  effects analysis  of  variance, where  interest centers   mainly   on  the
estimation  of  functions of  means.   Although  the  effects of  quantitative
analytical methods are  considered fixed (nonrandom), effects due to sampling
within sites are usually considered random, as are effects due   to differences
in day-to-day  setups of equipment and preparations.   Laboratory effects   are
often considered random, unless, for  example, only  a small, select group of
laboratories  can perform a test method.   For such  random effects, variances
(not means) are the statistical measure of primary interest.    Inference about
linear combinations of  variances leads  to complicated  formulas  about their
variances  (variances of variances).   The designs  proposed here then involve
mixed effects.
     A crossed (2-way) classification combines every level of one  factor with
every level of another factor.   For instance, each of a given  number of waste
samples might be split into four parts and tested with duplicated  analyses by
each of two test method procedures; in this case, the samples are crossed with
method of analysis, as in  the simple sinyle-site comparative case  of Section
4.
     Nested classifications  preclude the  formation of all  possible combina-
tions, by  pairing each level of one  factor with only one level  (rather than
all levels) of the other factor.   An example is waste samples  nested within a
waste  site, where  samples  are unique  to particular  hazardous  waste  site
(Figure 82-1).   With more  than two  factors, hybrids of  nested and crossed
classifications can occur.   Table B2-1  lists and describes several pertinent
factors that might be involved in the experimental designs.
                                     B-34

-------
                                                      OSWER Policy Directive
                                                            #9433.00-2

     a)  A Crossed Classification:   Test  Method  Crossed  With Waste Sample
Method
Sample
1
2
1
Yl,l
Y2,l
2
Yl,2
Y2,2
     b)  A Nested Classification:   Waste Sample Nested  Within  Waste  Site
Site 1
Sample 1
Yl,l
Sample 2
Yl,2
Site 2
Sample 3
Y2,3
Sample 4
Y2,4
        Figure B2-1.  Examples of crossed and nested classifications.
Legend. Y = test method's recovery of the hazardous waste property or con
            stltuent of interest.  In practice,  samples 3 and 4 are often
            referred to as samples 1 and 2,  i.e.,  the same labels are often
            used for samples from both sites, even though physically distinct
            sets of samples are involved.  In this case, ^2,3 and Y2,4 become
            Y2,l and Y2,2-
                                     B-35

-------
                                                       OSWER  Policy Directive
                                                              #9433.00-2
        TABLE B2-1.   SOME COMMON  FACTORS  IN  EXPERIMENTAL DESIGNS FOR
                             EQUIVALENCY  TESTING
  Factor
   Fixed or
Random Effect
  Crossed or
    Nested
Index
Number of
  Levels
  Method of
  Chemical
  Analysis

  Hazardous
  Waste Site

  Day

  Sample

  Concentration

  Laboratory


  Replicate
   Fixed
   Random
   or fixed

   Random

   Random

   Fixed

   Random
   or fixed

   Random
Crossing factor        m



Crossing factor        t


Crossing factor        d

Nested within site     s

Crossing factor        c

Crossing factor        1


Nested within cell     r
           M
           D

           S

           C

           L
Legend.   The indices are used as subscripts  for  model  terms  and  to designate
factors and  their combinations (interactions).    The  number   of levels  is  the
upper limit  on the index.   For instance,   for  a comparative  study there   are
usually two methods so m can be 1 or 2,  i.e., m  = 1, 2.   For a study with five
sites T = 5 and t runs from 1 to 5,  t =  1,  ...,5.   Notation  such as  Ymtcr is
used to denote the measurement using method m the £th  replicate of a sample at
concentration c from site t.
                                     B-36

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2
B2.2  Classification of Factors as Random or Fixed
     Factors  such as laboratory and  site are  usually regarded as   random in
order  that results may  be generalized  to arbitrary  laboratories   and  sites
rather  than  limited to  those  actually  employed  in  the  study.    If  the
petitioner does not mind the  conclusion being  restricted to those   sites  and
laboratories  used in the study, then  these may be treated as  fixed effects.
If the  laboratories in  the study  are  the only  ones capable  of   using  the
method,  then laboratory must be regarded  as fixed, since there is   no larger
reference population.   If the sites were selected out of  convenience, or  are
the only sites where the method is proposed for use, then site may  be treated
as a fixed effect.
     It is anticipated that most multisite and multilaboratory  petitions will
be from vendors of equipment  and professional associations, who will  want to
treat these  effects as random.   The statistical  analyses suggested here  are
for this  case, i.e.,  treat  laboratory and  site as  random,  while treating
method and concentration as fixed.
     The  absolute  and  comparative  cases  are  as  defined  in Section   4.
The  example  experimental  designs offered  here are  three-way factorial  de-
signs. The factors  are day, site, and concentration in the absolute case;  and
day, site, and method in the comparative case.  In both cases day and site  are
regarded as random and the third factor is regarded as fixed.  The analyses of
variance and tables of  expected mean  squares for other  three-way  factorials
(such as all factors fixed or two fixed, one random) can be found in Kleinbaum
and  Kupper  (4, p. 367).  For multilaboratory  petitions, the  design  can be
obtained  by  replacing  day  with  laboratory   in  either  the  absolute   or
comparative case.   Each laboratory should then analyze samples from different
                                     B-37

-------
                                                        OSWER  Policy  Directive
                                                              #9433.00-2
sites on different days, and the laboratories should not discuss  their results
or otherwise collude.

B2.3  Multisite Absolute Cases
     An important consideration in experimental  design  for equivalency testing
is whether  waste site is regarded as  a fixed or random effect.    If the  test
method is proposed  for use at a small,  select group of waste sites  - most  or
all of  which are  sampled for  the  equivalency data - then site  is  a  fixed
effect.   However, if the test method is planned for general use  at any  site,
then site should probably be modeled as a random effect.  This means  the  waste
sites  should  be  selected at  random.   The  statistical  treatment  in  the
following guidance assumes site is a random effect.
     Only one type of design is  discussed for multisite petition cases.   The
petitioner may propose others if financial, logistic, or  other considerations
dictate.   At least five sites  should be selected at random.    The example  of
Figure B2-2 involves taking thirty samples at random from within  each site and
spiking and analyzing  six per day  (two analyses at each of three concentra-
tions) over a period of five days.  The days do not  need to be consecutive.  In
fact, EPA  usually prefers days to be a week or more apart, involving separate
independent setups and preparations.   The recommended  spiking concentrations
are T/2, T, 3T/2 where T is the regulatory threshold for the analyte.

B2.4  Comparative Data Quality Objectives, Multisite Case
     This approach  compares  the  proposed  test  method  directly  with the
approved test  method.   Waste samples  are not  spiked in this  approach; the
dependent variable  is  the test  method measurement  (or  logarithm   thereof)
                                     8-38

-------
                                                           OSWER Policy Directive
                                                                #9433.00-2
Figure B2-2.  Experimental design for multisite absolute case:   concentra-
     tion (fixed effect) crossed with day and site (random effects).
      Expected Mean Squares for Multisite Absolute Case (3,  p.  203)
     Source of
     Variation
   d.f.
       Expected Mean Squares
 c  (concentration)
 s  (site)
 d  (day)
 cs
 cd
 sd
csd
witnin-cell (error)
   C-l
   S-l
   D-l
(C-D(D-l)
 CSD(R-l)
V+RV(csd)+RSV(cd)+RDV(cs)+RSDV(c)
V+CRV(sd)+CRDV(s)
V+CRV(sd)+CRSV(d)
V+RV(csd)+RDV(cs)
V+RV(csd)+RSV(cd)
V+CRV(sd)
V+RV(csd)
V
                                  8-39

-------
                                                      -  USWER  Policy  Directive
                                                              #9433.00-2

Figure B2-2 (continued)

                      Estimates of Variance Components


v       = MS(error)

v (csd) = [MS(csd) - MS(error)]/R

v (sd)  = [MS(sd) - MS(error)]/CR

v (cd)  = [MS(cd) - MS(csd)]/RS

v (cs)  = [MS(cs) - MS(csd)]/RD

v (d)   = [MS(d) - MS(sd)]/CRS

v (s)   = [MS(s) - MS(sd)]/CRD

v (c)   = [MS(c) - MS(cd) - MS(cs) + MS(csd)]/RSD

Figure B2-2  Legend:  Two replicate observations per cell.   Thirty samples  are
randomly  selected from each of five randomly selected sites,  with six  samples
per site analyzed per day (two at  each  concentration).  For   muHi laboratory
petitions  replace day  by  laboratory.  Each  laboratory  should  then  analyze
samples from different sites on different days.

               Outline of Analysis for Multisite Absolute Case

                    [Additional details are in Section B3]

1.  Screen for Outliers [Section B3.2]

2.  Check  for equality  of replicate  variances of  recoveries;   transform if
    necessary [Section B3.3]

3.  Prior to  testing  adequacy of  bias and  total system  variance, test  for
    model  simplification, beginning with third order interaction, then second
    order interaction, then main effects.

    Pooling of sum of  squares and deleting nonsignificant variance components
    for remaining tests is recommended.  In addition  to presumably increasing
    power  of  tests, this  can  lead  to  an  exact  test  of the  main  fixed
    concentration effect (an exact test does not exist for the full model;  see
    the table of  expected mean squares in  Figure B2-2).   Pooling is  done by
    combining  sums of squares for  nonsignificant terms with  the  error sum of
    squares and also  adding the corresponding degrees of freedom  to those  for
    error (Winer, 1962, p. 2U2).  The new  mean squared  error is  the ratio of
    the new  sum of  squares and the  new degrees of freedom.  The  significance
    level for preliminary  tests of these  secondary hypotheses should  be at a
    higher level (e.g., 20% to 30%) than tests of primary hypotheses.

4.  Model  validation:  Perform  model diagnostic  tests on  residuals  for  the
    simplified model, including a test of normality.  (5)

                                     B-40

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
Figure B2-2 (continued)
    Test  adequacy  of total   system variance:    Using  all  model  'terms  which
    remain after the preliminary  tests, i.e.,  were judged to  be  significant,
    estimate  the total system  variance by  estimating and  summing  the cor-
    responding variance components.   Use Satterthwaite's approximation (6,  p.
    369) to obtain a 95% lower confidence limit for total system variance.   If
    this  lower  limit is  greater  than  o2,  the adequacy  of precision   is
    rejected.   In some  cases, the improved approximation of  Welch should  be
    used (6, p. 370).  Additional  details are in Section 83.

    Test  method  unbiasedness:   In general, bias is  carried by   those  model
    terms involving only fixed effects.   In the present 3-way mixed model this
    is the grand mean  plus the  main concentration  effect;   therefore,  first
    for a main concentration effect.   Usually  the  second order (cd) interac-
    tion will  be nonsignificant  from step  3,  so that  an exact  test   of a
    concentration effect can be based on

                               F = MS(c)/MS(cs).

    If no exact test exists,  then a test can be based'on quasi-F ratios (Winer
    1962, p. 199).  If a difference among the concentrations exists, then each
    concentration must be tested  separately for  adequacy of bias.   Multiple
    comparisons procedure (6) may be  used to control the overall   error  rate.
    If no  difference exists, test the grand  mean   u i.e.,  test  H:  ju  -1)  <
                                     B-41

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
rather than the method's recovery (measurement/true value).  If  the petitioner
believes that  the proposed method  is less  biased than the  approved method,
then absolute criteria should be used to assess equivalency.
     If there are any common elements to the two test methods, then it is  best
to apply them concurrently (e.g., the same day) to a given waste sample.  This
proposed  experimental design  is  a type  of paired  comparison,   designed to
eliminate by blocking effects which  are not of primary interest.    Each waste
sample is split into four parts and two subsamples are tested by  each method,
thus controlling for sample effects by testing each sample by both methods.  A
waste sample should be tested  concurrently with  the two methods   to equalize
sample holding times.
    Since it is the objective of comparative DQOs to compare the  proposed and
approved  methods directly, waste  samples are  subsampled or  split  into two
parts to  control for differences among samples.   To approximate  the range of
measurements encountered  once a test  method is  in practice, one  varies the
types of  waste samples tested  by selecting  several samples  at   random  from
several randomly selected sites.   The randomization at all  levels of measure-
ment is  necessary if  the  petitioner  wants  to generalize  the   results  to
arbitrary waste sites.   The petitioner must be careful to obtain   representa-
tive  subsamples  or  splits of  waste.   As  a  further  precaution,  the two
subsamples should be randomized to the two test methods.  Figure B2-3 contains
an  example experimental  design with five  waste sites  and five  days  and an
outline of suggested analyses.   Additional details of analysis are in Section
B3.
                                     B-42

-------
                                                              OSWER Policy Directive
                                                                    #9433.00-2
Figure B2-3.  Experimental design for multisite comparative case:
       method (fixed effect) crossed with day and site (random effects)

       Expect Mean Squares for Multisite Comparative Case (3,  page 203)
Source of
Variation
m (method)
s (site)
d (day)
ms
md
sd
msd
within-cell
degrees of
freedom
1
S-l
0-1
S-l
D-l
(S-D(D-l)
(S-D(D-l)
SD(R-l)
Expected Mean Squares
V+RV(msd)+RSV(md)+RDV(ms)+RSDV(m)
V+2RV(sd)+2RDV(s)
V+2RV(sd)+2RSV(d)
V+RV(msd)+RDV(ms)
V+RV(msd)+RSV(md)
V + 2RV(sd)
V+RV(msd)
V
                                     B-43

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
Figure B2-3  (continued)
 Legend:   Two  replicate  observations per cell are recommended.   Five samples
 are  randomly   selected from each  of five  randomly selected  sites  with four
 representative subsamples extracted from each sample.   On each of five days,
 one  sample  from each site is analyzed,  two subsamples  by each method.   The
 design  over   sites  and  days  hopefully  incorporates  realistic  spread  of
 experimental conditions  representative  of the  proposed use  of  the method.
 Within each condition (site x day cell) for each method, replicate variance is
 calculated and these  are averaged over the cells.   The average within-method
 replicate variances are compared by a standard F-test.   The comparison of the
 two  methods is completed by exploiting the pairing.   For comparative studies,
 the  criterion  is usually that  the methods be essentially equivalent,  so that
 each model term involving method should be nonsignificant.

"Outline of Analysis for Multisite Comparative Case

 1.   Screen for outliers [Section B3.2].

 2.   Check  for equality of  replicate  variances  of  measurements  or  their
     logarithms; transform if necessary [Section B3.3].

 3.   Precision  check:   Compare replicate  variances for the two methods.   The
     replicate  variance  estimates  are  the  average  within-cell  variances,
     averaged over  the S x D cells,  with  degrees of freedom DS(R-l).   These
     can be  obtained as the  residual mean   squared errors from  fitting full
     (i.e., include  both main effects, site and day, and  interaction) two-way
     ANOVA models separately to each method.  The null hypothesis is

                            H:   02 /  0|l 1

     where "1"  indexes the proposed method and "2" indexes the approved method.
     The test statistic is              -   -
                                        F21 2
                                     = 0/02


     with DS(R-l), DS(R-l) degrees of freedom.

 4.   Other tests:   Every model term involving  method should be nonsignificant
     in order for  the methods  to be equivalent.   Starting with  the full 3rd
     order model, test the significance of third order interaction, then site *
     day interaction, then second order interactions involving method, then the
     main method effect.  Pooling of sum of squares and deleting (setting equal
     to  zero)  nonsignificant  variance  components  for  remaining  tests  is
     recommended.   In addition  to presumably increasing power of  tests, this
     can lead to  an exact test of the  main method effect (an exact  test does
     not exist  for the full model).  Pooling is done by combining (adding) sums
     of squares for nonsignificant terms with the error sum of squares and also
     adding the corresponding degrees of  freedom to  those for  error  (3, p.
     202).   The new  mean squared error is just  the ratio  of the new  sum of
     squares and the new degrees of freedom.   If at any stage a term involving
     method is  significant, the testing  can stop  with the conclusion  that a
     method related effect exists.

 5.   Model validation:   Perform diagnostic  tests on residuals  for the  final
     model including a normality test on residuals (5).

                                      B-44

-------
                                                        OSWER Policy Directive
                                                              #9433.UO-2
B2.5  Petitions Based on Quality Control  Data
        It is possible to characterize a  test method partially from evaluation
of relevant QC data taken  from the method's routine day-to-day  use, provided
that the waste samples tested are comparable to those wastes  being petitioned
for  testing.   However, it  must be  acknowledged that  QC  test data  can be
fraught with complexities, due to  the likely  effects of sitee,  waste sample,
lab technician, and so forth.   The QC data will, in most cases,  not exhibit
the symmetries and sample  size balances   possible in a  specifically designed
experiment.   The QC data experimental design may change  from day-to-day with
the number of routine analyses.   Also, site effects can take very complicated
forms, e.g., all tests performed one  week could involve only one  site, while
the following week's  tests might involve three sites.   The QC  program might
call for  spiking samples  at  low, medium,  and high  concentrations  in  the
working  range of the test method.   On a  slow day with only a  few scheduled
routine tests, perhaps only one or two QC spiked concentrations would actually
be used (unless the QC program  demands all three concentrations on  every day
of operation), resulting in empty cells on some days.
     In summary,  proper  statistical treatment  of QC  data  requires keeping
track of site, sample, spiked  concentrations and day of analysis.   Sample is
nested in site and probably in  day, while site and spiking  concentration are
crossed with  day.   To ignore  all these  effects and throw  the data  into a
one-way analysis of variance to estimate  between-day and within-day components
of  variance  is  risky  at  best.   Proper  analysis  of  such  data requires
statistical expertise.
                                      8-45

-------
                                                        OSWER Policy  Directive
                                                              #9433.00-2
B3.  STATISTICAL SUMMARIES AND ANALYSIS
     This section  further  details  steps  of the  statistical   analyses  for
multisite cases which were outlined in Figures B2-2 and B2-3.   The single-site
cases were treated in detail  in Section 4 of the main text.
     In the multisite case, if  results will  be generalized beyond  the sites
and days  employed for  the  experimental design,  these two  factors  must be
considered random. Adding a fixed factor (concentration or method) leads  to a
3-way mixed  model with two  random and  one fixed  factor.    The accompanying
statistical  analysis  can  be  complicated  and  is  performed  best  with  a
computerized statistical  package  such  as  SAS, applied  by  an  experienced
statistician.  A pertinent discussion of this design and its analysis is given
by Winer (3, pp. 202-207).
     The petitioner does have  design options  which lead to  simpler analysis
but were not treated in detail here because they are regarded  as economically
or logistically  unfeasible.    If samples  can be  obtained from  25  randomly
selected  sites and  analyzed on 25  separate days,  then the  site  factor is
confounded with the day factor and simpler statistics apply.   In fact, if the
concentrations are in an  adequately controlled  (e.g., narrow)  range  for the
test method, then preliminary tests for a concentration effect in the absolute
case may  indicate that  the  three concentrations  can be  regarded  as  just
replicates and the data conform to a one-way random effects ANOVA design.
                                     B-46

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2

B3.1  'Basic Approach

     The basic approach to statistical  analysis of test data is to —

     •  screen for outliers,
     •  check constancy of the waste samples'  replicate precision,
     •  transform test data (if necessary),
     •  fit the full statistical model,
     •  test for model simplifications,
     •  perform diagnostic checks on the simplified model,
     •  take appropriate remedial action if necessary, and
     •  use the  final selected  model  to test  hypotheses  about precision and
        bias.

The  final testing  of adequacy of  precision  and  bias may ce  approached as

follows:

     For absolute data quality criteria, the petitioner estt matrf  total  system

variance by  adding estimates of  each  component  of variance that  involves a

random  factor  and  is  judged  by  preliminary  tests  to   be  significant.

Satterthwaite's chi-square approximation (described below)  is used hTO test the

hypothesis  of  adequacy  of  total  system  variance.  Next,  the  petitioner

estimates  system  biases  as y  +  etc,  i.e.,  the means  for  the  different

concentrations.   If  preliminary  tests indicate  there is  no  concentration

effect, then a test on \t alone is appropriate.

     In the comparative case,  any statistical model terms  involving  the test

method  factor should not  be significantly  different than  zero.   Also, the

replication variances  for the  new  method  should  not exceed  that  of  the

proposed method.

     If  the experimental  design's  test data  do not  support  a  simplified

statistical  model,  then the  computations become  involved.   It  is  highly

recommended that a computerized statistical  package such  as SAS be  used for

computations, and that a statistician be directly involved with the design and

responsible for the statistical analysis of the experiment.
                                     B-47

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
B3.2  Approach to Outliers and Other Model Violations
     Outliers present  many problems in statistical  analysis.   Their presence
can  seriously  disturb  conventional   least squares  and  maximum  likelihood
analyses  such as ANOVA.   One would  like to delete outliers or  replace them
with  suitable values  before  statistical analysis   so that  the  analysis is
undisturbed. However, it is really  only relative  to a reasonable  model  that
outliers can be determined, i.e., they are best determined after the  model is
fit. To  break this circle,  it is  proposed that  preliminary  outlier checks
based on  empirical  univariate  and  bivariate distributions  (e.g.,  in  the
absolute case, recovery and  recovery  versus  concentration, site and  day) be
used and that a check be applied  to residuals from the final  selected model,
at which time normality and other diagnostic tests are also applied.
     The petitioner  is free  to  suggest  other  approaches such  as  use  of
Thompson's t-test or the Dixon ratio test.   EPA does, however, usually prefer
outlier tests which are 2-tailed at the 1 percent significance level  or less.
The suggested approach involves inspection of distributions of  recoveries for
each  designed factor  (from  day, site,  concentration, method)  as  well  as
recoveries pooled  over all conditions.   Any data  point more than 4 standard
deviations  from the mean is suspect.    Rather than  delete an outlier observa-
tion, it is preferable to replace  it  with a value regarded  as representative
for the corresponding experimental condition, so as  to preserve the balance of
the design.   In no  case should  all  values for  a  given  condition (cell) be
deleted because of  one outlier for the cell.   In any case, the raw data must
be reported  with clear indication  of what  points  were deemed  outliers, how
they were detected, and how they were  treated.
     For inputing values to replace rejected outliers, it is  recommended that
the additive (no interactions) model be fitted to the acceptable data and that

                                     B-48

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
order  statistics  appropriate  to that  model  be  used.  For instance,  if the
predicted value and estimated residual  error variance from the fitted additive
           ^
model  are y and s,  respectively,  and  if  replicates  (R)   were used  in the
                                                  A
design,  then  to  replace  a  high  outlier  use y + s/-*/~JT.   To replace a low
              /s
outlier, use (y - s
B3.3  Approach to Equal Variance Assumption
        A standard assumption for ANOVA procedures is that the error variances
or replication variances  are equal  under all  conditions.   Each experimental
design recommended  suggests  two replicate  measurements in  order  that this
assumption,  as  well  as  hypotheses  about  higher  order  interactions,  be
testable.  More than two replicates is not recommended.  If the experiment can
be expanded, it is generally more informative to add sites or days.
        If the waste samples  contain concentrations in the working  range for
the test  method, then it will often  be the  case that standard  deviation is
proportional to concentration.   If this  is so,  then one ba nefit  of working
with  recoveries in  the  absolute cases  and logarithms  of  measurements  in
comparative cases is that  variances will be approximately equal.   Therefore,
it  is recommended that  initial statistical  analyses use  recoveries  in the
absolute case and logarithms of  measurements in  the comparative case.   Some
other transformation  may be  used  if needed  to attain  normality  or  equal
variances.
     To test  for equality  of replicate  variances, calculate the within-cell
average (7-jsc|) and standard  deviation (Sisc|) (of  recoveries in  the absolute
case; of  logarithms  of measurements  in the comparative case) for each cell.
Plot log (Siscj) vs. log  (7isd) and use ordinary linear regression
                          log (Sisd) = a log (yisd) + b
                                     6-49

-------
                                                        OSWER  Policy  Directive
                                                              #9433.00-2

to  test  for  dependence  of  variation  on  level.    If   the  regression   is

significant, then try the power tranformation yl~a (log(y)  for a=l) on  the  raw

data and test again as above for equality of variances  (7,  page 232).

        Normality will  be tested after  the model   has  been fit  and possibly

simplified as part of the diagnostic tests on residuals.



83.4  Analysis for Multisite Comparative Case

            B3.4.1  Outlier Screen.  For each separate test  method,  petitioners

should study the distributions of logarithms of measurements by day,  by  site,

and pooled  over all conditions.   Any datum  more than  4  standard deviations

from the average is a candidate for an outlier.

            B3.4.2   Checks on Replicate Variances.  Waste sample  replicate

variances are  assumed to be approximately equal  in each  day  * site  cell  and

for each test method.  For most test methods, the logarithm of measurement  (or

a  power  near zero,  e.g.,  yO»l)  will  be approximately   normal  and  homo-

scedastic.

            B3.4.3  Comparison of Method Replicate Precisions.  The  two  van'-
                A       A
ance estimates, a2,  and 
-------
                                                        OSWER  Policy Directive
                                                              #9433.00-2
where "1" indexes the proposed test method,  and "2"  indexes  the   approved test
method.  If this hypothesis is rejected,  no  further  testing  is necessary.
            B3.4.4   Comparison of Method Biases.  An   ANOVA  table  for  this
experiment is  in Figure B2-3.   Each statistical  model  term  involving method
should be nonsignificant.   Otherwise, some   method-related  effect  appears to
exist.
     By  sequentially  testing  for  3rd   order interaction,  then  2nd  order
interactions, then main effects, it is possible to obtain exact  tests  for each
hypothesis.   (No exact  test of  the main method  effect exists in the  full
model.)  That is, if a term  involving method is significant,  the testing can
stop, with the conclusion that a method-related effect  exists.   Otherwise the
testing  continues  with  the  nonsignificant  interactions  neglected (i.e.,
assumed equal to zero) in  lower order terms which originally  contained them.
The sums  of squares for  nonsignificant  interactions   can be  pooled   with the
error sums of squares after each stage.
     The following sequence can be used.   Here pooling  is used only before the
test of the main method effect.
     To test significance of 3rd order interaction,  use F =  MS(MSD)/MS(error),
with  (S-1)(D-1), SD(R-l) degrees of freedom.   In most  cases,  this 3rd order
term will be not significant and the associated  a2     may be  crossed  out from
                                          •        MSD             .
the expected mean squares  for other sources of variation.  If  this 3rd order
interaction term is significant, then further testing is not necessary.
     Petitioners  should statistically test 2nd order   interactions involving
method by using—
       FI = MS(MS)/MS(MSD) with S-l, and  (S-D(D-l)  degrees  of freedom
       F£ = MS(MD)/MS(MSD) with D-l, and  (S-1)(D-1)  degrees  of freedom.
                                     B-51

-------
                                                        OSWER Policy Directive
                                                              #9433.00-2
If any 2nd and 3rd order interactions involving method are significant, then  a
method-related  effect  is  assumed to  exist.    If  these  2nd and  3rd  order
interaction tests are not significant, then the main method effect   remains to
be tested.   A pooling procedure is recommended for this to increase the  power
of the test potentially.
     The petitioner  should test the  main method  effect using  F   = MS(M)/MS
(error,pooled), with 1, 2SD(R-1) + SD - 1 degrees of freedom.
            83.4.5  Diagnostic Tests. Separately for each method and combining
data from the two methods,  petitioners should   plot residuals versus  day  and
versus site  and pooled  over all conditions.    Also, petitioners  should plot
residuals versus fitted values, then inspect for trends and outliers.  A test
for  normality such  as the Shapiro-Wilks  test should  be used on   the pooled
residuals. These graphic displays  and diagnostic tests should be  included in
the petition to EPA.
            B3.4.6   Conclusions.   If the replicate variance of the  new test
method  is significantly larger than that  of the approved test method,   or if
any significant method-related effect is  found, then  the new test  method is
rejected.

B3.5  Analysis For Multlsite Absolute Case
            B3.5.1   Outlier Screen. Petitioners  should look at the distribu-
tions  of recoveries by day,  by site,  by concentration, and pooled  over  all
conditions.   Any datum more than 4 standard deviations  from the average is  a
candidate for an outlier.
                                     B-52

-------
                                                        USWER Policy Directive
                                                              #9433.00-2
            83. 5. 2   Checks on Replicate Variances.    Waste  sample  replicate
variances  are  assumed  to be  approximately  equal   in  each day  *  site  *
concentration cell  and for  each test  method.    For most  test methods,   the
recoveries  or their  logarithms or  a  power near  zero,  e.g.,  y0-1  will  be
approximately normal and homoscedastic.
            83. 5. 3   Preliminary Tests of Simplifying Hypotheses. These can be
constructed from the table of expected mean squares,  proceeding from bottom to
top  to  form  F-ratios, e.g.,  third  order  interaction   is tested  by  F  =
MS(csd)/MS(within-cell),  with   (C-1)(S-1)(D-1)   and   CSD(R-l)   degrees  of
freedom.
            B3.5.4  Testing Adequacy of Total System Variance.   The basic re-
sult for combining mean squares is that if —
                           n-jx-j/a?, i = l,2,*..,fc

(6, p. 369) are independent  chi-squared variables with n-j  degrees  of freedom,
then --
                                   u = ng/r
is approximately  chi-squared with n degrees of freedom, where --
                                  r  -  u^.
                                  g  =  EgiXi,  and
                                  n  =  (Z9 a

In applying this, the  of are  replaced  by  their  estimators in the  formula
for n.
                                      B-53

-------
                                                       OSWER  Policy  Directive
                                                             #9433.00-2
                 NOTATION FOR SINGLE-SITE CASES OF  SECTION  Bl
     Symbols subscripted by "a" or "p"   have the  meanings   defined  below,  for

the approved (old) or proposed (new)  method, respectively.

Probability Distributions

    tdf.a       = tne 100(cx)th  percentile of a  _t  distribution  with  df  degrees
                  of freedom

    X2df,a      = tne 100(a)th  percentile  of  a  chi-square distribution with
                  df degrees of freedom

    Fdfi.dfj.a  = tne  100(o)th  percentile  of  an   F   distribution with  df,
                  numerator, df  denominator degrees  of freedom.


Population Parameters

    a           = variance = (standard  deviation)2

    u           = mean

    b           = bias =  difference  between the  mean  (expected  value)  of  the
                  measurement  system under  fixed conditions  and   the true
                  value.

    ao,bo       = data  quality   objectives,  i.e.,  desired  bounds  on   the
                  standard deviation  and bias of the  proposed method.

Sample Statistics for One-Hay ANOVAS

      1    2
     Y. , Y.    = duplicate  observations  (e.g.,   recoveries  or log measure-
      1    i      ments) on  ith day  for  a particular  concentration  in   the
                  absolute case or method in the comparative case

     D          = number of days

     N          = total number of observations

     Yij« sij   = tne  average  and  standard deviation of  the (usually, two)
                  observations in the (i,j) cell

     MSW, MSB   = within-day and between-day  mean squares for  one-way ANOVA
                  designs with groups defined by different days
      2
     s    = g   = estimate of  total  variance of  a   single  observation under
      TOT         the  model at hand  (referred to  in  some formulae as "g" for
                  compactness);  for  duplicate  observations  in  each cell
                  s    is the simple  average of  MSB and MSW.
                                      B-54

-------
                                                        OSWER  Policy Directive
                                                              #9433.00-2
                                  REFERENCES
.1.  Standard  Practice  for  Dealing  with  Outlying  Observations.  ANSI/ASTM
    E178-80.

2.  Box GEP,  Hunter WG, Hunter JS.  Statistics   for Experimenters:  An  Intro-
    duction to Design, Data Analysis and Model Building.     Wiley,   New   York.
    1978.

3.  Winer,  B.J.   Statistical Principles in Experimental  Design. McGraw-Hill,
    New York.  1962.

4.  Kleinbaum, D. G.  and Kupper, L. L.  Applied Regression  Analysis.  Duxbury
    Dress, North Scituate, Mass.  1978.

5.  Draper, N. R.  and Smith, H.  Applied Regression Analysis.  John Wiley  and
    Sons, New York.  1966.

6.  Graybill, F.A.   An Introduction Linear Statistical Models.   McGraw-Hill,
    New York.  1961.
                                      B-55

-------
                                        USWER Policy Directive
                                              #9433.00-2
                  APPENDIX C

             EXCERPT FROM SW-846:

SECTION TEN, QUALITY CONTROL/QUALITY ASSURANCE

-------
                                                               OSWER Policy Directive

                                 APPENDIX C
                                SECTION TEN

                    QUALITY CONTROL/QUALITY ASSURANCE


     Section 10.1 defines Quality Control  (QC) and Quality Assurance (QA).
Section 10.2 discusses how QC/QA procedures can be used to ensure achievement
of program goals.  The various QC/QA aspects of sampling are discussed in
Section 10.1.3 while Section 10.1.4 discusses and lists appropriate laboratory
QC/QA activities.  Section 10.1.5 discusses the criteria with which acceptable
data must comply and methods of data evaluation.


1U.1  Introduction

     Quality assurance (QA) is a system for ensuring that all information,
data, and resulting decisions compiled under a specific task are technically
sound, statistically valid, and properly documented.  Quality control  is the
mechanism through which quality assurance  achieves its goals.Qua)ity
control  programs define the frequency and  methods of checks, audits, and
reviews necessary to identify problems and dictate corrective action,  thus
verifying product quality.

     The soundness of an organization's QC/QA program has a direct bearing on
the integrity of its sampling and laboratory work.  Results of sampling or
analysis conducted without adequate quality control and assurance may  be
deemed unacceptable for RCRA evaluation purposes.  The following section
discusses some minimum standards for QC/QA programs.  Generators who are
choosing contractors to perform sampling or analytical work should make their
choice only after evaluating the contractor's QC/QA program against the
procedures presented in these sections.  Likewise, contractors that currently
sample and/or analyze solid wastes should  similarly evaluate their QC/QA
programs.                                                               •


10.2  Program Design

     The initial step for any sampling or  analytical work should be to
strictly define the program goals.  Once the goals have been defined,  a
program must be designed that will meet these program goals. .QC and QA
measures will be the mechanisms used to monitor the program and to ensure
that all data generated are suitable for their intended use.  A knowledgeable
person who is not directly involved in the sampling or analysis must be
assigned the responsibility of ensuring that the QC/QA measures are properly
employed.

     As  a  minimum, a proper QC/QA program would include the following:

      1.  The intended use(s) for the data, and the necessary level of
          precision and accuracy of the data for these intended uses.
                                     C-l

-------
                                                               OSWER Policy Directive
                                                                    #9433.00-2
      2.  A  representative  sampling plan that  includes provisions for:
          -  selecting  appropriate  sampling locations, depths, etc.
          -  providing  a  statistically sufficient number of sampling sites.
          -  measuring  all necessary ancillary  data.
          -  determining  climatic flow or other conditions under which
             sampling should be conducted.
          -  determining  which media are to be  sampled (e.g., wastewater,
             sediment,  effluent, soil).
          -  determining  which parameters are to be measured  (and where).
          -  selecting  appropriate  sample containers.
          -  selecting  the.frequency of sampling and length of sampling
             period.
          -  selecting  the types of sample (e.g., composites  vs. grabs) to be
             collected.
          -  sample preservation.
          -  chain-of-custody.
     3.  An.analytical plan that includes:
          -  chain-of-custody procedures.
          -  appropriate  sample preparation methods.
          -  appropriate  analytical methods.
          -  appropriate  calibration and analytical procedures.
          -  data handling,  review and reporting.
     4.  Planning for the inclusion of proper  and sufficient QC/QA activities,
         including the use  of QC samples throughout all  phases of the study
         to  ensure that -the  level  of quality of the data will meet the
         requirements of the intended use(s) of the data.
     All program details should be put in writing and assignments made to
appropriate personnel.
     If the above procedures are followed (i.e., an appropriate program is
designed, tasks are assigned to knowledgeable  personnel, and sufficient OC/OA
                                    c-2

-------
                                                                  OSWER Policy Directive
                                                                       #9433.00-2
steps are employed), the program should meet and possibly surpass its goals
in most cases; at worst the failure to meet the program goals will be detected
and the usefulness of any data will be quantified.


10.3  Sampling

     The quality of a sampling program.has a direct bearing on the legal,
physical, and chemical integrity of the samples.  If the representativeness
of the samples cannot be verified due to inadequate attention to sampling
procedures, then the usefulness of the analytical data will be limited,
regardless of the refinement of the analytical program.  It is imperative,
therefore, that no analytical program be conducted without an adequate
sampling plan which does or will document the degree of representativeness
of the parameters of interest.

     10.3.1  Design of a Sampling .Plan

     Section One of this manual discusses the considerations involved in
designing a representative sampling plan.  For each specific project, a
sampling plan should be designed prior to commencement of sampling.  If the
plan addresses the considerations discussed in Section One, then the resulting
samples should be representative of the waste of interest and therefore
suitable for evaluation of the waste according to RCRA criteria.


     10.3.2  Sample Collection

     A variety of different sampling devices are used in sampling depending
on the type of sample (solid, liquid, multiphased), the type of sample
container, and the sampling location.  Section One and portions of Section
Three of this manual describe different devices that are available.  The
appropriate sampling device must be selected and its use supervised by a
person thoroughly familiar with both the sampling and analytical requirements.
This familiarity is essential since (1) certain sampling devices are made of
materials that may contaminate samples, (2) cross contamination of samples
can occur if the sampling device is not cleaned properly, (3) routine sampling
methods may not be applicable when the waste is to be analyzed for a different
parameter (e.g., volatile organic compounds), and (3) the method of employing
the sampling devices may affect the integrity of the sample.


     10.3.3  Sample Preservation

     Some form of preservation is usually required for all samples.  The type of
sample preservation required will vary depending on the sample type and the
parameter to be measured.  Therefore, more than one container of the same
waste may be necessary if the waste is to be analyzed for more than one
parameter type.

                                    C-3

-------
                                                              OSWER Policy Directive
                                                                    #9433.00-2
     The analytical methods included ii tiris manual refer to the optimum
means of preservation.  Since the chemical make-up of certain samples can
alter the effectiveness of preservation measures, all sample analyses should
be performed as soon as possible after sampling and before any recommended
holding time has expired.


     10.3.4  Chain of Custody

     Although chain-of-custody procedures may not be required in all cases,
they often are an essential part of sampling/analytical schemes since these
procedures can document the history of samples.  Chain of custody establishes
the documentation and control necessary to identify and trace a sample from
sample collection to final analysis.  Such documentation includes labeling to
prevent mix-up, container seals to detect unauthorized tampering with
contents of the sample containers, secure custody, and the necessary records
to support potential litigation."

     A sample is considered to be under a person's custody if (1) it is in the
person's physical possession, (2) in view of the person, (3) secured by
that person so that no one can tamper with the sample, or (4) secured by that
person in an area that is restricted to authorized personnel.

     Refer to Section One for details of how to implement chain-of-custody
procedures.


10.4  Analysis

     An analytical program defines standard operating procedures to be used1
in waste analysis, appropriate QC/QA procedures, means for detecting out-of-
control situations, and remedial actions.  -A separate analytical program
should be developed for each different waste to be analyzed.  The program
should be thoroughly specified before sampling is begun, since the analytical
procedures to be used may affect the choice of sampling devices and procedures.

     The program should select methods that will provide data at the level of
accuracy and precision that will be required by users of the data for decision-
making purposes under RCRA.  Once the appropriate method(s) have been selected
it is imperative that the accuracy and precision of all analytical data be
thoroughly documented by means of a well-designed QC/QA program.

     Laboratory QC/QA activities normally include:

      1.  Use of EPA-acceptable sample preparation and analytical methods.

      2.  Calibration of laboratory instruments to within acceptable limits
          according to EPA or manufacturer's specifications before, after,
          and during (as acceptable) use.  Reference standards must be used
          when necessary.

                                    C-4

-------
                                                             OSWER Policy Directive
                                                                   #9433.00-2
      3.  Periodic inspection, maintenance, and servicing (as necessary) of
          all laboratory instruments and equipment.

      4.  The use of reference standards and QC samples (e.g., checks,
          spikes, laboratory blanks, duplicates, splits) as necessary to
          determine the accuracy and precision of procedures, instruments,
          and operators.

      5.  The use of adequate statistical procedures (e.g., QC charts) to
          monitor the precision and accuracy of the data and to establish
          acceptable limits.

      6.  A continuous review of results to identify and correct problems
        -  within'the measurement system (e.g., instrumentation problems,
          inadequate operator training, inaccurate measurement methodologies).

      7.  Documenting the performance of systems and operators.

      8.  Regular participation in external laboratory evaluations (including
          the EPA Performance Audit Program) to determine the accuracy and
          overall performance of the laboratory.  This should include performance
          evaluation and interlaboratory comparison studies, and formal
          field unit/laboratory evaluations and inspections.

      9.  Use of acceptable sample identification and, as necessary,  formal
          chain-of-custody procedures in the laboratory.

     10.  Maintenance and storage of complete records, charts, and logs of
          all pertinent laboratory calibration, analytical, and QC activities
          and.data.

     11.  Ensuring all data outputs are presented in their prescribed format.

     Specific Quality Control measures for each method can be found by
referring to the individual analytical methods included in this manual.


10.5  Data Handling

     The quality of all data must be assessed before the data are used.
Assessment should focus on five basic points.

      1.  Accuracy - Can the data's accuracy be determined, and is it
          acceptable for the planned use?  QC/QA procedures will be designed
          to measure the accuracy of all analytical data.

      2.  Precision - Can the data's precision be determined, and is  it
          acceptable for the planned use?  QC/QA should demonstrate the
          reproducibility of the measurement process.


                                   C-5

-------
                                                           OSWER Policy Directive
                                                                 #9433.00-2
      3.  Completeness - Are a sufficient amount of data available for the
          planned use?  QC/QA shall identify the quantity of data needed to
          meet the program goals.

      4.  Representativeness - How well do the data represent actual
          conditions at the sampling location, considering the original
          study design, sampling methods, analytical methods, etc., which
          were used?

      5.  Comparability - How comparable are data with respect to several
          factors, including:

          - consistency of reporting units?
          - standardized siting, sampling, and methods of analysis?
          - standardized data format?

     All these factors must be considered when designing a study, and QC/QA
procedures must specify a reviewing process for all data.

     Statistical procedures applicable to data evaluation include:

     1.  Central tendency and dispersion

         - Arithmetic mean
         - Range
      «  - Standard deviation
         - Relative standard deviation
         - Pooled standard deviation
         - Geometric mean

     2.  Measures of variability

         - Accuracy
         • Bias
         - Precision; within laboratory and between laboratories

     3.  Significance test

         - u-test
         - t-test
         - F-test
         - Chi-square test

     Specific data handling precautions are noted in the individual methods
described in this manual.
                                »U.S.  GOVERNMENT PRINTING OFFICEi I9«2-16 I•012/3I 9
                               C-6

-------
                                     OSWER Policy Directive
                                           .#9433.00-2
                APPENDIX D


EXAMPLE TEST METHOD EQUIVALENCY PETITION

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2

                                  APPENDIX D

                  EXAMPLE TEST METHOD EQUIVALENCY PETITION


     The following example represents a test method recently approved by EPA.
The original report submitted to  EPA1 has  been modified and  supplemented to
meet guidelines on petition format and content that were not available when it
was first submitted. Fictitious information  has been added in some  areas and
is identified by footnotes; other minor changes have been made for clarity and
for conformity to guidelines.

     This example demonstrates  the format and general  issues  appropriate for
an equivalency  petition.   It  does  not include  an experimental  design  or
statistical analyses  of test data  that the  petitioner must  submit.   These
analyses are often unique to the petition; therefore, analyses are excluded so
as  not to  mislead the  petitioner.  Section 4  and Appendix  B  provide  the
guidance needed  to establish  experimental designs  and  statistical analysis
procedures.
1 A Laboratory Comparison Between EPA Method 450.1 [also referred to as Method
  9020] and  Haloscan TM for  the Determination  of Total  Organic  Halogen in
  Groundwater,  Surface Water, and Waste Treatment Plant Effluents.   Prepared
  by  Richard  A. Cope, Ph.D.,  for Environmental  Research Group,  Inc.   Ann
  Arbor, Michigan.  1982.
                                     D-l

-------
                                                      jp,
                                               *OSWER Policy Directive
                                                       #94,33*00-2
                 TEST METHOD EQUIVALENCY PETITION

                           -TO-MEASURE
                        9

   TOTAL ORGANIC HALID£S^(TOX) BY NEUTRON. ACTIVIATION ANALYSIS
                     Petition Submitted By --


                ENVIRONMENTAL RESEARCH GROUP, INC.

                      117 North First Street

                    Ann Arbor, Michigan  48104
[Petitioner should provide name and telephone number* of a contact]
                              CDate]
                               D-2

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2
                                 INTRODUCTION
     There is currently one existing published procedure for the determination
of organic  halogens (TOX) in water samples:   EPA Method  9020.   Any time an
analytical method becomes accepted as  an approved method, it is  desirable to
provide  an  alternative  or  additional  analytical  method,  which  is  also
approved, that can be substituted as equivalent to the first method.   This is
especially true if the first  method specifies a specific brand  of instrumen-
tation such as the Dohrmann DX 20, which is specified in Method 9020.

     This petition presents an alternative method to 9020 for the  analysis of
total-' organ-ic halogens (TOX) using neutron activiation analysis.  ERG believes
the data show that-this-method  produces equivalent data to Method  9020 while
providing additional information  on the  organic halogen distribution  of the
sample.  ERG requests that this method, called HaloscanTMj be evaluated by the
EPA  and, if  found.adequate, be  accepted as  an approved  procedure  for the
screening of groundwater,  surface water,,  and waste treatment  plant effluent
for TOX.                  .                    :.;  •
              1.0  CERTIFICATION OF ACCURACY-AND..RESPONSIBILITY


          I certify under penalty of  law that I have personally  examined and
am  familiar with  the  information submitted^...!'n this., demonstration  and all
attached  documents  and  that  based  on  my  inquiry" of- .those  individuals
immediately responsible  for  obtaining the  information, I  believe  that the
submitted information is true, accurate, and complete.   I am aware that there
are  significant  penalties for  subrouting false  information  including  the
possibility of fine and imprisonment.,
                         ,"""'-'-  -        Signed,
                " •'»       ''*••• 1* - '
                                         mre-_

                                         Date
                   ;  2.4  DESCRIPTION OF PROPOSED ACTION
2.1  TEST METHOD DESCRIPTION                -    -J ._;      .,,,.-

     The  proposed Haloscan^M test  method is  presented below  in  the format
found  in  EPA's   "Test  Methods   for  Evaluating  Solid   Waste"  (SW-846).
                                     D-3

-------
                                                       OSWER  Policy Directive
                                                             #9433.00-2
                             PROPOSED METHOD
         TOTAL ORGANIC HALIDES (TQX)  BY NEUTRON ACTIVATION ANALYSIS

1.0  Scope and Application

     1.1   This  method  determines Total   Oryanic Halides  (TOX)   in drinking
waters, groundwaters,  and  river waters,   and in  wastewater  treatment  plant
effluents.  The method uses a carbon adsorption procedure identical to that  of
Method  9020  (TOX  analysis  using  a  microcoulometric-titration  detector),
irradiation  by neutron  bombardment,  and then  detection using  a  yamma  ray
detector.   The reliable limits of detection are 5 ppb for chlorine and 1  ppb
for iodine and bromine.

     1.2   This  method  detects  all  organic  halides  containing  chlorine,
bromine, iodine, and fluorine  that are adsorbed by granular  activated carbon
under  the  conditions  of  the  method.   Each  halogen  can  be   quantitated
independently.

     1.3   This method  is restricted to use by,  or under the supervision  of,
analysts experienced  in  the operation  of neutron  activation analysis  and
familiar with spectral interferences.

     1.4   This  method  may be  used in  place  of Method  9020 and  has  the
advantage  of  determining  the  individual  concentrations  of the  halogens
chlorine, bromine, and iodine in addition to TOX.


2.0  Summary of Method

     2.1   A sample  of water  that has  been  protected against  the loss   of
volatiles by the elimination of headspace in the sampling container,  and that
is free of undissolved solids, is passed through a column containing 40  mg of
granulated activated carbon (GAC).  The column is washed to remove any trapped
inorganic halides.   The GAC sample is exposed to thermal  neutron bombardment
creating a radioactive isotope.   Gamma ray emission, which is unique to each
halogen, is counted.  The area of the resulting peaks is directly  proportional
to the concentration of the halogens.


3.0  Interferences

     3.1   Method  interferences  may  be  caused by  contaminants,  reagents,
glassware, and other sample processing hardware.   All these materials must be
routinely demonstrated to be  free from interferences under the  conditions of
the analysis by running method blanks.

          3.1.1  Glassware must be scrupulously cleaned.   Clean all glassware
     as soon as possible after use by treating with cremate cleaning solution.
     This should be followed by  detergent washing  in hot water.    Rinse with
     tap water and distilled water, drain dry, and heat in a muffle furnace at
     400° C  for 15  to 30 min.   Volumetric ware  should not  be heated in  a

                                     D-4

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2

     muffle  furnace.    Glassware  should  be sealed  and stored  in  a   clean
     environment after drying and cooling to prevent any accumulation  of dust
     or other contaminants.

          3.1.2   The use of high-purity reagents and gases   helps to  minimize
     interference problems.

     3.2   Purity of the activated carbon must be  verified  before use.    Only
carboi; samples  that register less than 2,000  ng/40 mg  should  be used.    The
stock of  activated carbon should be stored  in its  granular form in  a  glass
container  with  a Teflon  seal.    Exposure  to  the  air must  be  minimized,
especially during and after milling and sieving the activated carbon.  No more
than a 2-week supply  should be  prepared in advance.   Protect  carbon at all
times from all  sources of halogenated organic vapors.   Store prepared carbon
and packed columns in glass containers with Teflon seals.

     3.3   It is possible that other radioisotopes, stray  radiation,  counting
geometries, and counting-equipment materials can affect gamma counting.   It is
essential that the data interpretation be performed by an  analyst experienced
at detecting these interferences.


4.0  Apparatus and Materials

     4.1  Adsorption system

          4.1.1   Dohrmann adsorption  module (AD-2),  or equivalent,  pressur-
     ized, sample and nitrate-wash reservoirs.

          4.1.2  Adsorption columns:  Pyrex, 5-cm-long x 6-mm-O.D. x 2-mm-I.D.

          4.1.3  Granular activated carbon (GAC):   Filtrasorb-400, Calgon-APC
     or equivalent, ground or milled,  and screened  to a 100/200  mesh range.
     Upon combustion of 40 mg of GAC, the apparent-halide background should be
     2,000 ng Cl'equivalent or less.

          4.1.4   Cerafelt  (available  from  Johns-Manville),  or equivalent:
     Form this material  into plugs  using a  2-mm-I.D.  stainless-steel  borer
     with ejection rod (available from Dohrmann) to  hold 40 mg  of GAC in the
     adsorption  columns.   CAUTION:   Do  not touch  this material with   your
     fingers.

          4.1.5  Column holders (available from Dohrmann).

          4.1.6   Volumetric flasks:   100-ml, 50-ml.   A general schematic of
     the adsorption system is shown in Figure 1.

     4.2   Containers suitable for containment of samples and standards during
irradiation (e.g., 1/5-dr polyethylene snap-cap vial).

     4.3   Sample  introduction  system and  a reactor  generating  a   thermal
neutron flux  capable  of  achieving  enough  halogen  activity   for  counting
purposes (e.g., a reactor having a  neutron flux of 5 x  10*2 neutrons/cm^/sec
and a pneumatic-tube sample introduction system).

                                      D-5

-------
                        N2
*-*-!
                                    Sample

                                    Reservoir

                                    (I of 4)
                                                      Nitrate Wash

                                                      Reservoir
o

a\
   GAG Column 1
                                 GAC Column 2
                                    Figure  1.   Schematic of absorption  system.
                                                                                                                        o
                                                                                                                        co
                                                                                                                      <£> O
                                                                                                                      -P> — •
                                                                                                                      CO -^.
                                                                                                                      CO O
                                                                                                                      • <<
                                                                                                                      o
                                                                                                                      o o
                                                                                                                        to
                                                                                                                        n

-------
                                                       OSWER  Policy  Directive
                                                             #9433.00-2

     4.4   A gamma ray detector and data handling  system capable  of resolving
the halogen peaks from potential  interferences and background.


5.0  Reagents

     5.1  Prepurified nitrogen.

     5.2  Deionized (DI) water.

     5.3  500 ppm N03 solution (KNOa dissolved in DI  water).

     5.4  50 percent v/v mixture acetone and nanograde hexane.

     5.5  0.1 M sodium sulfite (ACS reagent grade - 12.6  g/1).

     5.6  Nitric acid:  concentrated, reagent grade.

     5.7  25-pg Cl, 2.5-ycj Br, and 2.5-ug I standards.

     5.8  Radioactive standards to be used for calibrating gamma ray detection
          systems.


6.0  Sample Collection, Preservation, and Handling

     6.1   All samples  must have  been collected using  a sampling   plan  that
addresses the considerations discussed in Section One of  SW-846.

     6.2   All samples should be collected in bottles with Teflon  septa  (e.g.,
Pierce  #12722 or equivalent)  and be  protected from light.    If  this   is not
possible, use amber glass, 250-ml, fitted with Teflon-lined caps.  Foil  may be
substituted  for Teflon  if  the sample  is not  corrosive.   Samples must  be
protected against loss of volatiles by eliminating headspace  in the  container.
If amber bottles are not available, protect samples from  light.  The container
must be washed and muffled at 400° C before use to minimize contamination.

     6.3   All glassware  must be dried prior to  use according  to  the  method
discussed in 3.1.1.


7.0  Procedure

     7.1  Sample preparation

          7.1.1   Special care should be taken in handling the sample in order
     to minimize the loss of volatile organohalides.   The adsorption procedure
     should be performed simultaneously on the front and  back columns.

          7.1.2  Reduce residual  chlorine by adding sulfite (1 ml  of 0.1 M per
     liter of sample).  Sulfite should be added at the time of sampling  if the
     analysis is  meant to  determine  the TOX  concentration at  the  time of
     sampling.   It should  be recognized that TOX may  increase on  storage of
     the sample.  Samples should be stored at 4° C without headspace.

                                     D-7

-------
                                                  OSWER  Policy  Directive
                                                        #9433.00-2

     7.1.3   Adjust  the  pH  of  the  sample  to  approximately  2   with
concentrated HN03 just prior to adding the sample to the reservoir.


7.2  Calibration

     7.2.1   Check the adsorption efficiency of each newly prepared  batch
of carbon by analyzing  100 ml  of the adsorption-efficiency  standard,  in
duplicate, along with duplicates of the blank standard.   The net  recovery
should be within 5 percent of the standard value.

     7.2.2   Nitrate-wash blanks (method blanks):  Establish the  repeata-
bility  of the  method  background each  day by  first   analyzing several
nitrate-wash  blanks.   Monitor this  background by  spacing nitrate-wash
blanks between each group of eight analysis determinations.   The  nitrate-
wash  blank values are obtained on  single columns  packed with 40   mg  of
activated  carbon.   Wash with  the nitrate  solution as  instructed for
sample analysis, and then analyze the carbon.

     7.2.3   Prior to each day's operation, the instrument  is  calibrated
using  radioactive standards  (e.g., cobalt-60  and  radium-226 sources).
The instrument is calibrated such that gamma rays from the standards fall
within plus or minus one channel of their true energies.  A 100-sec  blank
is then counted to verify  that no  stray radioactive sources  are within
sensing distance  of the detector.   As data  are obtained throughout the
day, peak locations in the  standards are monitored to ensure  that  there
is no electronic drift of the instrument.   If drift is  noted,  the system
is recalibrated.
7.3  Adsorption procedure

     7.3.1   Connect two  columns in  series, each  containing  40 mg  of
100/200-mesh activated carbon.

     7.3.2   Fill the  sample reservoir,  and pass  a  metered amount  of
sample through the activated-carbon columns at a rate of  approximately 3
ml/min.   NOTE:   10U ml of sample is the preferred volume for concentra-
tions of TOX between 5 and 500 ug/1; 50 ml  for 501 to 1,000 g/1,   and 25
ml for 1,001 to 2,000 g/1.

     7.3.3   Wash  the  columns-in-series  with  at least  2  ml  of  the
5,000-mg/l  nitrate solution  at  a rate  of approximately  2  ml/min  to
displace inorganic chloride ions.
7.4  Activation

     7.4.1  After the quartz collection tube with the GAC is removed from
the extraction  unit, the GAC  and cerafelt  pads are extruded  using the
packing rod into a prewashed plastic container (e.g., 1/5-dr polyethylene
snap-cap vial).   The vial  is prewashed to remove inorganic  and organic
chlorine by soaking in distilled water followed by storage in a glass jar
containing 50  percent v/v acetone and hexane.   Just prior  to extrusion

                                D-8

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2

     the vial is removed by  forceps and  air-dried to remove  residual  water,
     acetone, and hexane.   The vial is snapped shut,  the hinge removed  with  a
     scalpel blade, the cap heat-sealed to the vial with an electric soldering
     gun reserved for that purpose,  and a  single-digit number placed  on  the
     vial with a Marks-A-Lot (or equivalent)  marker pen.

          7.4.2   Samples plus a similar vial  containing 25  pg Cl, 2.5  u9  Br,
     and 2.5 yg I standards are then introduced into the reactor,  generally by
     placing them together in a b-dr polyethylene vial and inserting them into
     a pneumatic-tube transfer "rabbit" for neutron irradiation.    Irradiation
     would be for  a 15-min  period at  a  thermal  neutron  flux of  b  x 1012
     neutrons/cm2/sec.   After  returning  from  the  reactor,  the  rabbit  is
     allowed  to  "cool"  for  20  min  to  allow   short-lived radioisotopes
     (primarily Al) present in the GAC to decay away.

     7.5  Detection

          7.5.1   Analysis  is  performed using  a  lithium  drifted  germanium
     Ge(Li)  gamma ray detector  with an  amplifier and a  4096-channel  memory
     unit  for data storage.   The analyses  can be performed either manually,
     with the operator changing samples and transferring the data   to magnetic
     tape, or automatically,  with both  functions  performed by  an automatic
     sample changer.

          7.5.2   Analysis begins  by counting the  standard and samples for  a
     suitable time  period (e.g., 200  sec "live"  time for the standards  and
     samples).   The operator  records the time intervals between   samples  and
     the "dead" time of each sample in a logbook for later use   in calculating
     halogen concentrations in each sample.
     7.6  Calculations

          7.6.1   Chlorine,  bromine,  and iodine  can be  analyzed  within   a
     200-sec counting period taking place 20 to 40 min after irradiation.

          7.6.2  Chlorine is analyzed using the 1642-KeV gamma ray produced  by
     37.1-min  ^C].   Bromine is  analyzed using  the 616-KeV gamma  ray  from
     17.7-min  H^Br, while  iodine  is analyzed  using the  442-KeV  gamma ray
     produced by 25-min 128i.

          7.6.3  The calculation used for quantitation is:

     ppm halogen = cts unk. x counting time std. x pg in std.   x ext
                   cts std.   counting time unk.   sample vol.
where
     cts unk. = the integrated area of  the appropriate gamma ray peak  in the
                unknown with background subtracted and the total  multiplied by
                1   +  [(percent   dead  time   unknown - percent   dead  time
                std.)/200].   The latter  correction is  usually  less  than  4
                percent and corrects for pile-up errors.


                                     D-9

-------
                                                       OSWER  Policy  Directive
                                                             #9433.00-2

     cts std. = the integrated area of  the appropriate gamma ray  peak   in  the
                standard with background subtracted.

     counting time std.  = the "live" counting time in  seconds of the standard.

     counting time unk.  = the "live" counting time in  seconds of   the unknown.

     ug in std. = the number of  micrograms of the stable  element  in question
                  in the standard (25 for Cl, 2.5  for  Br and  I).

     sample  vol. =  the volume of  sample passed   through the GAC  column,  in
                     mi 11-Miters.

     ext = tne  decay  correction  to  bring  all  statistics  back  to t = 0.x =
           0.693/tj/2 where t\/2 = the half-life,  in minutes.

     t =  the  time interval  in minutes  from  the end of the count   of  the
          standard until the end of the count of the sample.

          7.6.4   No further  calculations are necessary as long   as the final
     sample is counted within 40 min after the end of  irradiation.  If samples
     are counted after 40 min, the addition  to the 616-KeV peak of   80gr from
     the 619-KeV peak from 82gr becomes large enough that  a  correction  factor
     must be applied.   In practice, all counting  should be completed in less
     than 40 min after irradiation.
8.0  Quality Control

     (Refer to Section 3.4 of this example petition).


9.0  References


2.2 DESCRIPTION OF APPLICABLE SAMPLE OR MATRICES1

     ERG is  requesting that the  proposed test  method be approved  to screen
groundwater,  surface  water,  and  wastewater treatment  plant  effluent  for
organic halogens.  In  all cases,  these three  categories  represent  aqueous
liquids.   Only waste treatment plant effluent may be   deemed an actual waste.
Groundwater and surface water are merely analyzed for  hazardous contaminants.

     The groundwater analyzed for this petition was well water  collected from
aquifers  in  the  State of  Michigan.   Raw  well  water  (unchlorinated  and
unsoftened)  was  collected from  aquifers at  approximately  100- to 150-foot
depths.  These waters were generally hard and relatively high in iron.
  All information in Section 2.2 is fictitious.


                                     D-10

-------
                                                       USWtR  Policy  Directive
                                                             #9433.00-2

     The surface waters were  collected from  rivers in Michigan.    These  were
hard waters, quite clear in appearance, with only minor suspended  solids.

     The waste  treatment  plant  effluent  samples  were  provided   with   the
permission of municipal  treatment plants  in Michigan (all   tertiary systems)
from their effluent discharge into the rivers mentioned above.

  [NOTtJ:  The  number of samples taken from each type of water  and the  number
          of locations sampled is determined during  pre-petition  negotiations
          between EPA and the petitioner.]


2.3   ASSESSMENT OF  LIMITING OR INTERFERING FACTORS

     Quality control of  the extraction  unit (i.e.,  the  Dohrmann  adsorption
module (AD-2)) proved to be relatively straightforward with  three exceptions:
variation in sample size  passed through  the granular activated  carbon (GAC)
columns, the potential for channeling  in the  GAC columns, and the variation
between GAC  background levels  from  bottle to  bottle and  column   to  column
because of difficulty obtaining the same volume of GAC in each  column.

     In  general, the  automatic  shutoff feature  of the  AD-2 module  worked
satisfactorily,  but sample  would  continue to  pass slowly  through the GAC
column  after shutoff  due to  gravity  if the  GAC columns  were  not removed
immediately.   This  problem  could be  accounted for  if  the   excess  volume
remained within the volumetric flask and  could be measured,  but if   the flask
overflowed  the  sample had  to be  discarded.   To  prevent  overflow,  it  is
recommended that only  1 or 2 milliliters of  sample in  excess of that   to be
analyzed be placed in the sample storage tube.

     Two GAC tubes in sequence should always  be used but they  both   also  need
to be analyzed if any detectable halogen shows up in the upper  tube.  We found
occasionally (5 percent  occurrence) that channeling or some  other  phenomenon
would take  place in the upper tube  such that breakthrough to  the  lower  tube
would take place even though the GAC in  the upper tube was not yet saturated
with organic halogen.

     The  only area  creating  a  potential  problem  was  in  the  instrument
stabilization  step after insertion of the  sample into the boat but  prior to
initiation of the analysis step.   For good data reproducibility,  it was found
necessary to allow the MC-1 to stabilize to plus or minus  two-digit variation
in the  TOX Det mode.   This often  took 5 to 10 minutes  with  the  instrument
appearing to plateau at a level (often about 3.5 on the panel meter)  and  then
suddenly change to a lower value about 1.1 where pertinent stabilization would
take place.   For samples very high in organic halogens (>20  u9 ultimate meter
reading),  stabilization was  very slow and  may be  due to  a   continuous off
gassing of  small amounts of  volatile organic  halogens even prior   to  intro-
ducing the sample into the furnace.
                                     D-ll

-------
                                                       OSWER Policy Directive
                                                             #9433.00-2

     For the actual  neutron activation analysis, the amount  of radioactivity
induced in a sample is directly proportional  to the neutron flux it is exposed
to.   Since this  flux can vary depending on   how the sample is positioned   in
relation to the reactor core during irradiation, it is essential that  a known
standard be irradiated with every sample batch to act as a flux monitor. This
was done  throughout the  study.   Since the   sample acts  as a  point  source
emitting radiation (much as a light bulb viewed at a distance),  the intensity
of the  radiation falls off with the  square  of the distance from  the source.
Thus,  care  must  be taken  to  ensure  that  the standard  and  all  samples
associated with  the standard  are  counted  at  the same  distance  from  the
detector.   This is accompli shad by using an  automatic sample changer that  can
position the samples to within plus or minus  1 millimeter of the same position
each time.
2.4  TEST METHOD QUALITY CONTROL (QC) PROCEDURES

     2.4.1  All  QC data should be maintained and available for easy reference
or inspection.

     2.4.2  Before performing  any analyses, the  analyst must demonstrate the
ability to generate  acceptable accuracy and  precision with this procedure by
analyzing appropriate QC check samples.

     2.4.3  The  laboratory must  develop  and maintain  a statement of method
accuracy for  their laboratory.  The  laboratory  should  update  the accuracy
statement regularly as new recovery measurements are made.

     2.4.4  Employ a  minimum of  one blank  per sample  batch to determine if
contamination is occurring.

     2.4.5  Run check standard after approximately every 15 samples.

     2.4.6  Run 1 duplicate  sample for every 10  samples.  A duplicate sample
is a sample brought through the whole sample preparation process.

     2.4.7   It is  recoi»nend(Wt',>that " the laboratory adopt  additional  quality
assurance (QA) practices  for us!  with this method.   The specific  practices
that would be most productive will depend upon the needs of the laboratory and
the nature of the samples.   Field duplicates  may be analyzed to monitor  the
precision of the sampling technique.  Whenever possible, the laboratory should
perform analysis of  standard reference materials and participate  in relevant
performance-evaluati on studies.

     2.4.8   Quality control  fof the  analysis phase is  very straightforward
since the  instrument is a noncontact analyzer.   That is,  only the radiation
emitted from the sample - not the sample - should  touch the analyzer.    Since
contamination of the system is not  usually a problem (unless a  sample spills
on  it), the most serious QC  issues deal with uniform neutron  flux, counting
geometry, and spectral interpretation.  The amount of radioactivity induced in
a sample is directly proportional id ttte neutron flux it is exposed to.  Since
                                     0-12

-------
                                                       OSUER  Policy  Directive
                                                             #9433.00-2

this flux  can vary depending on how  the sample is positioned in   relation  to
the reactor core during irradiation, it is essential  that  a known   standard  be
irradiated with every sample batch to  act as  a flux monitor.    Care must  also
be  taken to  ensure that the  standard and  all samples   associated  with the
standard are counted at the same distance from the detector.


      3.0  STATEMENT OF NEED AND JUSTIFICATION FOR THE PROPOSED METHOD1
3.1  NEED

     As discussed in the Statement of Interest,  only one method  for determin-
ing  the presence  of organic halogens  is currently  approved by  EPA.    This
method (9020) is limited to one brand of instrumentation, the Dohrmann   DX 20.
We wish to provide an equivalent alternative to  this analytical   device, which
can be  substituted if  necessary.   This  alternative  is neutron  activation
analysis used via the Haloscan™ method.
3.2  JUSTIFICATION

     This section compares the two analytical  methods (the  currently  approved
Dohrmann TOX method (9020) and the  ERG HaloscanTM method for the  analysis  of
organically bound halogens (chlorine, bromine, and iodine))  in three  types  of
water samples:   groundwater, surface water,  and wastewater.    The purpose  of
this exercise was to determine  that the  two methods produce  essentially the
same total halogen result and  to verify  that the Haloscan™  method  provides
•••ditional information in that  each halogen  species is reported  out  indepen-
  •;itly of the others.

     In the  following  pages, ERG  presents  comparability  data  showing  the
results of these two test methods.   Precision, accuracy, and QC data  are  also
provided.   Sampling and  test data on each replicate  are provided in Section
3.3.
     [NOTE:  At  this  point, the  petitioner  should  list  the data  quality
             objectives  set  by  EPA  and describe  the  experimental   design
             established  for  this test  program, addressing  the  number and
             timing  of samples,  locations, replicates,  and  testing,  among
             other things.]
     discussion as well as the data in this section is hypothetical


                                     D-13

-------
                                                       OSWER  Policy  Directive
                                                             #9433.00-2

     Table 3-1A contains the test  results  from  ground water source  number 1
(Aquifer A)1.   In each  sample,  the ambient  levels  of  chlorine,  bromine,  and
iodine were above the detection  limit of the Dohrmann method.   Therefore,  it
was not necessary to spike samples.

     Tables 3-2A and 3-2B present the QC data collected during  the analyses  of
ground water  samples.2   QC  checks  were  made  as  outlined   in   Sections  7.4
through 7.7 of the test method description.
 [NOTE:   Here, the petitioner should present  detailed  information   on  statis-
          tical  procedures  used  to  evaluate test  data.    Also,   raw   data
          tables or  computer  printouts of  the  statistical   analyses should
          be provided.   Section 4  and  Appendix B provide   detailed informa-
          tion about this aspect of the petition.]
  1 Table 3-1A contains fictitious data for one source of groundwater.   In the
    interest  of conserving paper,  similar tables  were not prepared  for the
    remaining sources of ground water  and the sources for both  surface water
    and wastewater effluent.   Thus, an actual petition, using the test design
    in this petition would have several tables that are similar to  the format
    of Table 3-1A.

  2 A  complete petition would  contain similar  tables for surface  water and
    wastewater effluent samples.
                                     D-14

-------
                 TABLE 3-1A.
                         OSWER  Policy Directive
                               #9433.00-2

RESULTS OF GROUNDWATER  ANALYSES:
SOURCE NO. 1,  AQUIFER A (EXAMPLE TABLE)1
           Samples 1-1,  1-2,  1-3:  analyzed by ERG—1/1/85
           Samples 1-4,  1-5,  1-6:  analyzed by ABC Associates--l/l/85
  Method 9020

Replicate-1   1.53
        R-2   2.26

       Mean = 1.89
  Std. dev. = 0.52
  I  Sample 1-1  |
   (ug/100 ml)

               Haloscan"  (Specific halogens)
               cr=           sP1         r^
  Replicate-3  OS          <0.1
          R-4  2.30          <0.1

                         Mean  = 2.08
                    Std.  dev.  = 0.31
                                  Sample 1-2  |
                                  (pg/100 ml)
Method 9020

Replicate-1
        R-2
  Replicate-3
          R-4
               Haloscan"  (Specific  halogens)
               rr=          ^Fi-
       Mean =
  Std. dev. =
                         Mean =
                    Std.  dev. =
                                I  Sample 1-3  |
                                  Ug/100 ml)
Method 9020

Replicate-1
        R-2
  Replicate-3
          R-4
               Haloscan"  (Specific halogens)
               FT          BF          r^~
       Mean =
  Std. dev. =
                         Mean  =
                    Std.  dev.  =
                                                            (Continued)

1 The  number of  samples  taken   will  vary based  on prepetition negotiations,
  All data are fictitious.
                                     D-15

-------
            TABLE 3-1A.
                              OSWER Policy Directive
                                    #9433.00-2

RESULTS Of; GRU'JNDWATER ANALYSES
SOURCE NO. 1. AUUIFER A (EXAMPLE TABLE)-1 (CONT.)
  Method 9020

Replicate-1
        R-2
                              Samp'ie 1-4 |

                              (yg/100ml)
         Replicate-3
                 R-4
                      Haloscan  (Specific halogens)
                      Cl-          Br-          I-
       Mean =
  Std. dev. =
                               Mean =
                          Std. dev. =
  Method 9020
Replicate-1
        R-2
                              Sample 1-5 |

                              (ug/lOOml)
         Replicate-3
                 R-4
                      Haloscan   (Specific halogens)
                      CT=Br=         r~^
       Mean
  Std. dev.
                                 Mean =
                            Std. dev. =
  Method 9020

Replicate-1
        R-2
                              Sample 1-6 |

                              (ug/lOOml)
                      Haloscan   (Specific halogens)
                      (TFBr'         I~
         Replicate-3
                 R-4
       Mean =
  Std. dev. =
                                  Mean =
                             Std. dev. =
1 Data are fictitious.
                                     D-16

-------
                                           OSWER Policy Directive
                                                 #9433.00-2
TABLE 3-2A.  QC DATA—GROUNDWATER ANALYSIS VIA METHOD  9020
                     (EXAMPLE TABLE)
                        (yg/ 100ml)
Samples
^qu'ifer A
1-1
1-2
1-3
1-4
1-5
1-6

Aquifer B
2-1
2-2
2-3
2-4
2-5
2-6

Aquifer C
3-1
3-2
3-3
3-4
3-5
3-6

Blank
(1 per sample)





















Field Duplicate
(1 per sample)





















Lab Duplicate
(1 per 10)
N. A.

(R- )
(R- )
(R- )
(R- )
N.A.


(R- )
(R- )
(R- )
(R- )
N.A.

(R- )

(R- )
(R- )
(R- )
N.A.

(R- )
(R- )
Standard
(1 per 15)
N.A.

(R- )
N.A.

(R- )
N.A.

(R- )
N.A.

(R- )
N.A.

(R- )
N.A.

(R- )
N.A.

(R- )
(R- )
N.A.

(R- )
N.A.

N.A. = Not Applicable.
                                                     (Continued)
                           D-17

-------
TABLE 3-2A.  QC
                                                          OSWER Policy Directive
                                                                #9433.00-2

                                                     »„ METHOD 9020  (CW1T.,
                                      ug/100ml)
                                                     Lab Duplicate
                                                     (1 per 10
                                                     replicates)
                                                              Standard
                                                              (1 per 15
                                                              replicates)
N.A. = Not Applicable.
                                     D-18

-------
                                                     OSWER  Policy Directive
                                                           #9433.00-2
       TABLE 3-2B.  QC OATA--GROUNDWATER  ANALYSIS  VIA  HALOSCAN" METHOD
                               (EXAMPLE TABLE)
                                  (ug/lOOml)
Samples
Aquifer A
1-1
1-2
1-3
1-4
1-5
1-6

Aquifer B
2-1
2-2
2-3
2-4
2-5
2-6

Aquifer C
3-1
3-2
3-3
3-4
3-5
3-6

Blank
(1 per sample)





















Field Duplicate
(1 per sample)





















Lab Duplicate
(1 per 10
replicates)
N.A.

(R- )
(R- )
(R- )
(R- )
N.A.


(R- )
(R- )
(R- )
(R- )
N.A.

(R- )

(R- )
(R- )
(R- )
N.A.

(R- )
(R- )
Standard
(1 per 15
replicates)
N.A.

(R- )
N. A.

(R- )
N.A.
(R- )
N.A.

(R- )
N. A.

(R- )
N.A.

(R- )
N.A.

(R- )
(R- )
N.A.

(R- )
N.A.
N.A. = Not Applicable
(Continued)
                                    D-19

-------
                                                     OSWER  Policy  Directive
                                                           #9433.00-2

    TABLE 3-2B.   QC DATA--GROUNDWATER  ANALYSIS  VIA  HALOSCAN™  METHOD  (CONT.)
                              (EXAMPLE TABLE)
                                (
Samples
Aquifer D
4-1
4-2
4-3
4-4
4-5
4-6

Aquifer E
5-1
5-2
5-3
5-4
5-5
5-6

Blank
(1 per sample)














Field Duplicate
(1 per sample)














Lab Duplicate
(1 per 10
replicates)

(R- )
(R- )
N.A.

(R- )
(R- )
(R- )

(R- )
N.A.

(R- )
(R- )
(R- )
(R- )
Standard
(1 per 15
replicates)

(R- )
N.A.

(R- )
N.A.

(R- )
N.A.


(R- )
N.A.

(K- )
N.A.

(R- )
(R- )
N.A. = Not Applicable.
                                     D-20

-------