United States
Environmental Protection
Agency
Office of Policy, Planning,
and Evaluation
Washington, DC 20460
EPA 230/02-89-042
February 1989

-------
    Methods for  Evaluating the
Attainment of  Cleanup  Standards
Volume 1:  Soils and  Solid  Media
                        rt S Environmental
                        "  , , 5. Library
                             '
           Statistical Policy Branch (PM-223)
        Office of Policy, Planning, and Evaluation
         U. S. Environmental Protection Agency
               401 M Street, S.W.
              Washington, DC 20460

-------
                                 DISCLAIMER
This report was prepared under contract to an agency of the United States Government.
Neither  the  United  States  Government  nor any of its  employees, contractors,
subcontractors, or their employees makes any warranty, expressed or implied, or assumes
any legal liability or responsibility for any third parry's use or the results of such use of any
information, apparatus, product, model, formula, or process disclosed in this report, or
represents that its use by such  third parry would not infringe on privately owned rights.


Publication of the data in this document does not signify that the contents necessarily reflect
the joint or separate views and policies  of each co-sponsoring agency. Mention of trade
names or commercial products does not constitute endorsement or recommendation for use.

-------
                      TABLE OF CONTENTS

                                                                  Page
   Authors and Contributors	xv
   Executive Summary	xvi

1. INTRODUCTION	1-1

   1.1    General Scope and Features of the Guidance Document	1-1
       1.1.1   Purpose	:	1-1
       1.1.2   Intended Audience and Use	1-3
       1.1.3   Bibliography, Glossary, Boxes, Worksheets, Examples,
              and References to "Consult a Statistician"	1-6
   1.2    A Categorization Scheme for Cleanup Standards	1-6
       1.2.1   Technology-Based Standards	1-7
       1.2.2   Background-Based  Standards	1-7
       1.2.3   Risk-Based  Standards	1-8
   1.3    Use of this Guidance in Superfund Program Activities	1-8
       1.3.1   Emergency/Removal Action	1-8
       1.3.2   Remedial Response Activities	1-9
       1.3.3   Superfund  Enforcement	1-9
   1.4    Treatability Studies and Soils Treatment Technologies	1-9
       1.4.1   Laboratory/Bench-Scale Treatability Studies	1-10
       1.4.2   Field/Pilot-Scale Treatability Studies	1-10
       1.4.3   Soils Treatment by Chemical Modification	1-11
       1.4.4   Soils Treatment by In Situ Removal of Contaminants	1-11
       1.4.5   Soils Treatment by Incineration	1-13
       1.4.6   Soils  Removal	1-13
       1.4.7   Soils Capping	1-14
   1.5    Summary:	1-14

2. INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS	2-1

   2.1    Hypothesis Formulation and Uncertainty	2-2
   2.2    Power Curves as a Method of Expressing Uncertainty and
          Developing Sample Size Requirements	2-6
   2.3    Attainment or Compliance Criteria	2-8
      2.3.1   Mean	2-9
      2.3.2   Proportions or Percentiles	2-9
   2.4    Components of a Risk-Based Standard	2-11
   2.5    Missing or Unusable Data, Detection Limits, Outliers	2-12
      2.5.1   Missing or Unusable Data	2-12
      2.5.2   Evaluation of Less-Than-Detection-Limit Data	2-15
                                      111

-------
                           TABLE OF CONTENTS

                                                                 Page
      2.5.3   Outliers	2-16
   2.6   General Assumptions	2-17
   2.7   A Note on Statistical Versus Field Sampling Terminology	2-17
   2.8   Summary	2-18
3.  SPECIFICATION OF ATTAINMENT OBJECTIVES	3-1
   3.1   Specification of Sample Areas	3-1
   3.2   Specification of Sample Collection and Handling Procedures	3-4
   3.3   Specification of the Chemicals to be Tested	3-5
   3.4   Specification of the Cleanup Standard	3-5
   3.5   Selection of the Statistical Parameter to Compare with the
         Cleanup  Standard	3-6
      3.5.1   Selection Criteria for  the Mean, Median, and Upper
              Percentile	3-6
      3.5.2   Multiple Attainment Criteria	3-9
   3.6   Decision Making With Uncertainty: The Chance of Concluding
         the Site Is Protective of Public Health and the Environment
         When It Is Actually Not Protective	3-10
   3.7   Data Quality Objectives	3-11
   3.8   Summary	3-12
4.  DESIGN OF THE SAMPLING AND ANALYSIS PLAN	4-1

   4.1   The Sampling Plan	4-1
      4.1.1   Random Versus Systematic Sampling	4-2
      4.1.2   Simple Versus Stratified Sampling	4-4
      4.1.3   Sequential  Sampling	4-6
   4.2   The Analysis Plan	4-6
   4.3   Summary	4-7
5.  HELD SAMPLING PROCEDURES	5-1
   5.1   Determining the General Sampling Location	5-1
   5.2   Selecting the Sample Coordinates for a Simple Random Sample	5-3
   5.3   Selecting the Sample Coordinates for a  Systematic Sample	5-5
      5.3.1   An Alternative Method for Locating the Random Start
              Position for a Systematic Sample	5-10
   5.4   Extension to Stratified Sampling	5-13
   5.5   Field Procedures for Determining the Exact Sampling Location	5-13
   5.6   Subsampling and Sampling Across Depth	5-14
      5.6.1   Depth  Discrete  Sampling	5-15
      5.6.2   Compositing Across Depth	5-15
      5.6.3   Random Sampli ng Across Depth	5-17
                                     IV

-------
                           TABLE OF CONTENTS


                                                               Page

   5.7   Quality Assurance/Quality Control (QA/QC) in Handling the
         Sample During and After Collection	5-18
   5.8   Summary	5-18

6. DETERMINING WHETHER THE MEAN CONCENTRATION OF
      THE SITE  IS STATISTICALLY LESS THAN  A CLEANUP
      STANDARD	6-1

   6.1   Notation Used in This Chapter	6-1
   6.2   Calculating the Mean, Variance, and Standard Deviation	6-2
   6.3   Methods for Random  Samples	6-4
      6.3.1    Estimating the Variability of the Chemical Concentration
              Measurements	6-4
         6.3.1.1   Use of Data from a Prior Study to Estimate a	6-5
         6.3.1.2   Obtain  Data to Estimate a After a Remedial Action
                  Pilot	6-5
         6.3.1.3   An Alternative Approximation for &	6-6
      6.3.2    Formulae for Determining Sample Size	6-7
      6.3.3    Calculating the Mean, Standard Deviation, and Confidence
              Intervals	6-10
      6.3.4    Inference:  Deciding  Whether the Site Meets Cleanup
              Standards	6-11
   6.4   Methods for Stratified Random Samples	6-12
      6.4.1    Sample Size Determination 	6-13
      6.4.2    Calculation of the Mean and Confidence  Intervals	6-15
      6.4.3    Inference: Deciding Whether the Site Meets Cleanup
              Standards	6-18
   6.5   Methods for Systematic Samples	6-20
      6.5.1    Estimating Sample Size	6-20
      6.5.2    Concerns Associated with Estimating the Mean, Estimating
              the Variance, and Making Inference from a Systematic
              Sample	6-21
         6.5.2.1   Treating a Systematic Sample as a Random Sample	6-22
         6.5.2.2   Treating the  Systematic  Sample  as a Stratified
                  Sample	6-22
         6.5.2.3   Linearization and  Estimates from Differences
                  Between Adjacent Observations of a Systematic
                  Sample	6-25
   6.6   Using Composite Samples When Testing the Mean	6-26
   6.7   Summary	6-27

7. DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
      THE SITE IS LESS THAN A CLEANUP STANDARD	7-1

   7.1   Notation Used in This Chapter	7-2

-------
                           TABLE OF CONTENTS


                                                                 Page
   7.2    Steps to Correct for Laboratory Error	7-3
   7.3    Methods for Simple Random Samples	7-5
      7.3.1   Sample Size Determination	7-5
      7.3.2   Understanding Sample Size Requirements	7-6
      7.3.3   Estimating  the  Proportion Contaminated  and the
              Associated Standard Error	7-7
      7.3.4   Inference:  Deciding Whether a Specified Proportion of the
              Site is Less than a Cleanup Standard Using a Large
              Sample Normal Approximation	7-8
      7.3.5   Deciding Whether a Specified Proportion of the Site is
              Less than the Cleanup Standard Using an Exact Test	7-9
   7.4    A Simple Exceedance Rule Me'thod for Determining Whether a
          Site Attains the Cleanup Standard	7-11
   7.5    Methods for Stratified Samples	7-12
      7.5.1   Sample Size Determination	7-13
      7.5.2   Calculation of Basic Statistics	7-16
      7.5.3   Inference: Deciding Whether the Site Meets Cleanup
              Standards  	7-19
   7.6    Testing Percentiles from a Normal or Lognormal Population
          Using Tolerance Intervals	7-20
      7.6.1   Sample Size Determination	7-21
      7.6.2   Testing the Assumption of Normality	7-23
      7.6.3   Inference:  Deciding Whether the Site Meets Cleanup
              Standards Using Tolerance Limits	7-24
   7.7    Summary	7-26

8. TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
      SAMPLING	8-1

   8.1    Notation Used in This Chapter	8-2
   8.2    Description of the Sequential Procedure	8-3
   8.3    Sampling Considerations in Sequential Testing	8-4
   8.4    Computational Aspects of Sequential Testing	8-5
   8.5    Inference: Deciding Whether the Site Meets Cleanup Standards	8-7
   8.6    Grouping Samples in Sequential Analysis	8-8
   8.7    Summary	8-10

9. SEARCHING FOR HOT SPOTS  	9-1

   9.1    Selected Literature that Describes Methods for Locating Hot
          Spots	9-1
   9.2    Sampling and Analysis Required to Search for Hot Spots	9-1
      9.2.1   Basic  Concepts	9-1
      9.2.2   Choice of a Sampling Plan	9-4
      9.2.3   Analysis   Plan;	9-8
   9.3    Summary	9-8
                                      VI

-------
                         TABLE OF CONTENTS
                                                           Page
10. THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
      THE ATTAINMENT OF CLEANUP STANDARDS	10-1

   10.1   Background	10-2
      10.1.1   What Is Geostatistics and How Does It Operate?	10-2
      10.1.2   Introductory Geostatistical References	10-4
   10.2   Soils Remediation Technology and the  Use of Geostatistical
         Methods	10-4
      10.2.1   Removal	10-5
      10.2.2   Treatment Involving Homogenization	10-6
      10.2.3   Flushing	10-7
   10.3   Geostatistical Methods that Are Most Useful for Verifying the
         Completion of Cleanup	10-8
   10.4   Implementation of Geostatistical Methods	10-9
   10.5   Summary	10-12
BIBLIOGRAPHY	BIB-1

APPENDIX A STATISTICAL TABLES	A-l

APPENDIXB EXAMPLE WORKSHEETS	B-l

APPENDIX C BLANK WORKSHEETS	C-l

APPENDIX D GLOSSARY	D-l

INDEX	IND
                                  vu

-------
                           TABLE OF CONTENTS




                            LIST OF FIGURES


                                                                 Page

Figure 1.1    Steps in Evaluating Whether a Site Has Attained the Cleanup
             Standard	1-2

Figure 2.1    A Statistical Perspective of the Sequence of Ground Water
             Monitoring Requirements Under RCRA	2-4

Figure 2.2    Hypothetical  Power Curve	2-6

Figure 2.3    Hypothetical Power  Curve Showing False  Positive and
             False Negative Rates	2-7

Figure 2.4    Measures of Location: Mean, Median, 25th Percentile, 75th
             Percentile, and 95th Percentile for Three Distributions	2-10

Figure 2.5    Components of a Risk-Based Standard	2-13

Figure 3.1    Steps in Defining the Attainment Objectives	3-2

Figure 3.2    Geographic Areas and Subareas Within the Site	3-4

Figure 4.1    Dlustration of Random, Systematic, and Stratified Sampling	4-3

Figure 5.1    Map of a Sample Area with a Coordinate  System	5-2

Figure 5.2    Map  of a Sample  Area  Showing Random Sampling
             Locations	5-6

Figure 5.3    Examples of a Square and a Triangular Grid for Systematic
             Sampling	5-6

Figure 5.4    Locating a Square Grid Systematic Sample	5-8

Figure 5.5    Map  of a  Sample Site Showing  Systematic Sampling
             Locations	5-10

Figure 5.6    Method for Positioning Systematic Sample Locations in the
             Field	5-12

Figure 5.7    An Example Illustration of How to Choose an Exact Field
             Sampling Location from an Approximate Location	5-14

Figure 5.8    Subsampling and Sampling Across Depth	5-16

Figure 6.1    An Example of How to  Group Sample Points from a
             Systematic Sample so that  the Variance and  Mean Can Be
             Calculated Using the Methodology for a Stratified Sample	6-24
                                     Vlll

-------
                            TABLE OF CONTENTS


                                                                  Page

Figure 6.2    Example of a Serpentine Pattern	6-25

Figure 8.1    Graphic Example of Sequential Testing	8-4

Figure 9.1    A Square Grid of Systematically Located Grid Points with
             Circular and Elliptical Hot Spots Superimposed	9-4

Figure 9.2    Grid Spacing and Ellipse Shape Definitions for the Hot Spot
             Search Table in Appendix A	9-5

Figure 10.1   An Example of an Empirical Variogram and a Spherical
             Variogram Model	-.	10-3

Figure 10.2   Contour Map of the Probability in Percent of Finding the
             Value of 1,000 ppm or a Larger Value	10-11

Figure 10.3   Contour Map of the Probability in Percent of a False Positive
             in the Remedial Action Areas and the 1,000 Contour Line	10-11

Figure 10.4   Contour Map of  the Probability in  Percent  of a False
             Negative in the Remedial Action Areas and the 1,000  ppm
             Contour Line	10-11

Figure A.I    Power Curves fora= 1%	A-12


Figure A.2    Power Curves for a = 5%	A-13

Figure A.3    Power Curves for a = 10%	A-14


Figure A.4    Power Curves for a = 25%	A-15

Figure B.I    Example Worksheets: Parameters to Test in Each Sample
             Area and Map of the Site	B-2

Figure B.2    Example Worksheets: Sequence in Which the Worksheets
             are Completed	B-3
                                           IX

-------
                            TABLE OF CONTENTS
                              LIST OF TABLES
                                                                    Page

Table 1.1     EPA guidance documents that present methodologies for
             collecting and evaluating soils data	1-5

Table 2.1     A diagrammatic explanation of false positive and false
             negative  conclusions	2-5

Table 3.1     Points to consider when trying to choose among the mean,
             high percentile, or median	3-7

Table 3.2     Recommended parameters to test when comparing the
             cleanup standard to the average concentration of a chemical
             with chronic effects	3-9

Table 4.1     Where sample designs  and analysis  methods  for  soil
             sampling  are discussed in this document	4-7

Table 7.1     Selected information from Tables A.7-A.9 that can be used
             to determine the sample sizes required for zero or few
             exceedance rules associated with various levels of statistical
             performance and degrees  of cleanup	7-14

Table 9.1     Selected  references regarding the methodologies for
             identifying  hot  spots at waste sites	9-2

Table 9.2     Factors controlling the design of a hot spot search sampling
             plan	9-6

Table 10.1    Selected introductory and advanced references that introduce
             and discuss geostatistical concepts	10-5

Table 10.2    Introductory references  for indicator, probability, and
             nonparametric global estimation kriging	10-10

Table 10.3    Selected  geostatistical software	10-13

Table A.I     Table of t for selected alpha and degrees of freedom	A-l

Table A.2     Table of z for selected alpha or beta	A-2

Table A.3     Table  of k for selected alpha, PQ, and sample size where
             alpha = 0.10 (i.e., 10%)	A-3

Table A.4     Table  of k for selected alpha, PO, and sample size where
             alpha  =  0.05  (i.e., 5%)	A-4

Table A.5     Table  of k for selected alpha, PQ, and sample size where
             alpha  =  0.01  (i.e.,  1%)	A-5

-------
                              TABLE OF CONTENTS


                                                                      Page

Table A.6     Sample sizes required for detecting a scaled difference tau of
              the mean from the cleanup standard for selected values of
              alpha and beta	A-6

Table A.7     Sample size required for test for proportions with a = .01
              and (3 = .20, for selected values of PO and P!	A-7

Table A.8     Sample size required for test for proportions with a = .05
              and (3 = .20, for selected values of P0 and P!	A-8

Table A.9     Sample size required for test for proportions with a = .10
              and P = .20, for selected values of PQ and P!	A-9

Table A. 10    Tables for determining critical values for the exact binomial
              test, with a = 0.01, 0.05, and 0.10	A-10

Table A. 11    The false positive rates associated with hot spot searches as a
              function of grid spacing and hot spot shape	A-ll
                                        XI

-------
                           TABLE OF CONTENTS
                             LIST OF  BOXES


                                                                Page

Box 2.1      Estimating the Final Sample Size Required	2-15

Box 5.1      Steps for Generating Random Coordinates that Define
             Sampling  Locations	5-4

Box 5.2      An Example of Generating Random Sampling Locations	5-5

Box 5.3      Calculating Spacing Between Adjacent Sampling Locations	5-7

Box 5.4      Locating Systematic Coordinates	5-9

Box 5.5      Alternative Method for Locating the Random Stan Position
             for a Systematic Sample	5-11

Box 6.1      Calculating Sample Mean, Variance, Standard Deviation,
             and Coefficient of Variation	6-3

Box 6.2      An Alternate Approximation for &	6-6

Box 6.3      Formulae for Calculating the Sample Size Needed to
             Estimate the Mean	6-7

Box 6.4      Example of Sample Size Calculations	6-8

Box 6.5      Example: Determining  Sample Size for Testing the Mean
             Using the Power Curves	6-9

Box 6.6      Determining the Approximate Power Curve for a Specified
             Sample  Size	6-10

Box 6.7      Computing the Upper One-sided Confidence Limit	6-11

Box 6.8      An Example Evaluation of Cleanup Standard Attainment	6-12

Box 6.9      Calculating the Proportion of the Volume of Soil	6-13

Box 6.10     Calculating  Desired Sample  Size for Each Stratum of a
             Stratified Random Sample	6-14

Box 6.11     An Example Sample Size Determination for a Stratified
             Sample	6-15

Box 6.12     Formula for the Mean Concentration from a Statified Sample....6-16

Box 6.13     Formula for the Standard Error from a Stratified Sample	6-17

Box 6.14     Formula for  Degrees of Freedom from a Stratified Sample	6-17
                                     xu

-------
                            TABLE OF CONTENTS


                                                                 Page

Box 6.15     Formula for the Upper One-sided Confidence Interval from a
             Stratified Sample	6-18

Box 6.16     An Example Illustrating the Determination of Whether the
             Mean from a Stratified Sample Attains a Cleanup Standard	6-19

Box 6.17     Estimating the Mean, the Standard Error of the Mean, and
             Degrees of Freedom When a Systematic Sample Is Treated
             as a  Stratified  Sample	6-24

Box 6.18     Formula for Upper One-sided Confidence Interval for the
             True Mean Contamination, When a Systematic Sample Is
             Treated as a Stratified Sample	6-24

Box 6.19     Computational Formula for Estimating the Standard Error
             and Degrees of Freedom from Samples Analyzed in a
             Serpentine  Pattern	6-26

Box 7.1      Illustration of Multiple Measurement Procedure for Reducing
             Laboratory Error 	7-4

Box 7.2      Computing the Sample Size When Testing a Proportion or
             Percentile	7-6

Box 7.3      Example of How  to Determine Sample  Sizes When
             Evaluating Cleanup Standards Relative to a Proportion	7-7

Box 7.4      Calculating the Proportion Contaminated and the Standard
             Error of the Proportion	7-8

Box 7.5      Calculation of the Upper Confidence Limit on a Proportion
             Using a Large Sample Normal  Approximation	7-9

Box 7.6      An Example of Inference Based on the Exact Test	7-11

Box 7.7      Computing the Sample Size for Stratum h	7-16

Box 7.8      Sample Size Calculations for Stratified Sampling	7-17

Box 7.9      Calculating an Overall Proportion of Exceedances and the
             Standard Error of the Proportion from a Stratified Sample 	7-18

Box 7.10     Calculating the Upper Limit of the One-sided Confidence
             Interval on an Estimate of the Proportion	7-19

Box 7.11     Inference for Proportions Using Stratified Sampling	7-20

Box 7.12     Calculating the Sample Size Requirements for  Tolerance
             Intervals	7-22

Box 7.13     Calculating Sample Size for Tolerance  Intervals—Two
             Examples	7-23


                                     xiii

-------
                            TABLE OF CONTENTS

                                                                   Page
Box 7.14     Calculating the Upper Tolerance Limit	7-24
Box 7.15     Tolerance Intervals: Testing for the 95th Percentile with
             Lognormal Data  	7-25
Box 8.1      Defining Acceptance and Rejection Criteria for the Sequential
             Test of Proportions  	8-6
Box 8.2      Deciding When the Site Attains the Cleanup Standard	8-7
Box 8.3      An Example of Sequential Testing	8-9
Box 8.4      Example of Sequential Test'Using Grouped Samples	8-10
Box 9.1      Approximating  the Sample Size When Area and Grid
             Interval Are Known	9-7
Box 10.1     Steps for Obtaining Geostatistical Software from EMSL-LV ..10-12
                                      xiv

-------
               AUTHORS AND  CONTRIBUTORS
             This manual represents the combined efforts of several organizations and
many individuals.  The names of the primary contributors, along with the role of each
organization, are summarized below.

Dynamac Corporation, 11140 Rockville Pike, Rockville, MD  20852 (subcontractor to
Westat) — sampling, treatment, chemical analysis of samples. Key Dynamac Corporation
staff included:

                   David Lipsky              Wayne Tusa
                                Richard Dorrler

EPA, OPPE, Statistical Policy Branch — project management, technical input, peer
review. Key EPA staff included:
                                Barnes Johnson

SRA Technologies, Inc., 4700 King Street, Suite 300, Alexandria, VA  22302, EPA
Contract  No. 68-01-7379, Task  11  -- editorial and graphics support, final report
preparation. Key SRA staff included:

                   Marcia Gardner            Jocelyn Smith
                   Sylvia Burns              Lori Hidinger

Westat, Inc., 1650 Research Boulevard, Rockville,  MD  20850, Contract  No.  68-01-
7359, Task 5 - research, statistical procedures, draft report. Key Westat staff included:

                   John Rogers              Jill Braden
                   Paul Flyer                Ed Bryant
                                  AdamChu
                                          xv

-------
                      EXECUTIVE  SUMMARY
             This document provides regional project managers, onsite coordinators, and
their contractors with sampling and analysis methods for evaluating whether a soils
remediation effort has been successful.  The verification of cleanup by evaluating a site
relative to a cleanup standard or applicable and relevant or appropriate requirement (ARAR)
is discussed in section  121 of the Superfund Amendments and Reauthorization Act
(SARA). In section 121 of SARA the "attainment" of cleanup standards and ARARs is
mentioned repeatedly. This manual, the first in a series, provides a technical interpretation
of what sampling and data analysis methods are acceptable for verifying "attainment" of a
cleanup standard in soils and solid media.

             Statistical methods are emphasized because there is a practical need to make
decisions regarding whether a site has met a cleanup standard in spite of uncertainty. The
uncertainty arises because Superfund managers are faced with being able to sample and
analyze only a small portion of the soil at the site yet having to make a decision regarding
the entire site. Statistical methods are designed to permit this extrapolation from the results
of a few samples to a statement regarding the entire site.

             The methods in this document approach cleanup standards  as having three
components that influence the overall stringency of the standard.  The first component is
the magnitude, level, or concentration that is deemed protective of public health and the
environment.  The second component of the standard is the sampling  that is done to
evaluate whether a site is above or below the standard. The final component is how the
resulting data are compared with the standard to decide whether the remedial action was
successful.  All three of these components are important.  Failure to address any of the
three components can result in far less cleanup than desired. Managers must look beyond
the cleanup level and explore the sampling and analysis that will allow evaluation of the site
relative to the cleanup level.

             For example, suppose that a cleanup level is chosen and that only a few
samples are acquired.  When the results are available, it is found that the mean of those
samples is just below the cleanup level, and therefore, the site is judged  as having been
successfully remediated.  Under this scenario, there may be a large chance that the average
                                       xvi

-------
                            EXECUTIVE SUMMARY
of the entire site, as opposed to the samples, is well above the cleanup level. Uncertainty
was not considered, and therefore, there is a large chance that the wrong decision was
made and the site-wide average is not below the cleanup level.

              These concepts and solutions to the potential pitfalls are presented in a
sequence that begins with an introduction to the statistical reasoning required to implement
these methods. Then the planning activities are described; these require input from both
nonstatisticians and statisticians.  The statistical aspects of field sampling are presented.
Finally, a series of methodological chapters are presented which consider the cleanup
standard as: (1) an average condition; (2) a value to be rarely exceeded; (3) being defined
by small discrete hot spots of contamination that should be found if present; or (4) broad
areas that should be defined and characterized.   A more  detailed discussion of the
document follows.

              Chapter One introduces the  need for the guidance and its application with
risk-based standards, under various soils treatment alternatives, and in various parts of the
Superfund program. Standards development and usage depends on certain factors, and the
three categories of standards used by EPA are discussed:  technology-based, background-
based, and risk-based standards.

              The statistical methods described in this manual are useful in various phases
of treatment, testing, piloting, and  full-scale implementation of various treatment
technologies.  In addition, the methods in this manual  apply in various programmatic
circumstances including both Superfund and Enforcement lead sites and removal actions.

              Chapter Two addresses statistical concepts as they relate to the evaluation of
cleanup attainment  Discussions of the form of the null and alternative hypothesis, types of
errors, statistical power curves, and special data like less-than-detection-limit values and
outliers are presented.

              A site manager inevitably confronts the possibility of error in evaluating the
attainment of the cleanup standard:  is the site really contaminated because a few samples
are above the standard?  Conversely, is the site really "clean" because the sampling  shows
the majority of the samples to be within the cleanup standard?  The statistical methods
demonstrated in the guidance document allow decision making under uncertainty and valid
                                       xvii

-------
                            EXECUTIVE SUMMARY
extrapolation of information that can be defended and used with confidence to determine
whether the site meets the cleanup standard.

              The procedures in this  guidance document  favor protection  of the
environment and human health. If uncertainty is large or the sampling inadequate, these
methods conclude that the sample area does not attain the cleanup standard. Therefore, the
null hypothesis, in statistical terminology, is that the site does not attain the cleanup
standard until sufficient data are acquired to prove otherwise.

              Chapter Three discusses the steps in specifying attainment objectives.
Definition of the attainment objectives is the first task in the evaluation of whether a site has
attained a cleanup standard. Attainment objectives are not specified by statisticians, but
must be provided by  risk assessors, engineers, and soil scientists. Specifying  attainment
objectives includes specifying the chemicals of concern and the cleanup levels, as well as
choosing the area to be remediated.

              Chapter Four presents approaches to the design of remedial verification
sampling and analysis plans. The specification of this plan requires consideration of how
the environment and  human health are to be protected and how the sampling and analysis
are to achieve adequate precision at a reasonable cost.

              Sampling designs considered in  this guidance document are random
sampling, stratified sampling, systematic sampling, and sequential sampling. Differences
in these approaches, including advantages and disadvantages, are both discussed and
graphically displayed. With any plan, the methods of analysis must be consistent with the
sample design.

              A primary objective of the analysis plan involves making  a decision
regarding how to  treat the applicable cleanup standard. For example, is the cleanup
standard a value that  should rarely be exceeded (1 or 5 percent of the time) at a remediated
site? Or, alternatively, should there be high confidence that the mean of the site is below
the cleanup standard?  Should  there be no  hot spots with concentrations in excess of the
cleanup standard? Or should the analysis plan employ a combination of these criteria.
                                       xvm

-------
                            EXECUTIVE SUMMARY
             Chapter Five discusses the statistical aspects of field sampling procedures.
The procedures used to establish random and systematic sample locations are discussed.  In
addition to selecting sampling locations, the advantages and disadvantages of methods for
subsampling across depth are discussed and illustrated.  Three approaches presented are
depth discrete sampling, compositing across depth, and random sampling across depth.

             Chapter Six  describes  procedures  for  determining  whether there is
confidence, based on  the results of a set of samples, that the mean concentration of the
contaminant in a sample area is less than the cleanup standard. Basic formulas are given
and used in examples to illustrate the procedures. The primary point is that to ensure with
confidence that the site mean is below the cleanup standard, the sample mean must be well
below the cleanup standard by a distance determined by a confidence limit.

             The following topics—determination of sample size; calculation of the mean,
standard deviation, and confidence interval;  and deciding if the sample area attains the
cleanup standard-are discussed for these three sampling plans:

             •       Simple random sampling;
             •       Stratified random sampling; and
             •       Systematic sampling.

             Chapter Seven presents several approaches that allow evaluation of whether
a specified proportion or percentage of soil at a hazardous waste site is below the cleanup
standard. The methods described apply if there is interest in verifying that only a small
proportion or percentage of the soil at the site exceeds the cleanup standard.

             One way to implement these methods is to use simple exceedance rules.  A
sample size and number of exceedances are specified that coincide with an acceptable level
of certainty and level of cleanup. If the prespecified number of samples is obtained and the
number of exceedances is less than or equal to the allowed number of exceedances the site
is judged clean.  If there are  more exceedances than allowed then cleanup cannot be
verified. The more exceedances allowed, the more soil samples that need to be collected to
maintain the statistical performance of the method.
                                       xix

-------
                            EXECUTIVE SUMMARY
              Chapter Eight deals  with sequential sampling as a method for testing
percentiles. Unlike the fixed-sample-size methods discussed in the two previous chapters,
with sequential sampling, a statistical test is performed after each sample or small batch of
samples is collected and analyzed.  The test then makes one of three decisions: the site has
attained the cleanup standard, the site has not attained the cleanup standard, or select
another sample.  Sequential sample findings can respond quickly to very clean or very
contaminated sites and therefore offer cost savings. Although these procedures yield a
lower sample size on the average than that for fixed-sample-size procedures, in order to be
practical, they require "rapid turn-around" laboratory methods.

              Chapter Nine illustrates the design of sampling  plans to search for  hot
spots.  The conclusions that can be drawn regarding the presence or absence of hot spots
are discussed.  Hot spots are generally defined as relatively small,  localized, elliptical areas
with contaminant concentrations in excess of the cleanup standard. Tables are provided to
help determine grid spacing  and detect hot spots of various sizes with different
probabilities.

              Chapter Ten discusses the use of geostatistical methods, which provide a
method for mapping spatial data that enables both interpolation between existing data points
and a method for estimating the precision of the interpolation. Geostatistical applications
are described as a two-step process.   First, the  spatial  relationship  is modeled as a
variogram and then the variogram is used by a kriging algorithm to estimate concentrations
at points that were not sampled.  Indicator and probability kriging are most useful for
remedial verification purposes.

              Geostatistical methods  have many applications  in soil  remediation
technology, especially when the extent of contamination needs to be characterized.  This
chapter includes guidance to help decide whether geostatistical data analysis and evaluation
methods are appropriate for use with soils remediation activities that  involve removal,
homogenization, and flushing.

              Before being applied the kriging techniques will require further study on the
part of the user.  Reference documents are listed. Because kriging cannot be conveniently
or practically implemented without a computer and the appropriate software, a first-level
familiarity with the  methodology along  with use of a software package is desirable to
                                        xx

-------
                           EXECUTIVE SUMMARY
explore example applications and data sets.  EPA has developed the first version of a
geostatistical software package which can be obtained by following instructions at the end
of Chapter 10.
                                      xxi

-------
                         1.   INTRODUCTION
             Congress revised the Superfund legislation in the Superfund Amendments
and Reauthorization Act of 1986 (SARA).  Among other provisions of SARA, section 121,
Cleanup Standards, discusses criteria for selecting applicable and relevant or appropriate
requirements (ARARs) and includes specific language that requires EPA cleanups to attain
ARARs.

             Neither SARA nor EPA regulations or guidances specify how to determine
attainment or verify that the cleanup  standards have been met.  This document offers
procedures that can be used to determine whether, after a remediation action, a site has
attained an appropriate cleanup standard.


1.1          General Scope and Features of  the Guidance Document

1.1.1       Purpose

             This document describes methods for testing  whether soil chemical
concentrations at a site are statistically below a cleanup standard or ARAR.  If it can be
reasonably concluded that the remaining soil or treated soil at a site has concentrations that
are statistically less than relevant cleanup standards then the site can be judged protective of
human health and the environment. Figure 1.1  shows the steps involved in this evaluation
which requires specification of attainment objectives, sampling protocols, and analysis
methods.

             For example, consider the situation where several samples were taken. The
results indicate that one or  two  of the samples exceed the standard: How should this
information be used to decide whether the standard has been attained?  Some possible
considerations include:  the mean of those samples could be compared with the standard;
the magnitude of the two sample values that are larger than the standard might be useful in
making a decision; or the area where  the two large sample values were obtained might
provide some insight. The following factors are important in reaching the decision as to
whether a cleanup standard has been attained:
                                      1-1

-------
                       CHAPTER 1: INTRODUCTION
Figure 1.1    Steps in Evaluating Whether a Site Has Attained the Cleanup Standard



Clean Up All or Part of
the Sample Area
•
'
Reassess Cleanup
Technology
'
'
Identify Areas of High
Contamination
Chapters 6, 7, 8, 9 & 10
4
k

^


C Start J
4
Define Attainment
Objectives
Chapter 3
4
Specify Sample Design
and Analysis Plan
Determine Sample Size
Chapters 4, 6, 7, 8 & 9
i
Collect Data
Chapter 5
4
Determine If the
Sample Area Attains the
Cleanup Standard
Chapters 6, 7, 8, 9 & 10
/ Is the Ny
/ Cleanup \.





X Standard /
No \ Attained? /
Tves


\~ End J

                                     1-2

-------
                        CHAPTER 1: INTRODUCTION


             •      The spatial extent of the sampling and the size of the sample area;
             •      The number of samples taken;
             •      The strategy of taking samples; and
             •      The way the data are analyzed.

             Simply to require that  a  Superfund  site be  cleaned until  the  soil
concentration of a chemical is below 50 mg/kg is incomplete.  Statements suggesting that
the site will be remediated until the soil concentration of a chemical is 50 mg/kg reveal little
in terms of the environmental results anticipated, the future exposure expected, the resultant
risk to the local population, or the likelihood that substantial contamination will remain after
a decision is made that the site has been fully remediated.  A  specific sampling and data
analysis protocol must accompany the risk-based standard  for the standard to be
meaningful in terms of benefit or actual risk.

             This document does not attempt  to suggest which standards apply or when
they apply (i.e., the "How clean is clean?" issue).  Other Superfund guidance documents
(e.g., USEPA,1986c and USEPA, 1986d) perform that function.


1.1.2        Intended Audience  and Use

             Management/supervisory personnel will find the executive summary and
introductory chapters useful.  However, this manual is intended primarily for Agency
personnel (primarily onsite coordinators and  regional project managers), responsible
parties, and  their contractors who are involved with monitoring the progress of soils
remediation  at Superfund sites.  Although selected introductory statistical concepts are
reviewed, the document is directed toward  readers that have  had prior training or
experience applying quantitative methods.

             This document discusses data analysis and statistical methods for evaluating
the effectiveness of Superfund remedial actions. However, there are many other technical
aspects to this problem. Input from soil scientists,  engineers,  geologists, hydrologists,
geochemists, and analytical chemists is essential.  There must be dialogue among this
group,  including the statistician, so that each member understands and considers the point
                                       1-3

-------
                         CHAPTER 1: INTRODUCTION
of view of the others. It is only through collaboration that an effective evaluation scheme
can be developed to measure the effectiveness of a remedial action.

             This document does not intend to address the issues that the other members
of the team specialize in such as:

             •      Soil sample acquisition protocols;
             •      Areas of the vadose zone of concern under different situations;
             •      The influence of soil chemistry;
             •      Waste types based on industrial processes;
             •      Leaching procedures that approximate the expected weathering
                    processes and risk assessment assumptions;
             •      Chemical analysis methods useful given particular soils matrices; or
             •      Approaches to soils remediation.

             Table 1.1 lists other relevant EPA guidance documents  on sampling and
evaluating soils and solid  media that apply to both the statistical and other technical
components of a sampling and analysis program.

             The selection of  statistical methods for use in assessing  the attainment of
cleanup standards depends  on the characteristics of the data.  In soils, concentrations of
contaminants change relatively slowly, with little variation from season to season.  In
ground water, the number of measurements available for spatial characterization is limited
and seasonal patterns may  exist in the data.  As a result of these differences,  separate
procedures are recommended for the differing problems associated with soils and solid
media, and  ground water,  surface water, and air.  These media  will be addressed in
separate volumes.
                                        1-4

-------
                       CHAPTER 1:  INTRODUCTION
Table 1.1
Tide
EPA guidance documents that present methodologies for collecting and
evaluating soils data
                    Sponsoring
                    Office
Date
ID
Number
Preparation of Soil
Sampling Protocol:
Techniques and
Strategies

Soil Sampling Quality
Assurance User's Guide

Verification of PCB
Spill Cleanup by
Sampling and Analysis

Guidance Document for
Cleanup of Surface
Impoundment Sites

Test Methods for
Evaluating Solid
Waste

Draft Surface Impoundment
Clean Closure Guidance
Manual

Data Quality Objectives
for Remedial Response
Activities: Development
Process

Data Quality Objectives
for Remedial Response
Activities: Example
Scenario RI/FS Activities
at a Site with Contaminated
Soils and Ground Water
                    EMSL-LV
                    ORD

                    EMSL-LV
                    ORD
                    OTS
                    OPTS
                    OERR
                    OSWER
                     OSW
                     OSWER
                     OSW
                     OSWER
                     OERR
                     OSWER
August
1983

May
1984
August
1985
June
1986
November
1987
March
1987
March
1987
EPA 600/
4-83-020

EPA 600/
4-84-043
EPA 5607
5-85-026

OSWER
DIRECTIVE
9380.0-6
SW-846

OSWER
DIRECTIVE
9476.0-8.C
 EPA 5407
 G-87/003
                     OERR
                     OSWER
 March
 1987
 EPA 5407
 G-87/004
             It must be emphasized that this document is intended to provide flexible

guidance and general dkection. This manual is not a regulation and should not be imposed

as a regulation.   Finally, this document should not  be  used as a "cookbook" or a

replacement for engineering judgment.
                                     1-5

-------
                         CHAPTER 1: INTRODUCTION
1.1.3       Bibliography,  Glossary,  Boxes,  Worksheets,  Examples, and
             References to "Consult a Statistician"
             The document includes a bibliography which provides a point of departure
for the user interested in further reading.  There are references to primary textbooks,
pertinent journal articles, and related guidances.

             The glossary is  included to provide short, practical definitions of
terminology used in the manual.  The glossary does not use  theoretical explanations or
formulae and should not be considered a replacement for more complete discussions in the
text or alternative sources of information.

             Boxes are used throughout the  document to separate  and highlight
calculation methods and example applications of the methods. A listing of all boxes and
their page numbers is provided on pages xii - xiv.

             A series of worksheets is included to help organize calculations. Reference
to the pertinent sections of the document appears at the top of each worksheet.

             Example data and calculations are presented in the boxes and worksheets.
The data and sites are hypothetical, but elements of the examples correspond closely to
actual sites.

             Finally, the document often directs the reader to "consult a statistician"
when more difficult and complicated situations are encountered. A directory of Agency
statisticians is available from the Statistical Policy Branch (PM-223) at EPA Headquarters.
1.2          A Categorization Scheme for Cleanup Standards

              Superfund remediations require standards for assessing the success and
completion of the cleanup. The criteria for choosing the type of standard and setting the
magnitude of the standard come from different sources, depending on many factors
including the nature of the contamination, negotiations with potentially responsible parties,
and public comment on alternatives identified by EPA.
                                       1-6

-------
                         CHAPTER 1:  INTRODUCTION
             Many programs throughout EPA  use numerical standards variously
described as ARARs, concentration limits, limitations, regulatory thresholds, action levels,
and criteria. These standards are often expressed as concentration measures of chemicals
or chemical indicators. Standards development and usage depends on the media to which
the standard applies, the data used to develop the standard, and the manner of evaluating
compliance with the standard. The following discussion categorizes the standards used by
EPA and compares the features of each category.
1.2.1       Technology-Based Standards

             Technology-based standards are developed for the purpose of defining the
effectiveness of pollution abatement technology from an engineering perspective.  For
example, waste water treatment plants operating under the National Pollution Discharge
Elimination System (NPDES) must be designed and operated under a numerically
prescribed level of technological performance depending on  the particular industrial
category.  Technology-based standards such as the NPDES standards are developed and
applied using statistical methods that consider variability in the operation of the treatment
system. The likelihood of exceeding the standards is rare if the technology is installed and
operated properly.  Often Superfund sites require the installation of waste water treatment
systems and compliance with NPDES standards.


1.2.2       Background-Based Standards

             Background-based standards are developed using site-specific background
data.  An example is the background ground water concentration standards that hazardous
waste land disposal facilities use under Resource Conservation and Recovery Act (RCRA)
permits.  The background data are used to establish a standard  for the facility, which
accounts for the presence of any existing contamination hydraulically upgradient of the
facility. Background standards are applied on a site-specific basis, but because they are
developed using statistical methodologies, the standards can be associated with a known
false positive and false negative rate.
                                       1-7

-------
                         CHAPTER 1: INTRODUCTION
1.2.3       Risk-Based Standards

             A third class of standards, risk-based standards are developed using risk
assessment methodologies. Chemical-specific ARARs adopted from other programs often
include at least a generalized component of risk. However, risk standards may be specific
to a site, developed using a local endangerment evaluation.

             Risk-based standards are expressed as a concentration value.  However,
cleanup standards based on risk as applied in the Superfund program are not associated
with a standard method of interpretation when applied in the field.  Therefore,  risk-based
standards, when applied  in the field, do not consider false positive and false negative
errors. Although statistical methods are used to develop elements of risk-based standards,
the estimated uncertainties are not carried through the analysis or used to qualify the
standards for use in a field sampling program.   Even though risk standards are not
accompanied by measures of uncertainty, field data, collected for  the purpose of
representing the entire site and validating cleanup, will be uncertain.  This document allows
decision making regarding site cleanup by providing methods that statistically compare risk
standards with field data in a scientifically defensible manner that allows for uncertainty.


1.3         Use of this Guidance in Superfund Program Activities

             Standards  that apply to Superfund activities normally fall into the third
category of risk-based standards. There are  many Superfund activities where risk-based
standards might apply.   The following discussion provides suggestions for using the
methods described in this document in the implementation and evaluation of Superfund
activities.
1.3.1        Emergency/Removal  Action

              Similar to the guidance regarding sampling strategies associated with PCB
spills (USEPA, 1985), cleanup activities associated with the methods in this document will
be useful for circumstances that are encountered during emergency cleanups and removals.
In many cases, because of the time, safety, and exposure constraints associated with
emergency activity, initial cleanup will focus on areas visually or otherwise known to be
                                       1-8

-------
                        CHAPTER 1: INTRODUCTION
contaminated.  The methods described in this document will, however, be most useful in
verifying the initial cleanup of obvious contamination.


1.3.2       Remedial  Response  Activities

             The  objective of remediation is to ensure that release of and exposure to
contaminants is curtailed. Remedial efforts are normally long-term and require diverse,
innovative technology. As discussed in section 1.4, soil or solid media remediation can be
addressed using a variety of technologies.  Numerical standards are used to define the
degree of curtailment. The methods described in this document can help to evaluate the
utility of the remediation technology in treating contaminants with respect to a particular
numerical standard.


1.3.3        Superfund Enforcement

              The methods described in this document will also be useful for providing
more technically exacting negotiations, consent decree stipulations, and responsible party
oversight. Questions such as "How much is enough?" and "When can I stop cleaning?"
are constantly emerging at the enforcement negotiation table. More specific questions such
as "How much should you sample?", "What sampling pattern or method of sampling
design should be applied?" and "How can I minimize the chance of saying the site is still
dirty when it is basically clean?" are addressed here, as well as the ultimate question: "How
do I know when the standard has been attained at the entire site, knowing that the decision
is based on a body of data that is incomplete and uncertain?"


1.4          Treatability Studies and Soils  Treatment Technologies

              In addition to discussing the methods described in this document and their
relationship to aspects of the Agency's Superfund program, it is also important to discuss
how the methods will function when applied in treatability studies and under various soils
treatment technologies.
                                       1-9

-------
                         CHAPTER 1:  INTRODUCTION
1.4.1        Laboratory/Bench-Scale  Treatabiiity Studies

              Feasibility studies often include small bench-scale laboratory evaluations of
how various treatment agents and concentrations of agents will perform. Suppose that the
contaminant and soil characteristics at the site indicate that two fixation media offer a
promising remediation approach. A treatability study examining several concentrations of
the two media is proposed.

              Under this scenario, the methods described in this manual could be applied
to the sampling program used to obtain soils material for the treatability study. Treatability
studies require "worst case" material—that is,  soils with the highest concentrations or with
the most tightly bound contaminants. Therefore, "worst case" sample areas within the site
must be delineated, using data from prior remedial investigations. Once the "worst case"
sample area is defined, the soils can be sampled as described in this manual, the treatability
study executed, and the resulting data analyzed using the methods described in this
document to examine whether the method has sufficiently treated the soil to allow
attainment of the relevant cleanup standard.


1.4.2        Field/Pilot-Scale  Treatability  Studies

              Once the feasibility study establishes an effective approach to treatment, it
may be implemented as an onsite pilot using the chemical/physical/biological remedy with
construction-scale onsite machinery. The approach favored in the bench-scale laboratory
experiment may be chosen if the cost is reasonable. Machinery such as cement mixers, soil
washers, soil mixing augers,  incinerators,  vacuum extraction manifold  networks, or
infiltration or injection  systems are used  in a pretest fashion.  With an associated
monitoring program, the methods in this guidance can be applied to determine whether the
method will attain the desired level of cleanup.

              The primary difference between the laboratory testing results and those
obtained from field-scale pilot application is that far greater variability will be encountered
in the onsite pilot.  Unless the treatment method is exceptionally effective relative to risk-
based standards in the laboratory, the variability encountered in the field may obscure the
treatment's effectiveness. This document guides the user to methods that will help in such
                                       1-10

-------
                         CHAPTER 1: INTRODUCTION
a situation. In addition, if a reasonable sampling program is conducted at the pilot-scale,
these data can be used to estimate sample sizes for the sampling program associated with
the full-scale implementation of the technology.


1.4.3        Soils Treatment  by Chemical Modification

             Soils are often treated by chemical fixation or stabilization.  This technology
uses a cement or grout-like material mixed with the contaminated soil or sediment. Once
the mixture reacts, it solidifies, and contaminants are retained in the matrix and resist
leaching.  When this technology is used, the methods in this manual  can be  applied,
keeping in mind, however, the following caveats.

              Once the material  has solidified onsite, it cannot be sampled easily. The
ability to stabilize the site may be compromised if cores were obtained throughout the area.
In addition, the resulting monolith may  be capped, which would restrict access to the
solidified matrix. Because it cannot be sampled after fixation, monitoring plans should be
developed before the mixing occurs.  The sampling could occur by taking samples at
randomly located positions across the site  and then pouring cylindrical casts of the material
immediately after it is mixed prior to setup. Enough casts must be obtained for the initial
evaluations of the  site and for monitoring the aging process of the stabilized material.
During analysis, concentrations are measured in leachate obtained from an accepted
extraction procedure.  Evaluation of the  leachate from the casts allows determination of
whether the lithified material attains or continues to attain the relevant cleanup standard.


1.4.4        Soils Treatment by In  Situ Removal of Contaminants

              Several soil treatment technologies, including vacuum  extraction, soils
leaching, and bioremediation, remove the contaminants without massive soil movement.
The efficacy of these systems can be evaluated using the methods in this document, with
the exceptions noted below.

              Vacuum extraction is used to remove volatile compounds. Ambient air is
drawn  down through the soil into a well network and then into  an adjustable manifold
                                      1-11

-------
                         CHAPTER 1:  INTRODUCTION
system attached to a vacuum pump.  Air is then sent through carbon columns to remove the
volatile compounds.

              Soils leaching technologies are generally designed to extract contaminants
that are water soluble.  Soils leaching also relies on a network of wells attached to a
manifold system.  The system includes infiltration areas where aqueous solutions are
allowed to recharge into the soils system.  A pumping system is attached to the manifold
and the water, after migration from the infiltration area to each well, is extracted and sent to
a waste water treatment system.

              Soils bioremediation can be used to degrade contaminants. Microorganisms
use the contaminant as an energy source.  One or more injection wells introduce water
possibly enriched  with oxygen,  nutrients, microorganisms, or other essential growth-
promoting materials. The injection wells are installed on one side of the contaminated area
and monitoring wells are installed in various patterns throughout and possibly beyond  the
area of contamination. Again, a manifold system might be used for injection or sampling,
and extraction wells may be used to direct or improve water movement.

              With these technologies, something other than direct soils sampling may be
used to evaluate effectiveness of the remediation, for example mass balance  differencing.
In this case, the methods herein may not always apply.  However, monitoring of the soil
relative to a risk-based cleanup standard is the most direct and protective measure  of any
soils cleanup technology.

              Another concern is that when these systems are in place, the above-ground
or slightly buried piping network will restrict the access of soils sampling equipment.  For
example, vehicle-deployed augers  may not be able to reach certain areas.  Engineering
specifications should call for easy disassembly whenever possible.  In cases where this is
not possible, the guidance can still be applied after exclusion of certain soil areas because of
inaccessibility.

              A third consideration is that, after implementing the soil remediation
technology, the soils concentration  profile may begin to take on a regular spatial pattern.
This occurs because removal wells are often arranged in a grid pattern and each well has a
zone of influence where the concentrations have been reduced substantially. The result is a
                                       1-12

-------
                         CHAPTER 1:  INTRODUCTION
series of areas with high and low concentrations across the site.  As discussed in the
sampling chapter, under these circumstances systematic sampling should not be used
because  all or many of the samples may be located in areas with high  or low
concentrations. Random sampling is recommended to avoid this problem.

             A final concern is that the soils system must be at steady state during the
sampling program. This requires shutting down the extraction process and allowing the
system  to return  to its  original balance.   This may take some time depending  on
characteristics of the system. In some cases when progress is being measured over time,
methods pertaining to ground water in Volume 2 of this series might be more appropriate.


1.4.5       Soils Treatment  by Incineration

             Soils incineration involves the burning of soils in a furnace at high
temperatures to degrade the contaminants into a  nontoxic form.  The product of the
incineration is an ash. If questions arise as to whether the ash material contains chemicals
in excess of applicable standards, then  this manual might be useful. Sampling will have to
be designed based on site-specific circumstances. If the treatment is highly effective and
uniform, only a few samples may be  necessary to verify effectiveness. However, if the
standard is quite low and the measurement technology is variable at low concentrations,
more samples may be required.


1.4.6       Soils  Removal

             In  the soils removal approach to site cleanup, soils are permanently or
temporarily removed from the site. Sampling must be done to verify that enough soil has
been removed, and to ensure that clean soil is not needlessly removed.  Under the
circumstances associated with soils removal, there is no homogenization of the soil through
a fixation process or artificial regularity to the soil profile caused by local extraction. In this
case, geostatistical applications (Chapter 10) are useful for characterizing the contaminant
profile.  A new concentration profile can be estimated with each succeeding layer that is
removed. In addition, geostatistical applications can help to identify hot spots that should
be removed and sampling and analysis to detect hot spots might be useful (Chapter 9).
                                      1-13

-------
                         CHAPTER 1: INTRODUCTION
Finally, the simpler, more conventional evaluation methods that comprise the bulk of this
manual can also be used. Exner et al. (1985) describe an application of these evaluation
methods to a soils removal scenario at a Superfund site with dioxin contamination.


1.4.7       Soils Capping

             A final category of soils remediation is to cap a site with impermeable layers
of clay and synthetic membranes. This prevents surface water from recharging to the
ground water through contaminated soils. Often caps are added as an additional measure in
conjunction with  other approaches.  The methods in this document can be used to
determine whether caps have met an engineering specification. For example, if the cap is
intended to be constructed with no more than a 10~~7 cms/sec permeability, samples might
be obtained to document that permeability has been attained.  Sampling may be difficult
because it might disturb the integrity of the cap; however, it is possible  that a pilot-scale
procedure could be implemented to verify attainment of the standard.


1.5         Summary

             This document deals with statistical methodology and procedures for use in
assessing whether, after remediation, the treated soil or remaining soil attain the cleanup
standards that are protective of public health and the environment as required by section
121 of SARA.

             Use of the  document is  intended  primarily  for Agency personnel,
responsible parties, and contractors who are involved with monitoring the progress of soils
and remediation at Superfund sites. Although selected introductory  statistical concepts are
reviewed, the document is directed toward users having prior training or experience in
applying quantitative methods.

             Important factors  in determining whether a cleanup standard has been
attained are:
             •      The spatial extent of the sampling and the size of the sample area;
             •      The number of samples taken;
                                      1-14

-------
                         CHAPTER 1: INTRODUCTION


             •      The strategy of taking samples; and
             •      The way the data are analyzed.

             The three types of EPA cleanup standards are technology-based standards,
background-based standards, and risk-based standards.  Superfund activities usually
employ risk-based standards.   By providing methods that statistically compare risk
standards with field data in a scientifically defensible manner that allows for uncertainty,
this document allows decision making regarding site cleanup. The statistical methods can
be applied to the implementation and evaluation of:

             •      Emergency/removal action,
             •      Remedial response activities, and
             •      Superfund enforcement.

Also discussed are the functions of the statistical methods described in the document in the
context of a variety of treatability studies and soils treatment technologies.
                                      1-15

-------

-------
    2.  INTRODUCTION TO  STATISTICAL  CONCEPTS
                            AND DECISIONS
              When it comes to verifying cleanup, suppose that no exceedances of the
cleanup standard are to be allowed.  In that case, one of the most frequently asked
questions regarding the use of statistical techniques in the evaluation of cleanup standards
is:
              Why  should I  use  statistical methods and  complicate  the
remedial verification  process?

              Allowing no exceedances of a standard is a perfectly acceptable decision
rule to use. In fact, that simple rule is a statistical procedure because errors are possible.
However, there is a chance that no exceedances will be discovered, yet a substantial portion
of the site is above the cleanup standard.  This is  clearly not a desirable environmental
result.  With small sample sizes the chance of missing contamination is greater than with
larger sample sizes. This is intuitive; the more you search for contamination and do not
find it, the more confident you become in your conclusion that the site is clean.

              Alternatively, consider the situation where a reasonable number of samples
is taken and (me sample exceeds the cleanup standard.  In this case, you would conclude
that the site continues to be dirty under the no exceedance rule. However, the problem is
that this conclusion may be in error. Either laboratory error occurred or some rare and
insignificant parcel of contamination could have been discovered.  Revisiting the remedial
method after many years or dollars of implementation is not reasonable because of the
possibility that an error was made. As  sample sizes are increased, the chances of finding
one of the few obscure samples  above the cleanup standard increases.  How can you
balance the two sets of possibilities: the chance that the site is contaminated even when the
sampling shows attainment of the cleanup standard, and the chance of contamination when
the  majority of samples taken show the site to be clean?

              The answer  is to evaluate the potential magnitude of these two errors and
balance them using the statistical strategies described in this manual.  Statistical methods
perform a powerful and useful function--they allow extrapolation from a set of samples to
the entire site in a scientifically valid fashion.
                                      2-1

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
             Consider the following circumstance.  The surface layer of soil from the
bottom of a 4-hectare lagoon at a Superfund site will be sampled using cores with a 4-cm
area. Given the size of the core and lagoon there will be approximately 10 million sample
locations; however, concentration measurements will only be  made on  100 of the 10
million. Statistical sampling and analysis methods provide an approach for choosing which
100 of the 10 million locations to sample so that valid results can be  presented and
statements can be made regarding the characteristics of the 10 million potential samples or
the entire site.

             Clearly, because of the extrapolation exercise, the statements or inferences
regarding the 10 million  sample locations  have uncertainty.  Statistical methods enable
estimation of the uncertainty. Without the statistical methods, uncertainty still exists; but
the uncertainty cannot be estimated validly.

             This  chapter will elaborate on statistical concepts and their specific
application to the evaluation of cleanup standards. Statistical concepts such as the form of
the null and alternative hypothesis, types of errors, statistical power, and handling peculiar
data structures like less-than-detection-limit values and outliers  are discussed to promote
understanding.  However, it is not necessary to read this chapter to apply the methods in
this manual.
2.1           Hypothesis Formulation and Uncertainty

              With any statistical procedure, conclusions will vary depending on which
soil sample locations are selected. Therefore, based on the data collected, the investigator
may conclude that:

              •      The site attains the cleanup standard;
                    The site does not attain the cleanup standard; or
              •      More information is required to make a decision with a specified
                    level of confidence.
                                        2-2

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
              When the results of the investigation are uncertain, the procedures in this
 guidance document favor protection of the environment and human health and conclude that
 the sample area does not attain the cleanup standard. In the statistical terminology applied
 in this document, the null hypothesis is that the site does not attain the cleanup standard.
 The null hypothesis is assumed to be true unless substantial evidence shows that it is false.
 Let <)> represent the  true (but unknown) value of a particular soil property,  such as the
 mean concentration of a specified chemical over the entire site. The null hypothesis is:

              HQ: $ > Cleanup Standard (CONTAMINATED or DIRTY),

 and the alternative hypothesis is:

              HI: < Cleanup Standard (CLEAN).

 This document describes  how  to gather and analyze data that will provide  evidence
 necessary to contradict  the null hypothesis and demonstrate that the site indeed attains the
 cleanup standard.  Figure  2.1 shows  how the null and alternative hypothesis change as
 contamination is detected and subsequently corrected. This illustration specifically pertains
 to ground water evaluations for land disposal facilities operating under the Resource
 Conservation and  Recovery Act  (RCRA), but the concept is similar for  the  soils
 contamination situation. Initially, the the null hypothesis is that there is no contamination
 (A-C). Once a statistical demonstration can be made that the downgradient concentrations
 are first above background-level concentrations (B) and also above a relevant action limit or
 other standard (D), then  the null  hypothesis is that the  site is contaminated.  Most
 Superfund sites that require cleanup  are in the situation described by D-E. The site must, at
 that point, be remediated (E,F) and proven to be clean (G)  before the null hypothesis as
 described above can be rejected and the site declared clean.

              If the null and alternative hypothesis described above were reversed, then a
 situation similar to C would designate a satisfactory cleanup.  As can be seen by comparing
 C  with G, the improper specification of the null and alternative hypothesis during a
corrective action can result  in very different levels of cleanup.
                                       2-3

-------
    CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
  Figure 2.1  A Statistical Perspective of the Sequence of Ground Water Monitoring
            Requirements Under RCRA
HI
    Upgradient       Downgradient
    (background)
                                          DETECTION
                                          MONITORING
                                          NO RELEASE


                                          TRIGGER
                                          COMPLIANCE
                                          MONITORING
                                          COMPLIANCE
                                          MONITORING
                                          TRIGGER
                                          CORRECTIVE
                                          ACTION
                                           CORRECTIVE
                                           ACTION
                                           BEGINS
                                           CORRECTIVE
                                           ACTION
                                           CONTINUES
                                           RETURN TO
                                           COMPLIANCE
                                           MONITORING
NULL  ;
HYPOTHESIS
CLEAN
                                                           ALTERNATIVE
                                                           HYPOTHESIS

                                                           CONTAMINATED
NULL
HYPOTHESIS

CONTAMINATED
 ALTERNATIVE
 HYPOTHESIS

 CLEAN
         CONCENTRATION
 (Notice that until contamination above a risk standard is documented (D) the null hypothesis is
 that the facility is clean. Once the facility has been proven to be in exceedance of a health criteria
 then the null hypothesis is that the facility is contaminated until proven otherwise (G).)
                                     2-4

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
              When specifying simplified Superfund site cleanup objectives in consent
 decrees, records of decision, or work plans, it is extremely important to say that the site
 shall be cleaned up until the sampling program indicates with reasonable confidence that the
 concentrations of the contaminants at the entire site are statisticallv less than the cleanup
 standard.  This prescription will result in the  site being designated clean only after a
 situation similar to G is observed.  However, attainment is often wrongly described by
 saying that concentrations at the site shall not exceed the cleanup standard.  This second
 prescription can result in a situation similar to C being designated as clean.

              As discussed in the introduction to this chapter, variation in sampling and
 lab analysis introduces uncertainty into the decision concerning the attainment of a cleanup
 standard.  As a result of the uncertainty and the null/alternative hypothesis arrangement
 discussed above, the site can be determined clean when, in fact, it is not, resulting in a
 false positive decision (or Type I error).  The converse of a false positive decision is a
 false negative decision (or Type II error), the mistake of saying the site needs additional
cleanup when, in fact,  it meets the standard.   The Greek letter alpha (a)  is used to
represent the probability of a false positive decision and beta ((3) is used to represent the
probability of false negative decision. The definitions above are summarized in Table 2.1.
Table 2.1      A diagrammatic explanation of false positive and false negative conclusions
               Decisi6n based on
               the sample data is:
                   Clean
                    Dirty
                                                The true condition is:
                                             Clean
                         Dirty
   Correct
 Power (1 - (5)
 False negative
(Probability is (3
  False positive
(Probability is a)
    Correct
Certainty (1 - a)
                                        2-5

-------
  CHAPTER 2:  INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
             It can be seen that if both a and P can be reduced, the percent of time that
the correct decision will be made will be increased Unfortunately, simultaneous reduction
usually can be achieved only by increasing sample size, which may be expensive.
2.2
            Power  Curves as  a Method of  Expressing Uncertainty  and
            Developing Sample Size Requirements
             The probability of declaring the sample area clean will depend on the
population mean concentration.  If the true population mean is above the cleanup standard
the sample area will rarely be declared clean (this will only happen if the mean of the
particular set of samples is by chance well below the cleanup standard). If the population
mean is much smaller than the cleanup standard, the sample area will almost always be
judged clean.  This relationship can be demonstrated by Figure 2.2. The figure illustrates
a power curve that shows the probability of deciding that the site attains the cleanup
standard on the vertical axis and the  true, but always unknown,  population mean
concentration on the horizontal axis.

Figure 2.2    Hypothetical Power Curve
Probability
of Deciding
  the Site
Attains the
  Cleanup
 Standard
                                                          Cleanup
                                                          standard
                                  0.4       0.6       0.8
                               Population Mean Concentration
                                      2-6

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
             If the population mean concentration in the sample area is equal to or just

above the cleanup standard (i.e., does not attain the cleanup standard), there is still a small

5-percent probability of declaring the sample area clean; this is the false positive rate
denoted by a.


             If the population mean is equal to 0.6 ppm (i.e., attains the cleanup standard

of 1.0 ppm), the probability of declaring the sample area clean is 80 percent. Conversely

the probability of declaring the site dirty, given that it is actually clean, is 20 percent. This

is the false negative rate for a population mean of 0.6 ppm.  Note that the probability of

declaring the site clean changes depending on the population mean. These false positive

and false negative rates are shown in Figure 2.3.
Figure 2.3     Hypothetical Power Curve Showing False Positive and False Negative
              Rates
               1
 Probability
 of Deciding
   the Site
 Attains the
   Cleanup
  Standard
               g   I False negative rate for
                  J a mean of .6 ppm = 20%
0.8
0.7 4-
0.6

0.5
0.4
0.3
0.2
0.1
  0
Power at M. i=80%
                    False positive rate
                    at the cleanup
                    standard = 5%
                                             =.6

Cleanup
standard
                         0.2      0.4       0.6       0.8       1
                              Population Mean Concentration, ppm
                                                           1.2
                                                       1.4
             The following items specify the shape and location of the power curve:

             •      The population coefficient of variation;

             •      The method of sample selection (the sampling plan);

             •      The statistical test to be used;
                                       2-7

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS


             •      The false positive rate; and
             •      The sample size.

             In summary, there are two important uses of power curves. The first is to
further facilitate understanding of the concept that, although the site may actually be clean, a
set of samples from the site can be obtained that suggest the site is dirty. The cleaner the
site, the less chance of this happening. Conversely, a site may be dirty, but the particular
set of samples suggest the site is clean.  Again the dirtier the site, the less chance of this
occurring.  The chances of these errors are controlled  by the position and shape of the
power curve relative to the cleanup standard. Figures A.I - A.4 illustrate several families
of power curves.  The ideal shape of a power curve is a step function that has a 1.0
probability of declaring the site clean whenever the true concentration is  less than the
cleanup standard and a zero probability of declaring the site clean when the concentration is
greater than the cleanup standard.

              The second use of a power curve is to help decide on an appropriate sample
size for a sampling program.  The lower the variability and the more samples taken, the
closer the power curve will come to approaching the ideal step function described above.
In addition, the trade-off between the false positive and negative rate influences the position
of the power curve. Use the power curves in Appendix A to assist with the sample size
determination process in one of two ways:

              •      Select the power curve desired for the statistical test and determine
                    from this the sample size that is required; or
              •      Select  the sample size  to be  collected and determine what  the
                   . resulting power curve will be for the statistical procedure.

              Chapters 6, 7, and 8 provide  specific methods for making sample size
determinations.


2.3          Attainment or Compliance  Criteria

              The characteristic of the chemical concentrations to be compared to the
cleanup standard must be specified in order to define a statistical test to determine whether a
sample  area  attains the cleanup standard.  Such  characteristics  might  be the mean
                                       2-8

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
concentration, the median, or the 95th percentile of the concentrations. In other words, it
must be decided whether the cleanup standard is intended to be applied as a mean value
such that the mean of the site must be below the cleanup standard or whether the cleanup
standard is a high percentile value that must rarely be exceeded at only 5 or 10 percent of
the site.  Figure 2.4 illustrates these characteristics on three distributions.  Section 3.5
offers a more detailed discussion of these parameters.
2.3.1        Mean

              The location or general magnitude of a set of data is often characterized by
the mean of the distribution. The mean of the concentration distribution is the value that
corresponds to the "center" of the distribution in the sense of the "center of gravity." In
determining the mean from a highly skewed lognormal distribution, small amounts of soil
with concentrations far above the mean are balanced by large amounts of soil with
concentrations close to, but below, the mean.

              Whether the mean is a useful summary of the distribution depends on the
characteristics of the sample area and the objectives of the cleanup. In a sample area with
uniform contamination and very little spread or range in the concentration measurements,
the mean will work well. If the spread in the data is large relative to the mean, the average
conditions will not adequately reflect the most heavily contaminated parts of the population.
If interest is in the average exposure or the chronic risk, the mean may be an appropriate
parameter.

              When using the mean, consideration should be given to the number of
measurements that are likely to be recorded as below the detection limit. With many
observations below the detection limit, the simple estimate of the population mean cannot
be calculated (see the discussion in section 2.5.2).


2.3.2        Proportions  or Percentiles

              High percentiles or proportions pertain  to the tail of a distribution and
control against having large concentration values. The 50th percentile, the median, is often
a useful alternative to the mean.
                                       2-9

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
Figure 2.4    Measures of Location: Mean, Median, 25th Percentile, 75th Percentile, and
              95th Percentile for Three Distributions

-1
0
                             Hypothetical Distribution
                      2345
                       Concentration ppm
                 12345
                       Concentration ppm
                            Lognormal Distribution
                 12345
                        Concentration ppm
                                                          Legend:
                                                            Measures of Location:
                                                                   25th Percentile
                                                       Median (50th Percentile)


                                                       Mean


                                                       75th Percentile


                                                       95th Percentile
Measure of Spread:

    ± 1 Standard Deviation
       Around the Mean

      «	H	
                                        2-10

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
              Methods are available for proportions that are unaffected by concentrations
 below the detection limit, as long as the detection limit is below the cleanup standard. The
 likelihood of having many data values below the detection limit makes the proportion of
 soil units above the cleanup standard an appealing parameter to use in assessing attainment.
 If the cleanup standard is only slightly above the detection limit, then it will always be
 possible to calculate the proportion of soil samples above the cleanup standard.

              Knowing the maximum concentration of the hazardous contaminant at a
 waste site would be helpful in making decisions. Unfortunately, in realistic situations the
 maximum cannot be determined from a sample of data.  A test of proportions, using an
 upper percentile of the concentration distribution, can serve as a reasonable approximation
 of the maximum value.
2.4          Components of a Risk-Based  Standard

              Chapter 1 introduced the concept of a risk-based standard and its application
to Superfund activities.  Here we will describe how statistical sampling and analysis
methods can be used to adjust the stringency of a risk-based standard.

              A hypothetical example of a risk-based standard is as follows:  a soil
concentration of arsenic greater than 20 ug/kg at a specific smelter subjects workers to a 1
in a million chance of oral cancer during a lifetime.  It is commonly thought that the only
way to change the stringency of the 20 ug/kg standard is to change the magnitude of the
number, 20.  In other words, a less stringent standard is obtained by changing the risk-
based standard to 25 ug/kg with an associated increase in the probability of acquiring oral
cancer. This is true, but there are other ways to influence the stringency of the standard.

              There  are three components of a risk-based standard that can be used to
adjust its stringency.  Bisgaard and Hunter (1986) provide discussion of these components
and their application.  The three components are:

              1)     The magnitude of the Concentration  Threshold Level (Cs):
             2)     The method for obtaining data or the Sampling Plan: and
                                      2-11

-------
  CHAPTER 2:  INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
             3)    The evaluation scheme, Decision Rule, that will be used to compare
                   the data with the threshold level.

Figure 2.5 illustrates the relationship among these components. The choice of a numerical
level  is one element  of a risk standard.  Other  questions must also be answered
regardingsampling: How many samples?  In what area will the samples be obtained? In
what pattern will the samples be chosen? In addition, after the data are obtained a decision
framework must be developed  to analyze the data.  Will no more than one exceedance in 10
samples be permitted or will no more than 10 exceedances in 100 be allowed?  That is,
what level of confidence is required to conclude that the site is clean? Answers to these
questions influence the spread of the distribution in Figure 2.1 in  D, E, F, and G and,
therefore, the steepness of the curve used for the Decision Rule in Figure 2.5, which is a
power curve similar to Figure  2.2.

             The following scenario describes the impact that the sampling plan and
decision rule can have on the actual degree of cleanup. A stringent chemical concentration
level is imposed as a requirement at a site (component 1). In contrast, five samples will be
obtained after remediation to verify attainment of the standard (component 2), and  80
percent confidence that the new  site mean is less than the standard will be required
(component 3).  The health effect results obtained by imposing a stringent numerical level
standard are weakened  because  the area has not been thoroughly sampled and the
associated level of confidence in the conclusions is relatively low.  In this case, a poor
sampling plan and low required level of confidence have influenced the actual degree of
cleanup in spite of the stringency of the numerical standard.
2.5          Missing or Unusable  Data, Detection Limits,  Outliers


2.5.1       Missing or Unusable  Data

             In any sampling program, physical samples will be obtained in the field and
then, some time during processing, a problem develops and a reliable measurement is not
available.  Samples can be lost, be labeled incorrectly, exceed holding times, be transcribed
                                      2-12

-------
 CHAPTER 2:  INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
Figure 2.5      Components of a Risk-Based Standard
                 maximum
                 acceptable
                 risk
                                                            estimate of
                                                           true process      *2. Sampling Plan
                                                                decision
                                                             (of whether the
                                                             site has attained
                                                            cleanup standard)
                                                       probability of saying
                                                       the site is "dirty" when
                                                       rt is really "clean"
                                                            probability of saying
                                                            the site is "dean" when
                                                            it is really "dirty"
1.  Concentration
   Threshold Level
                                                                               Decision Rule
                                       2-13

-------
  CHAPTER 2:  INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
incorrectly, or not satisfy quality control specifications. Clearly, missing data are not
available and cannot be used in data analysis. Data that do not satisfy the most rigorous
quality control specifications may or may not be usable; however, this depends on the
requirements as specified in the Quality Assurance Project Plan.

             One of the primary problems with missing data is the possibility that bias is
imposed on statistical estimates. For example, if the presence of high concentrations of a
specific contaminant  causes  laboratory-interferences that prevent samples  with the
contaminant from satisfying quality control specifications, then the data set will not
adequately reflect the presence of the contaminant  Careful attention should be paid to the
pattern of missing data to determine if the missing samples have a similar attribute such as
location, time, or chain of custody.  If so, then they may all have a special concentration
profile, and their absence may be affecting or biasing the result summary.

             However, the main question  is  how can  planning  help  to prevent the
problem of an excessive number of missing values.  One method can be used to help plan
for missing values.  The method can  be used if the approximate proportion of missing
values can be anticipated, based on prior experience with or a professional judgment of a
sampling team, laboratory, and data analyst.  The number of samples needed to conduct a
particular statistical evaluation is inflated  by the expected rate of missing values. More
sample results than needed will not be a problem because precision will increase; on the
other hand, too few sample results will  be a problem, and may result in  more  treatment
being required.

              The equation for the simplest situation requires prior estimation of the
sample size for the statistical procedures (nd). This is discussed above and throughout the
document.  Also, the rate at which missing or unusable values occur must be determined
(R). The final sample size required (nf) is then estimated using the simple  equation in Box
2.1.

             Throughout this guidance document, when sample size formulae, tables,
and graphs are used, the resulting sample sizes (nd and nnd) required  for a statistical
analysis having a specified precision can be increased using these equations in anticipation
of missing data.
                                      2-14

-------
   CHAPTER 2:  INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
                                     Box 2.1
                      Estimating the Final Sample Size Required

                                  nf=nd/(l-R)
               A similar equation is used for each of the h strata in  a stratified
     sampling plan:
                                nhf=nhd/d-Rh)
 2.5.2        Evaluation  of Less-Than-Detection-Limit Data

              The science and terminology associated with less-than-detection-limit
 chemistry are unstandardized.  There are a variety of opinions, methods, and approaches
 for reporting chemicals present at low concentration.  The problem can be segmented.
 First, there is the problem of how a chemist determines the detection limit value and
 EXACTLY what it means when values are reported above and below a detection limit.
 This question is not the subject of this document, but it is important. There is substantial
 literature on this subject and Bishop (1985) and Clayton £t al. (1986) offer useful insight
 and access to other references.

              The second problem is:  How should less-than-detection-limit values be
 evaluated along with Other values larger than the detection limit when both are present in a
 data set? This subject also is supported by a considerable amount of literature.  Examples
 include Gilbert and Kinnison  (1981);  Gilliom and Helsel (1986); Helsel and Gilliom
 (1986); and Gleit (1985).  This aspect of the detection limit problem is discussed briefly in
 the following paragraph.

              Fortunately, because of the null and alternative hypothesis  arrangement,
having concentrations  less then a detection limit is no problem when a proportion is being
tested, provided the detection limit is less than Cs. When the proportion or percentile is
being  tested, the important attribute of each data value is whether it is larger or smaller than
the Cs, rather than the magnitude of the value. In fact, a site can be evaluated easily relative
                                      2-15

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
to a high percentile in spite of a data set that includes many values less than the detection
limit, which is expected when  a cleanup technology has uniformly  reduced most
concentration measurements to less than the detection limit

             When the mean is being used as the basis of comparison with a cleanup
standard, the magnitude of each value is important.  When values are reported as being
less than a detection limit, it is generally recommended that they be included in the analysis
as values at the detection limit. This method accommodates detection limits that vary across
samples, and the method is simple to use.  In addition, this approach, although statistically
biased, errs in favor of health and environmental protection because of the construction of
the null and alternative hypothesis described earlier.  In some cases a less-than-detection-
limit value may be quite large relative to other measured values in a data set. In this case it
may be best to delete such a value.  Other methods are available for statistically addressing
less-than-detection-limit values as described above, but they may not be as conservative
with respect to environmental protection.


2.5.3       Outliers

             Measurements that are extremely large or small relative to the rest of the data
gathered and that are suspected of misrepresenting the true concentration at the sample
location are often called "outliers." If a particular observation is suspected to be in error,
the error should be identified and corrected, and the corrected value used in the analysis.  If
no such verification is  possible, a statistician should be consulted to provide modifications
to the statistical analysis that account for the suspected "outliers."  Methods to detect and
accommodate outliers are  described in Barnett and Lewis (1978) and Grubbs (1969).

             The  handling of outliers is a  controversial topic.  This document
recommends that  all  data  not  known to be  in error should be considered valid
because:

             •       The expected distribution of concentration values may be skewed
                     (i.e., nonsymmetric) so that large concentrations,  which look like
                     "outliers" to some analysts, may be legitimate;
                                       2-16

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
                    The procedures recommended in this document are less sensitive to
                    extremely low concentrations than to extremely high concentrations;
                    and
              •      High concentrations are of particular concern for their potential
                    health and environmental impact.

 2.6          General Assumptions

              The statistical procedures recommended in this guidance document must be
 applicable to many different  field situations; therefore, the procedures that  have been
 chosen are generally based on few assumptions. Situations  in which other statistical
 procedures might be used to provide more accurate or more cost-effective results will be
 noted with references.

              This document  assumes that (1) the  sources of contamination  and
 contaminating chemicals are known, (2) the sources of contamination have been removed,
 or there is no reason to believe that the concentrations of contaminant in the soil will
 increase after treatment, and (3)  chemical  concentrations  do not exhibit  short-term
 variability over the sampling  period. The methods presented can be used if sources of
 contamination exist or concentrations are expected to increase. However, sampling may
 have to be repeated and the  results carefully interpreted and presented to  reflect the
 possibility of additional contamination.

              When statistical tests are repeated to evaluate several chemicals,  such as
 testing that concentration levels for two chemicals both attain the cleanup standard, it is
 assumed that the sample  area will  be declared to attain the cleanup standard only if all
 statistical tests used are consistent with this conclusion. For other procedures that might be
 used to combine the results of individual  tests,  it would be advantageous to consult a
 statistician.
2.7          A Note on  Statistical  Versus  Field  Sampling Terminology

             The term sample is used in two different ways.  One refers to a physical
soil sample collected for laboratory analysis, and the other refers to a collection of data
called a statistical sample.  To avoid confusion, definitions of several terms follow.
                                      2-17

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
             Physical sample or soil sample: A portion of material (such as a soil
core, scoop, etc.) gathered at the waste site on which laboratory measurements are to be
made.  This may also be called a soil unit.

             Statistical sample:  A statistical sample consists of the collection of
multiple physical samples obtained for assessing attainment of the cleanup standard. The
units included in a statistical sample are selected by probabilistic means.

             Sample:   The word "sample" in this manual will generally have  the
meaning of "statistical sample."

             Sample size: The number of soil units being measured or the  size of the
statistical sample. Thus, a sample of size 10 consists of the measurements taken on 10 soil
units.

             Size of the physical sample:  This term refers to the volume or weight
of a soil unit or the quantity of soil in a single physical sample.

             The following terms refer to the manner in which the statistical sample of
physical  samples  is collected:  random  sample, systematic sample, stratified
sample, judgment sample. These sample designs are discussed in Chapter 4.
2.8          Summary

             Errors are possible in evaluating whether or not a site attains the cleanup
standard. For example, consider the errors associated with an extreme decision rule where
no exceedances of a standard are allowed. The site may be dirty even when substantial
sampling shows no exceedances; however, one sample may exceed the cleanup standard
and the site is judged dirty even when the site is acceptably clean.

             Statistical methods provide approaches for balancing these two decision
errors and allow extrapolation in a scientifically valid fashion. This chapter reviews the
                                      2-18

-------
  CHAPTER 2: INTRODUCTION TO STATISTICAL CONCEPTS AND DECISIONS
statistical concepts that are assumed and used as part of the procedures described in this

guidance document. These include:

                    A false positive decision-that the site is thought to be clean when it
                    is not;

              •      A false negative decision—that the site is thought to be contaminated
                    when it is not;

              •      The factors that specify the shape and location of the power curve
                    relative to the cleanup standard and to sample size determination;

              •      The mean--the value that corresponds to the "center" of the
                    concentration distribution;

              •      Proportions  or  percentiles—a value that can be used effectively,
                    based  on the  distribution  of contaminant concentration,  to
                    approximate the  maximum concentration of the  hazardous
                    contaminant.


              The components of a risk-based standard and how these components relate

to one another are reviewed and graphically illustrated. Methods to help plan for missing

or unusable data, less-than-detection-limit data, and outliers are discussed, followed by the

general assumptions associated with the statistical procedures explained in this document.

These assumptions are that:

              •      All of the sources of contamination and contaminating chemicals are
                    known;

              •      These sources have been  removed, so that the contamination will not
                    increase after treatment; and

              •      Chemical concentrations do not exhibit short-term variability over
                    the sampling period.
                                      2-19

-------

-------
  3.   SPECIFICATION  OF  ATTAINMENT OBJECTIVES


             The specification of attainment objectives must be completed by personnel
familiar with:
             •      The engineering aspects of the remediation;
             •      The nature and extent of contamination present;
             •      Health and environmental risks of the chemicals involved; and
             •      The costs of sampling, analysis, and cleanup.

             Attainment objectives are the procedures and criteria that must be defined to
guide waste site managers and personnel in the process of sampling and data analysis to
achieve a predetermined cleanup standard. Meeting these objectives and criteria enable the
waste site to be judged sufficiently remediated.

             As indicated in Figure 1.1, defining attainment objectives is the first task in
the evaluation of whether a site has attained a cleanup standard.  Figure 3.1 divides the box
devoted to the establishment and definition of cleanup objectives into its components.


3.1          Specification of  Sample Areas

             Three terms describing areas within the waste site are:
             •      Sample area;
             •      Strata; and
             •      Sample location.

             These terms are used in establishing the attainment objectives and the
sampling and analysis plans. Sample area specification is discussed below and methods
for defining strata and sample locations are discussed in Chapter 5.

             The waste site should be divided into sample areas. Each sample area will
be evaluated separately for attainment of a cleanup standard and will require a separate
statistical sample.
                                      3-1

-------
         CHAPTER 3:  SPECIFICATION OF ATTAINMENT OBJECTIVES
Figure 3.1    Steps in Defining the Attainment Objectives
r Start J
4
Define the sample areas.
(section 3.1)
1
Specify the sample handling
and collection procedures.
(section 3.2)
1
Specify the chemicals to be tested.
(section 3.3)
*
Establish the cleanup standard.
(section 3.4)
1
Specify the parameter to be compared
to the cleanup standard.
(section 3.5)
1
Specify the probability of mistakenly
declaring the sample area clean.
(section 3.6)
4
Review all elements of the
attainment objectives.

/ Are
>^ change
NV ream
                                                                  Yes
                                                                  No
                                                      (Specify sampling  j
                                                      I  and analysis plan.  I
                                      3-2

-------
          CHAPTER 3: SPECIFICATION OF ATTAINMENT OBJECTIVES
             Consider the following example, which emphasizes the importance of the

sample area definition. A site consists of an open field with little contamination and a

waste pile covering one-quarter of the site. If sampling and data analysis were executed

without respect to the waste pile, it might be maintained that the mean concentration of the

site was statistically lower than the standard.  The site wide mean was "excessively" low
because the waste pile data were "diluted" by many open field measurements. The solution

is to define the waste pile as one sample area and the open field as another.  Attainment

decisions will be made independently for each area.


             Because of the potential for this problem, it  is important to ensure that

sample areas  are clearly defined during the design phase.  Parties must agree that if the

sample area is judged clean, no more cleanup is required in any part of the sample area.

There are several considerations associated  with the definition of sample areas.


             1)     It is generally useful to define multiple sample areas within a waste
                    site. These areas should be defined so that they are as homogeneous
                    as possible with respect to prior waste  management activities. For
                    example, if a PCB  transformer disposal area and a lead battery
                    recycling area are  located on the same site, they should not  be
                    included in the same sample area.

             2)     It may also be useful to define sample  areas by batches of material
                    that will receive a treatment action, for example, dump truck loads
                    (see Exner et al., 1985)  or the minimum sized areas that can  be
                    stabilized or capped.

             3)     A site may be comprised of areas that require different sampling or
                    treatment technologies. For example, disturbed versus natural soils,
                    wetlands versus firm terrain, or sandy versus clay soils may suggest
                    establishment of different sample areas.

             4)     Finally, while more (smaller) sample  areas provide more flexible
                    response to changing conditions, sampling costs will increase with
                    the number of sample areas.

             5)     Sample area definitions also require that the depth or depth intervals
                    of interest be specified. This is discussed in greater detail in section
                    5.6.


             Figure 3.2  shows how  different geographic sample areas relate to one

another.
                                       3-3

-------
         CHAPTER 3: SPECIFICATION OF ATTAINMENT OBJECTIVES
Figure 3.2    Geographic Areas and Subareas Within the Site
         Site boundary
                           Map of the waste site

        Waste site with two sample areas, SA1 and SA2. Separate attainment
        decisions are made for each sample area. Sample area SA1 is divided into
        two strata, ST1 and ST2. (See Chapter 4 for more on stratified sampling
        [multiple strata per sample area].) Stratum ST1 has randomly selected soil
        sample locations indicated by "*".
3.2
Specification  of  Sample Collection and  Handling  Procedures
             Deciding whether a sample area attains the cleanup standard requires that

measurements be made on a statistical sample of soil units, and that these measurements be
compared to the cleanup standard. An important task for any decision procedure is to

define carefully what is being measured;  questions that must be answered include:


             •      What is meant by a soil unit or soil sample?
                                      3-4

-------
         CHAPTER 3:  SPECIFICATION OF ATTAINMENT OBJECTIVES
             •      How  is the  soil sample collected  and what equipment and
                    procedures are used?

             •      How  is  the soil sample  handled  between collection and
                    measurement?

             •      How are the laboratory measurements to be made and what accuracy
                    is to be achieved?


             The above questions are not addressed in this document.  Consult the

guidances listed in Table 1.1 for more information.
3.3         Specification of the Chemicals to be Tested


             For each sample area, the chemicals to be tested in each soil unit should be

listed. When multiple chemicals are tested, this document assumes that all chemicals must

attain the cleanup standard for the sample area to be declared clean.



3.4         Specification of the Cleanup Standard


             Concentration measurements for each physical sample will be compared to

the appropriate, relevant, or applicable cleanup standard chosen for each chemical to be

tested.  Cleanup standards are determined by EPA during the site-specific endangerment

assessments.  The cleanup standard for each chemical of concern must be stated at the

outset of the remedial verification investigation.  Final selection of the cleanup standard

depends  on many factors as discussed in  USEPA (1986c).  Selection of the cleanup
standard depends on the following factors:


             •      The availability and value of other appropriate criteria;

             •      Factors related to toxicology and exposure, for example, the effect
                    of multiple  contaminants, potential use of the waste site  and
                    pathways of exposure, population sensitivities to the chemical;

             •      Factors related to uncertainty, for example, the effectiveness of
                    treatment alternatives, reliability of exposure data, and the reliability
                    of institutional controls; and

             •      Factors related to  technical limitations, for example,  laboratory
                    detection limits,  background contamination levels,  and technical
                    limitations to restoration.
                                       3-5

-------
         CHAPTER 3: SPECIFICATION OF ATTAINMENT OBJECTIVES
             Throughout this document, the cleanup standard will be denoted by Cs.
3.5          Selection of  the Statistical  Parameter  to Compare  with  the
             Cleanup Standard
3.5.1       Selection Criteria for the Mean, Median, and Upper Percentile

             Criteria for selecting the parameter to use in the statistical assessment
decision are:

             •      The criteria used to develop the risk-based standards, if known;
             •      The lexicological effect of the contaminant being measured (e.g.,
                    carcinogenic, systemic toxicant, developmental toxicant).
             •      The relative sample sizes required or the relative ease of calculation;
             •      The likelihood of concentration measurements below the cleanup
                    standard; and
             •      The relative spread of the data.

             Table 3.1 presents these criteria and when they support or contradict the use
of the mean,  upper percentile, and median.  The  median may offer  a reasonable
compromise because the median is the 50th percentile and a measure of central tendency.
Table 3.2 illustrates the broad potential utility of the median.
                                      3-6

-------
          CHAPTER 3: SPECIFICATION OF ATTAINMENT OBJECTIVES
Table 3.1      Points to consider when trying to choose among the mean, high percentile,
              or median
Parameter
                    Points to Consider
Mean
1) Easy to calculate and estimate a confidence interval.

2) Requires fewer samples than other parameters to achieve similar
confidence.

3) Useful when the cleanup standard has been based on
consideration of carcinogenic or chronic health effects or long-term
average exposure.

4) Useful when the soil is uniform with little spread in the sample
data.

5) Not as useful when contamination exists in small areas within a
larger area that is being sampled because the mean can be "diluted"
or reduced by the inclusion of clean  areas in the sample area.

6) Not very representative of highly variable soils because the most
heavily contaminated areas are not characterized by a mean.

7) Not useful when there are a large proportion of less-than-
detection-limit values.
Upper
Proportion/
Percentile
1) Can be expressed in terms that have more meaning than tests of
the mean. Volumes or areas can be expressed relative to the total
volume or area of concern, and this can be a proportion of importance.
For example, if no more than 10,000 m^ in a total volume of
1,000,000 m3 can exceed a cleanup standard, then this becomes a
test to verify with reasonable confidence that no less than 99 percent
of the site is below the cleanup standard.

2) Will provide the best control of extreme values when data are
highly variable.

3) Some methods are unaffected by less-than-detection-limit values,
as long as the detection limit is less than the cleanup standard.

4) If the health effects of the contaminant are acute or worst-case
effects, extreme concentrations are of concern and are best evaluated
by ensuring that a large proportion of the site is below a cleanup
standard.
                                       3-7

-------
          CHAPTER 3: SPECIFICATION OF ATTAINMENT OBJECTIVES
Table 3.1      Points to consider when trying to choose among the mean, high percentile,
              or median (continued)
Parameter
                    Points to Consider
Upper
Proportion/
Percentile
(continued)
5) Similar to the mean, if contamination exists within a small area,
but if the sampling program is conducted to include a much larger
surrounding area with little contamination, the proportion will be
affected or "diluted."

6) The proportion  of the site that must be below the cleanup standard
must be chosen.

7) When statistical methods are used that require few assumptions,
a larger sample size will be required than for tests based on the
mean.
Median
1) Has benefits over the mean because it is not as heavily
influenced by outliers and highly variable data, and can be used with
a large number of less-than-detection-limit values.

2) Has many of the positive features of the mean, in particular its
usefulness for evaluating cleanup standards based on carcinogenic
or chronic health effects and long-term average exposure.

3) Has positive features of the proportion, including its reliance on
fewer assumptions.

4) Retains some negative features of the mean in that the median
will not control extreme values.
                                        3-8

-------
          CHAPTER 3:  SPECIFICATION OF ATTAINMENT OBJECTIVES
Table 3.2      Recommended parameters to test when comparing the cleanup standard to
              the average concentration of a chemical with chronic effects
Data Variability
Large Coefficient
of Variation
(Perhaps cv > .5)
Small Coefficient
of Variation
(Perhaps cv < .5)
Proportion of the data with concentrations
below the detection limit:
Low High
(Perhaps < 50%) (Perhaps > 50%)
Mean
(or Median)
Mean
(or Median)
Upper Percentile
Median
3.5.2
Multiple Attainment Criteria
             This guidance document addresses testing for a single parameter--the mean
or a specified percentile of the distribution—that is below the cleanup standard. However,
in some situations two or more parameters can be chosen.  The sample area would be
declared clean if all parameters were significantly less than the cleanup standard.  For
example, there may be interest in providing protection against excessive extreme  and
average concentrations. Therefore, the mean and an upper percentile can be tested using
the rule that the sample area attains the cleanup standard if both parameters are below the
cleanup standard.  When testing both parameters, the number of samples collected will be
either the number required for the test of the mean or the number required for the test of the
percentile (whichever number is larger).

             Other more complicated criteria may be used to assess the attainment of the
cleanup criteria. Multiple criteria are established in the following examples. In each case it
is desirable that:

             •      Most of the soil has concentrations below the cleanup standard and
                    that the concentrations above the cleanup standard are not too large.
                                      3-9

-------
          CHAPTER 3:  SPECIFICATION OF ATTAINMENT OBJECTIVES
                    This may be accomplished by testing whether the 75th percentile is
                    below  the cleanup standard and whether the  mean  of those
                    concentrations above the cleanup standard is less than twice the
                    cleanup standard.  This combination of tests can be performed with
                    minor modifications to the methods presented in this document.

                    The mean concentration be less than the cleanup standard and that
                    the standard deviation of the data be small, thus limiting the number
                    of extreme concentrations. This may be accomplished by testing if
                    the mean is below the cleanup  standard and the coefficient of
                    variation is below some low level (.5 for example). This document
                    does not  address testing the standard deviation, variance, or
                    coefficient of variation against a cleanup standard.

                    The mean concentration be less than the cleanup standard and that
                    the remaining contamination  be  uniformly  distributed across the
                    sample area relative to the overall spread of the data. Testing these
                    criteria may be accomplished by testing for a mean below the
                    cleanup standard  and  variability between strata means that is not
                    large compared to the variability within strata (analysis of variance).

                    The mean concentration be less than the cleanup standard and that no
                    area of contaminated soil (assumed to be circular) be larger than a
                    specified size. Testing these criteria involves testing for hot spots,
                    which  are discussed in Chapter 9 and more extensively in Gilbert
                    (1987).
3.6          Decision Making With Uncertainty: The Chance of Concluding

              the Site Is  Protective of Public Health and  the Environment

              When  It Is Actually  Not Protective


              As discussed in Chapter 2, the validity of the decision that a site meets the

cleanup standard depends on how well the  samples of soil represent the site,  how

accurately the soil samples are analyzed, and  other factors, all of which are subject to

variation.   Different sampling patterns will yield different results and  repeated

measurements on individual soil samples will yield different concentrations. This variation

introduces uncertainty into the decision concerning the attainment of a cleanup standard.


              As a result of this uncertainty, one may decide that the site is clean when it

is not.  In the context  of this document, this mistaken conclusion can be referred to as a

false positive finding (the chance  or probability of a false positive is indicated by the
Greek letter alpha, a).  There are two important points surrounding false positives:
                                      3-10

-------
          CHAPTER 3: SPECIFICATION OF ATTAINMENT OBJECTIVES
              •      First, from an environmental and health protection perspective, it is
                    imperative to reduce the chance of a false positive. In direct terms a
                    false positive is the chance of deciding a Superfund site is clean
                    when it still poses a health or environmental threat.  Of course, a
                    low false positive rate does not come without a cost. The additional
                    cost required to lower the false positive rate comes from additional
                    samples and more accurate sampling and analysis methods.

              •      Second, the definition of a false positive in this document is exactly
                    opposite  the more familiar definition of a false positive under
                    RCRA detection and compliance monitoring. This is because the
                    null and alternative hypotheses are reversed, once a site has been
                    verified to  have contamination.  Under the RCRA detection
                    monitoring  situation,  EPA was concerned about a  high  false
                    negative rate; here EPA is concerned about a high false positive rate.


              In order to design a statistical test for deciding whether the sample area

attains the cleanup standard, those individuals specifying the sampling and analysis

objectives should select and specify  the false positive rate for  testing the site. While

different false positive rates can be used for each chemical, it is recommended that all

chemicals in the sample area use the same rates.  This rate is the maximum probability that

the sample area will be declared clean by mistake when it is actually dirty. For a further

discussion of false positive rates, see Sokal and Rohlf (1981).
3.7          Data Quality Objectives


             The Quality Assurance Management staff within EPA  has developed
requirements and procedures for establishing Data Quality Objectives (DQOs) when
environmental data are collected to support regulatory and programmatic decisions.  The
DQOs are a clear set of statements addressing the following issues (see USEPA, 1987a and
USEPA, 1987b).

             •      The decision to be made;

                    The reasons environmental data are needed and how  they will be
                    used;

             •      Time and resource constraints on data collection;

             •      Detailed description of the data to be collected;
                                      3-11

-------
         CHAPTER 3:  SPECIFICATION OF ATTAINMENT OBJECTIVES
                    Specifications regarding the domain of the decision;

                    The consequences of an incorrect decision attributable to inadequate
                    environmental data;

                    The calculations, statistical or otherwise, that will be performed on
                    the data in order to arrive at the result, including the statistic that will
                    be used to summarize the data and the "action level" (cleanup
                    standard) to which the summary statistic will be compared; and

                    The level of uncertainty that the decision maker is willing to accept
                    in the results derived from the environmental data.
             The specification of attainment objectives that have been discussed in this

chapter and the sampling and analysis plan discussed in the next chapter are an important

part of the Data Quality Objectives process. Completion of the DQO process will provide
the required information for the specification of attainment objectives.



3.8          Summary


             The following steps must be taken to evaluate whether a site has attained the

cleanup standard:

             •      Define the attainment objectives;

             •      Specify sample design and analysis plan, and determine sample size;

             •      Collect the data; and

             •      Determine if the sample area attains the cleanup standard.

             This chapter discusses attainment objective specifications. Attainment
objectives are specified by RPMs, RPs, and their contractors. They are not statistically

based decisions.


              •      Define the sample area.  The waste site should be divided  into
                    sample areas.  Each sample area will be evaluated separately for
                    attainment of a cleanup standard and will require a separate statistical
                    sample.   It is important to ensure  that sample areas are clearly
                    defined during the design phase.

              •      Specify  the  sample handling and  collection procedures.
                    An important task for any decision procedure is  to define carefully
                    what is being measured.
                                       3-12

-------
CHAPTER 3: SPECIFICATION OF ATTAINMENT OBJECTIVES
          Specify the chemicals to be tested.  Chemicals to be tested in
          each soil unit should be listed.

          Establish  the  cleanup  standard.   Cleanup standards are
          determined by EPA using site-specific risk assessments or ARARs.
          The cleanup standard for each chemical of concern must be stated at
          the outset of the remedial verification investigation.

          Specify  the parameter  to be compared  to  the  cleanup
          standard. In other words: "Does the cleanup standard represent an
          average condition (mean) or a level to be rarely exceeded (high
          percentile)?  Criteria for selecting the parameter to use in the
          statistical assessment decision are:

                 The criteria used to develop the risk-based standards, if
                 known;

                 Whether the contaminant being measured has an acute or
                 long-term chronic effect;

                 The relative sample sizes required or the relative ease of
                 calculation;

                 The likelihood of concentration measurements below the
                 detection limit; and

                 The relative spread of the data.

          Specify  the  probability  of  mistakenly  declaring  the
          sample area clean.  Select and specify the false positive rate for
          testing the site.  It is recommended that all chemicals in the sample
          area use the same rates. This rate is the maximum probability that
          the sample area  will be declared clean by mistake when it is actually
          dirty.

          Review  all elements of  the attainment objectives.
                            3-13

-------

-------
    4.   DESIGN OF  THE SAMPLING AND ANALYSIS
                                   PLAN
             Once the attainment objectives are specified by program and subject matter
personnel, statisticians can be useful for designing important components of sampling and
analysis plans.

             The methods of analysis must be consistent with the sample design and the
attainment objectives. For example, data that are collected using stratified sampling cannot
be analyzed using the equations for simple random sampling.  The sample design and
analysis plan must coincide. If there appears to be any reason  to use different sample
designs or analysis plans than those discussed in this manual, or if there is any reason to
change either the sample design or the analysis plan after field data collection has started, it
is recommended that a statistician be consulted.

             This chapter presents some approaches to the design of a sampling and
analysis plan and presents the strengths and weaknesses of various designs.
4.1          The Sampling Plan

             The following sections provide background discussion guiding the choice
of sampling plan for each sampling area.  Chapter 5 discusses the details of how to
implement a sampling plan. For more details, see Kish (1965), Cochran (1977), Hansen si
al (1953), or the EPA guidances in Table 1.1.

             The sample designs considered in this document are:

             •      Simple random  sampling called random  sampling in this
                    document;
             •      Stratified random sampling called stratified sampling in this
                    document;
             •      Simple systematic sampling called systematic sampling in this
                    document; and
                                     4-1

-------
        CHAPTER 4:  DESIGN OF THE SAMPLING AND ANALYSIS PLAN
             •      Sequential random sampling called sequential sampling in this
                    document.

             Randomization is necessary to make probability or confidence statements
about the results of the sampling. Both random and random start systematic sample
locations have random components.  In contrast, sample selection using the judgment of
the sampler has no randomization.  Results from such samples cannot be generalized to the
whole sample area and no probability statements can be made when judgment sampling is
used.  Judgment sampling may be justified, for example, during the preliminary
assessment and site investigation stages if the sampler has substantial knowledge of the
sources and history of contamination.  However, judgment samples should not be used to
determine whether the cleanup standard has been attained.

             Combinations of the designs referred to above can also  be  used.  For
example, systematic sampling could be used with stratified sampling. In the situation
where cleanup has occurred, if the concentrations across the site are relatively low and
uniform and the site is accessible, the sample designs considered in this document should
be adequate. If other more complicated sample designs are necessary, it is recommended
that a statistician be consulted on the best design, and on the appropriate analysis method
for that design. Figure 4.1 illustrates a random, systematic, and stratified sample.
4.1.1       Random  Versus Systematic Sampling

             Random selection of sample points requires that each sample point be
selected independent of the location of all other sample points. Figure 4.1 shows a random
sample. Note that under random sampling no pattern is expected in the distribution of the
points.  However, it is possible (purely by chance) that all of the sample points will be
clustered in, say, one or two quadrants of the site. This possibility is extremely small for
larger sample sizes.
                                      4-2

-------
       CHAPTER 4: DESIGN OF THE SAMPLING AND ANALYSIS PLAN
Figure 4.1    Illustration of Random, Systematic, and Stratified Sampling (axes are
            distance in meters)
                              Random Sampling
                  0    25   50   75   100  125  150  175
               0
               o-
               o-
                              Systematic Sampling
                      25   50   75   100  125  150  175

                           Stratified Random Sampling
                 0    25  50    75   100  125  150  175
                   Legend:
                         •Sample Area Boundary
                         "Strata Boundary
                         Randomly Selected Sample Location
                         Sample Location Determined
                         Systematically
                                   4-3

-------
        CHAPTER 4: DESIGN OF THE SAMPLING AND ANALYSIS PLAN
             An alternative to random sampling is systematic sampling, which distributes
the sample more uniformly over the site. Because the sample points follow a simple pattern
and are separated by a fixed distance, locating the sample points in the field may be easier
using a systematic sample than using a random sample.  In many circumstances, estimates
from systematic sampling may be preferred.  More discussion of systematic versus random
sampling can be found in Finney (1948), Legg, §1 §L (1985), Cochran (1977), Osborne
(1942), Palley and Horwitz (1961), Peshkova (1970), and Wolter (1984).
4.1.2       Simple Versus Stratified Sampling

             The precision of statistical estimates may be improved by dividing a sample
area into more  homogeneous strata.  In this way, the variability due  to soil, location,
characteristics of the terrain, etc. can be controlled, thereby improving the precision of
contamination level estimates.  Homogeneous areas from which separate samples are
drawn are referred to as "strata," and the combined sample from all areas is referred to as a
"stratified sample."

             Like systematic sampling, stratification provides another way of minimizing
the possibility that important areas of the site will not be represented in the sample. Note in
Figure 4.1 that  the two strata represent subareas for which representation in the sample
will be guaranteed under a stratified sampling design.

             The main advantage of stratification is that it can result in a more efficient
allocation of resources than would be possible with a simple random sample.  For example,
suppose that, based on physical features, the site can be divided into a hilly and a flat area,
and that the hilly area comprises about 75 percent of the total area and is more expensive to
sample than the flat area.  If there is no reason to analyze the two subareas separately,  we
might consider selecting a simple random sample of soil units across  the entire site.
However, with a simple random sample, about 75 percent of the sample would be in  the
hilly, and therefore more expensive, areas of the site. With stratified sampling, the sample
can be allocated disproportionately to the two subareas, i.e., sample fewer units from hilly
areas and more from flat areas.  In this way, the resulting cost savings (over a simple
random sample) can be used to increase the total sample size and, hence, the precision of
estimates from the sample.
                                       4-4

-------
        CHAPTER 4: DESIGN OF THE SAMPLING AND ANALYSIS PLAN
              The above illustration is highly simplified.  In addition to differential
stratum costs, factors such as the relative sizes of the strata and the variability of the
contaminant under study in the different strata will affect the optimum allocation.  The
illustration  does, however, point out that stratification can be used to design a more
efficient sample, and is more than simply a device to ensure that particular subareas of the
site are represented in the sample. A formal discussion of stratified sampling, and the cost
and variance considerations used to determine an optimum allocation, is beyond the scope
of this manual. However, sections 5.4 and 6.4 offer a discussion of the basic principles
used to guide the design of a stratified sample.

              Although stratified sampling is more difficult to implement in the field and
slightly more difficult to analyze, stratified sampling will provide benefits if differences in
mean concentrations or sampling costs across the sample area exist and can be reasonably
identified using available data. It is important to define strata so that the physical samples
within a stratum are more similar to each other than to samples from different strata.
Factors that can be used to define strata are:

              •     Sampling depth (see section 5.6 for details);
              •     Concentration level;
              •     Physiography/topography;
              •     The presence  of other contaminants that affect the analytical
                    techniques required at the lab;
              •     The history and sources of contamination over the site;
              •     Previous cleanup attempts; or
              •     Weathering and run-off processes.

              There are two fundamental and important points to remember when defining
areas that will become different strata:

              •      The strata must not overlap-no area within one strata can be within
                    another strata; and
              •      The sum of the sizes of the strata must equal the area of the sample
                    area.
                                       4-5

-------
        CHAPTER 4: DESIGN OF THE SAMPLING AND ANALYSIS PLAN
             In other words, the strata must collectively account for the entire sample
area of interest—no more, no less.
4.1.3       Sequential Sampling

             For most statistical methods, the analysis is performed after the entire
sample has been collected and the laboratory results are complete. In sequential random
sampling, the samples are analyzed as they are collected. A statistical analysis of the data,
after each sample is collected and analyzed, is used to determine if another sample is to be
collected or if the sampling program ten >inates with a decision that the site is clean or dirty.
(Sequential sampling is the subject of Chapter 8.)
4.2          The  Analysis Plan

              Similar to sampling plan designs, planning an approach to analysis and the
actual analysis begin before the first sample is collected. The first task of the analysis plan
is to determine how the cleanup standard should function. In other words, what is the
cleanup standard: a value that should be rarely exceeded; an average value; or a level that
defines the presence of a  hot spot? This must be decided because it determines whai
analysis method will be used to determine attainment.

              Second, the analysis plan must  be developed in conjunction with the
sampling plan discussed earlier in this chapter.  For example, plans to conduct stratified
sampling cannot be analyzed using the equations for random sampling.

              Third, the first actual  step required in the  analysis plan should be a
determination of the appropriate sample size. This requires calculations  and evaluation
before the data are collected. Often the number of samples is determined by economics and
budget rather than an evaluation of the required accuracy. Nevertheless, it is important to
evaluate the accuracy associated with a prespecified number of samples.
                                       4-6

-------
        CHAPTER 4: DESIGN OF THE SAMPLING AND ANALYSIS PLAN
             Fourth, the analysis plan will describe the evaluation of the resulting data.
Chapters 6 through 10 offer various analytical approaches, depending on attainment
objectives and the sampling program.  Table 4.1 presents where in this document various
combinations of analysis and sampling plans are discussed.
Table 4.1     Where sample designs and analysis methods for soil sampling are discussed
             in this document
Type of
Evaluation
Test of the Mean
Test of
Percentiles
Hot Spot
Evaluation
Geostatistics
Analysis
Method
Test for means
Nonparametric
Tolerance Intervals
Sequential Sampling

Indicator Kriging
Chapter Location
Sample Design
Random
6.3.3
7.3.3
7.3.6


Stratified
6.4.2
7.5.2


Systematic
6.5.2
7.6
9.2.1
10.3
Sequential

8.2


4.3
Summary
             Design of the sampling and analysis plan requires specification of attainment
objectives by program and subject matter personnel.  The sampling and analysis objectives
can be refined with the assistance of statistical expertise. The sample design and analysis
plans go together, therefore, the following methods of analysis must be consistent with the
sample design:

             •      Random sampling;
             •      Stratified sampling;
                                      4-7

-------
        CHAPTER 4: DESIGN OF THE SAMPLING AND ANALYSIS PLAN


             •      Systematic sampling; and
             •      Sequential sampling.

Random selection of sample points requires that each sample point be selected independent
of the location of all other sample points. An alternative to random sampling is systematic
sampling, which distributes the sample more uniformly over the site. Systematic sampling
is preferred in hot spot searches and in geostatistical studies.

             Like systematic sampling, stratified sampling minimizes the possibility that
important areas  of the  site will not be represented  by dividing a sample area into
homogeneous subareas. The main advantage of stratification is that it can result in a more
efficient allocation of resources than would be possible with a random sample.

             Sequential sampling (Chapter 8) requires that the samples be  analyzed as
they are collected.

             Decisions required to plan an approach to analysis are:

             •      Determine the analysis method that is most useful;
             •      Develop the plan in conjunction with the sampling plan;
             •      Determine the appropriate sample size; and
             •      Describe how the resulting data will be evaluated.
                                       4-8

-------
              5.  FIELD  SAMPLING PROCEDURES
              The procedures discussed in this chapter ensure that:

              •      The method of establishing soil sample locations in the field is
                    consistent with the planned sample design;
              •      Each sample location is selected in a nonjudgmental and unbiased
                    way; and
              •      Complete documentation of all sampling steps is maintained.

              The procedures discussed in this chapter assume that the sampling plan has
 been selected; the boundaries of the waste site, the sample areas, and any strata have been
 defined; a detailed map of the waste site is available; and the sample size is known. Sample
 size determination is discussed in Chapters 6, 7, 8, and 9. Also, if sequential sampling or
 hot spot searches are planned, the reader should refer to Chapters 8 and 9, respectively, for
 additional guidance on field sampling.
5.1          Determining the General Sampling  Location

              Locating the soil samples is accomplished using a detailed map of the waste
site with a coordinate system to identify sampling locations. Recording and automation of
station-specific data should retain coordinate information, especially  if geostatistical
manipulations are performed (see Chapter 10) or a geographic information system will be
used.

              Soil sample locations will be identified by X and Y coordinates within the
grid system. It is not necessary to draw a grid for the entire waste site; it is only necessary
to identify the actual coordinates selected.  Figure 5.1 is an example of a map with a
coordinate system.  In this example, the  origin of the coordinate system is at the lower
lefthand corner of the map; however, this may not be true for coordinate systems based on
measurements from a reference  point on the ground,  i.e., a benchmark or  a standard
coordinate system such as latitude and longitude.
                                       5-1

-------
                 CHAPTER 5: FIELD SAMPLING PROCEDURES
Figure 5.1    Map of a Sample Area with a Coordinate System

          100
           75
           50
                                                                          Boundary of
                                                                         the sample area
0
25    50
75
                                           100    125    150    175
                                      X Coordinate
              The boundaries of the sample areas (areas within the site for which separate
attainment decisions are to be made) and strata within the sample areas (if stratified
sampling is required) should be shown on the map.  The map should also include other
important features that will be useful in identifying sample locations in the field.

              Accurate location of sampling points can be expensive and time consuming.
Therefore, a method is suggested which uses the coordinate system to identify the general
area within which  the soil sample is to be collected, followed by a second stage of
sampling, described in section 5.5,  to identify the sample point accurately.

              The X and Y coordinates of each sample location must be specified.  This
distance between coordinates on each axis represents a reasonable accuracy for measuring
distance in the field, and is represented by M.  If distances can be measured easily to within
2 m, but not to within 1 meter, the  coordinates should be provided to the nearest 2 m (M =
2 m). The sampling coordinates can be identified with greater accuracy when the distances
to be measured between reference  points are short, the measuring equipment is accurate or
easy to use, or there are few obstructions to line-of-sight measuring such as hills, trees, or
bushy vegetation.  For example, the location within a small lagoon, say, 30 by 30 m, can
                                        5-2

-------
                 CHAPTER 5: FIELD SAMPLING PROCEDURES
be established to within 5 cm.  On the other hand, in a 10 hectare field it may only be
reasonable to identify a location to within 10 m.
5.2          Selecting the Sample Coordinates for a Simple Random Sample

             A random sample of soil units within the sample area or stratum will be
selected by generating a series of random (X,Y) coordinates, finding the. location in the
field associated with these (X,Y) coordinates, and following the field procedures described
in section 5.5 for collecting soil samples. If the waste site contains multiple sample areas
and/or strata, the same procedure described above is used to generate random pairs of
coordinates with the appropriate range until the specified sample size for the particular
portion of the site has been met. In other words, a separate  simple random sample of
locations should be drawn for each sample area or stratum. To simplify the discussion, the
procedures below discuss selection of a random sample in a sample area.

             The number of soil samples to be collected must be specified for each
sample area. In what follows, the term nf will be used to denote the number of samples to
be collected in the sample area.

             To generate the nf random coordinates (Xj,Yj), i = 1 to nf, for the sample
area, determine the range of X and Y coordinates that will completely cover the sample
area.  These coordinate ranges will define a rectangle that circumscribes the sample area.
Let the coordinate ranges be Xmin to Xmax and Ymin to Ymax.  Thus, the point (Xmin,
Ymin) represents the lower lefthand corner of the rectangle, and (Xmax, Ymax) represents
the upper righthand corner of the rectangle.  The nf sample coordinates (Xj.Yj) can be
generated using a random number generator and the steps described in Box 5.1. Box 5.2
gives an example of generating random sample locations.
                                      5-3

-------
   CHAPTER 5:  FIELD SAMPLING PROCEDURES
                      Box 5.1
     Steps for Generating Random Coordinates That
              Define Sampling Locations

1)   Generate a set of coordinates (X,Y) using the following
     equations:

           X = Xmin + (Xmax - Xmin)*RND     (5.1)

           Y = Ymin + (Ymax - Ymin)*RND     (5.2)

     RND is the next unused random number between 0 and 1 in a
     sequence of random numbers.  Random numbers can be
     obtained from calculators, computer software, or tables of
     random numbers.

2)   If (X,Y) is outside the sample area, return to step 1 to generate
     another random coordinate; otherwise go to step 3.

3)   Define (Xj,Yj) using the following steps:

     Round X to the nearest unit that can be located easily in the field
     (see section 5.1);  set this equal to Xj

     Round Y to the nearest unit that can be located easily in the field
     (see section 5.1); set this equal to Yj.

4)   Continue to generate the next random coordinate, (Xj+i,Yj+i).
                         5-4

-------
                 CHAPTER 5: FIELD SAMPLING PROCEDURES
                                    Box 5.2
               An Example of Generating Random Sampling Locations

              To  illustrate  the selection of simple random sample of locations,
    assume that seven soil units will be selected from the site in Figure 5.2.  Pairs of
    random numbers (one X coordinate and Y coordinate for each pair) identify each
    sample point.  X  will be measured on the map's coordinate system in the
    horizontal direction and Y in the vertical direction.  It is assumed for this example
    that selected coordinates can be identified to the nearest meter. The first number of
    pair, Xj[,  must be between 0 and 190 (i.e., Xmjn = 0 and Xmax = 190) and the
    second, Yj, between 0 and 100 (Ymjn = 0 and Ymax = 100) for this example. If
    the X and Y coordinates for any pair identify a location outside the area of interest,
    they are ignored and the process is continued until the sample size nf has been
    achieved.

XYpair
1
2
3
4
5
6
7
8
9
Random
X coordinate
67
97
190
17
94
123
25
35
152
Random
Y coordinate
80
4
88 (outside of sample
15 (outside of sample
76
49
52
39
14




area)
area)





              It took nine attempts to secure seven coordinates that fall within the
    sample area. The randomly selected coordinates for pairs 3 and 4 fall outside the
    waste site and are to be discarded. The remaining seven locations are randomly
    distributed throughout the site.

        These locations can now be plotted on the map, as shown in Figure 5.2.
5.3
Selecting the Sample Coordinates for  a Systematic  Sample
             A square grid and a triangular grid are two common patterns used in

systematic or grid sampling.  These patterns are shown in Figure 5.3. Note that the rows

of points in the triangular grid are closer (.866L) than the distance between points in a row

(L) and that the points in every other row are offset by half a grid width.
                                      5-5

-------
                CHAPTER 5: FIELD SAMPLING PROCEDURES
Figure 5.2    Map of a Sample Area Showing Random Sampling Locations

          Locations of the random samples are indicated by a •. The numbers
          reference the XY pairs in Box 5.2.
              100
                   0     25     50    75     100   125    150    175
                0
                                     X Coordinate


Figure 5.3    Examples of a Square and a Triangular Grid for Systematic Sampling

             Square Grid                   Triangular Grid
f 	 1
1 J
• 	 '
» 4
k . A 4h
' T
L
, . .1
* .*. .* *
* 	 /\ 	 /\ 	 * * T
\ .- \ -866L
• ¥ • • -L
t -r t
              h-L— 4
             The size of the sample area must be determined in order to calculate the
distance, L, between the sampling locations in the systematic grid.  The area can be
measured on a map using a planimeter. The units of the area measurement (such as square
feet, hectares, square meters) should be recorded.

             Denote the surface area of the sample area by A.  Use the equations in Box
5.3 to calculate the spacing between adjacent sampling locations.
                                     5-6

-------
                 CHAPTER 5: FIELD SAMPLING PROCEDURES
                                    Box 5.3
              Calculating Spacing Between Adjacent Sampling Locations
                          for the Square Grid in Figure 5.3

                                                            (5.3)

                        for the Triangular Grid in Figure 5.3

                                                            (5.4)
The distance between adjacent points, L, should be rounded to the nearest unit that can be
easily measured in the field.

             After computing L, the actual location of one point in the grid should be
chosen by a random procedure.  First, select a random coordinate (X,Y) following the
procedure in Box 5.1.  Using this location as one intersection of two gridlines, construct
gridlines running parallel to the coordinate axes and separated by a distance L.  The
sampling locations are the points at the intersections of the gridlines that are within the
sample area boundaries.  Figure 5.4 illustrates this procedure.  Using this procedure, the
grid will always be oriented parallel to the coordinate axes. The grid intersections that lie
outside  the sample area are ignored.  There will  be some variation in sample  size,
depending on the location of the initial randomly drawn point.  However, the relative
variation in number of sample points becomes small as the number of desired sample points
increases.  For unusually shaped sample areas (or strata), the number of sample points can
vary considerably from the desired number.

             The coordinates for the sample points will be all coordinates (Xj.Yj) such
that

             •      (X,,Yj) is inside the sample area or stratum;
             •      Xj = X + j*L, for some positive or negative integer j, and;
             •      Yj = Y + k*L, for some positive or negative integer k.
                                      5-7

-------
                 CHAPTER 5: FIELD SAMPLING PROCEDURES
Figure 5.4    Locating a Square Grid Systematic Sample
(1)  S elect initial random point.
   100
(2)   Construct coordinate axis going
     through initial point.
                                        100

                                               25  50   75   100  125  ISO  175
                                                          X
(3)   Construct lines parallel to
     vertical axis, separated by
     a distance of L.
(4)   Construct lines parallel to
     horizontal axis, separated by
     a distance  of L.
          25  50   75   100  125  150  175
                                              25  SO   75  100  125   150   175
                                        5-8

-------
                 CHAPTER 5: FIELD SAMPLING PROCEDURES
             Box 5.4 and Figure 5.5 give an example of locating systematic coordinates
and the resultant sampling locations plotted on a map of the site.
                                   Box 5.4
                         Locating Systematic Coordinates

              Using the map in Figure 5.1 and a planimeter, the area of the sample
    area is determined to be 14,025 sq. m.  If the sample size is 12, the spacing
    between adjacent points is:
    [JT   1
= A/ — = M
                           14025   ,,,        ...
                                 = 34 m, rounded to the nearest meter
              Using the procedure in Box 5.1, a random coordinate (X,Y) = (11,60)
    is generated.  Starting from this point, the following sampling points can be
    calculated:

                                 (79,94)    (113,94)  (147,94)
              (11,60)   (45,60)   (79,60)    (113,60)  (147,60)  (181,60)
                       (45,26)   (79,26)    (113,26)  (147,26)  (181,26)

              These points are shown in Figure 5.5.  The intended sample size was
    12; however, because of the random selection process and the irregularity of the
    sample area boundary,  there are 14 sample points within the sample area.  A
    sample will be collected at all 14 locations.
                                      5-9

-------
                CHAPTER 5:  FIELD SAMPLING PROCEDURES
Figure 5.5    Map of a Sample Site Showing Systematic Sampling Locations
                0
                   0
                                        I
25    50     75     100   125    150    175
             X Coordinate
     Initial Randomly Selected Sample Point
5.3.1        An  Alternative Method for Locating the Random Start Position
             for  a Systematic Sample

             An alternative method may be used to locate the random start position for a
systematic triangular grid sample  (J. Barich, Pers. Com.,  1988).  This approach, as
detailed in Box 5.5, determines a random start location by choosing a random angle A and
a random distance  Y from point X.  This approach is useful under circumstances where a
transit and stadia rod are available for turning angles, measuring distances, and establishing
transects. This method is essentially equivalent to the method described above.
                                     5-10

-------
             CHAPTER 5: FIELD SAMPLING PROCEDURES
                                Box 5.5
                Alternative Method for Locating the Random
                   Start Position for a Systematic Sample

          Figure 5.6 and the following steps explain how  to implement the
sequence.

          1)  Establish the main transect with endpoints X  and X' using any
convenient reference line (e.g., established boundary). Notice that the transect X-
X' must be longer than the line indicated in Figure 5.6  in order to site all of the
transects that intersect the sample area.

          2) Randomly choose a point Y between X and X'.

          3) Randomly choose an angle A between 0° and 90°.

          4) Locate transect with endpoints Y and Y', A degrees from transect X
and X1.  If this transect intersects the boundary  of the sample area, mark the
transect.

          5)  Locate another transect beginning at point Y and 90° +A (i.e.,
perpendicular) from that transect that intersects the boundary of the sample area;
then mark the transect Y-Y'.  If this transect intersects the boundary of the sample
area then mark the transect.

          6)  Move away from point Y  on transect X-X' a distance D, where
D=L/sin(A).  L is the desired interval between sampling points along the grid
pattern.

          7)  At the point D units away  from Y, establish two more transects:
one A degrees from transect X-X' and parallel to transect Y-Y', and the  other
90°+A degrees from X-X' also beginning at the point D units from point Y.

          8)  Continue to move intervals of distance D along the transect X-X'
until two transects intersect within the boundary of the sample area.  Establish the
first sample location at that point. Then measure along that transect from the first
sampling location a distance of L and establish more transects and grid points
using the approach described in the previous method for systematic samples.
                                  5-11

-------
             CHAPTER 5: FIELD SAMPLING PROCEDURES
Figure 5.6    Method for Positioning Systematic Sample Locations in the Field
      Sample Locations
                                               Sample Area
                                                Boundary
 A
A    A   A
                  Y+D Y+2D
A
 Xf
 Where D = L/sin A
        Y is chosen randomly
        A is chosen randomly
        L is determined from sample size calculations
        D is a physical sampling location
                               5-12

-------
                 CHAPTER 5: FIELD SAMPLING PROCEDURES
5.4          Extension  to  Stratified Sampling

              The extension of these procedures to stratified sampling is straightforward.
Each stratum is sampled separately using the methods discussed above. Different random
sequences (or random numbers for locating the grids) should be used in each stratum
within the sample area. The sampling approach chosen for one stratum does not have to be
used in another stratum. For example, if a sample area is made up of a small waste pile and
a large 200-acre hillside, then it would be possible to use systematic sampling for the
hillside and random sampling for the waste pile.
5.5          Field Procedures for Determining the Exact Sampling  Location

              The grid points specified for the coordinate system or other reference points
(e.g., trees, boulders, or other landmarks) provide the starting point for locating the sample
points in the field.  The location of a sample point in the field will be approximate because
the sampling coordinates were rounded to distances  that are easy to measure, the
measurement has some inaccuracies, and there is judgment on the part of the field staff in
locating the sample point.

              A procedure to locate the exact sample collection point is recommended to
avoid subjective factors that may affect the results. Without this precaution, subtle factors
such as the difficulty in collecting a sample, the presence of vegetation, or the color of the
soil may affect where the sample is taken, and thus bias the results.

              To locate the exact sample collection point in the field,  use  one of the
following procedures (or a similar procedure) to move from the location identified when
measuring from the reference points  to the final sample collection point.  In the methods
below, M is the accuracy to which distances can be easily measured in the field.

              •      Choose a random compass direction (0 to 360 degrees or N, NE, E,
                    SE, etc.) and a random distance (from zero to M meters) to go to the
                    sample location (as illustrated in Figure 5.7).
             •      Choose a random distance  (from -M meters to M meters) to go in the
                    X direction and a random distance (from -M meters to M meters) to
                    go in the Y direction, based on the coordinate system.

                                      5-13

-------
                CHAPTER 5: FIELD SAMPLING PROCEDURES
Figure 5.7    An Example Illustration of How to Choose an Exact Field Sampling
             Location from an Approximate Location
             Approximate Location
                  Exact
                  Sampling
                  Location
                                                            Randomly chosen
                                                            angle between
                                                            0°and360°
                               Randomly chosen, accurately
                               measured distance from the
                               approximate sampling
                               location
             For either of these procedures, the random numbers can be generated in the
field using a hand-held calculator or by generating the random numbers prior to sampling.
The sample should be collected as close to this exact sampling location as possible.
5.6
Subsampling and Sampling Across Depth
             Methods for deciding how and where to subsample a soil core are important
to understand and include in a sampling plan.  These methods  should be executed
consistently throughout the site.  The field methods that are used will depend on many
things including the soil sampling device, the quantity of material needed for analysis, the
contaminants that are present, and the consistency of the solid or soils media that is being
sampled. The details of how these considerations influence field procedures are not the
subject of this discussion, but they are important and related to the discussion. More detail
can be obtained in the Soil Sampling Quality Assurance User's Guide (USEPA, 1984).

             This discussion describes methods for soil acquisition across depth once an
exact auguring or coring position has been determined and describes how these approaches
                                     5-14

-------
                 CHAPTER 5:  FIELD SAMPLING PROCEDURES
 influence the interpretation of sampling results. There are several approaches that might be
 considered each with advantages and disadvantages; these are outlined in Figure 5.8.


 5.6.1        Depth  Discrete Sampling

              The first approach is to decide before sampling on an exact position or
 positions across depth that will be retained for analysis.  For example, it may be decided
 that throughout the site a split spoon will be driven so that the soil within the following
 intervals is retained and sent to the laboratory for separate analysis: at elevations 1.5 m to
 1.4 m, -0.5m  to -0.6 m, and -4.5 m to  -4.6 m (relative to a geodetic or site standard
 elevation).  The  size of the interval would  depend on the volume required by  the
 laboratory. In this example,  all  the soils  material  within each interval is  extracted and
 analyzed. Advantages  of this approach are that each depth can be considered a different
 sample area and conclusions regarding the attainment of cleanup standards can be made
 independently for each  soil horizon.  This is also a preferred method when the presence of
 volatiles  in the soils media prevents the application of compositing methods.


 5.6.2        Compositing Across  Depth

              Other approaches to sample acquisition  within  a core are  based on
 compositing methods. Compositing methods are generally to be approached with caution
 unless the statistical parameter of interest is the mean concentration. If the mean is the
 statistic of interest, then the variance of  the mean contributed by differences in location
 across the site from composited samples  will be lower than the same variance associated
 with the mean from noncomposited samples.  However, compositing will restrict the
 evaluation of the proportion of soil above an established cleanup standard because of the
 physical  averaging that occurs in the compositing process.  Clearly compositing is not
 recommended if the compositing process  will influence the mass of material in the sample
 as in the case of volatile organics within a  soils matrix.   Numerous authors have
 contributed to the  understanding of the effects of compositing (Duncan, 1962; Elder ejaL,
 1980; Rohde, 1976; Schaeffer and Janardan, 1978; and Schaeffer & flL, 1980), and these
references or a statistician should be consulted if complicated compositing strategies are
planned.
                                      5-15

-------
               CHAPTER 5: FIELD SAMPLING PROCEDURES
Figure 5.8   Subsampling and Sampling Across Depth
   Subsample
   at the same
   depth(s) in
   each soil
   sample core
   Random
   subsamples
   from each
   core are
   mixed
   Entire core
   is mixed
   A single
   randomly
   selected
   location is
   sampled in
   each core
                   Soil Sample
                       Core
                Relative           Possible
                Depth in          Subsampling
Laboratory            Sample
  Samples             Results
              Possible
             Subsampling
                        Xdi

                        Xd2


                        Xd3
                        Xi
                        Xi
                                           Mixed
                         Xi
                                   5-16

-------
                 CHAPTER 5: FIELD SAMPLING PROCEDURES
              Under one compositing method, segments of the soil core are retained from
randomly or systematically identified locations. Then only the sampled portions are
homogenized and then subsampled.  Another approach calls for retaining the entire core
and homogenizing all of the material and then subsampling.  The latter approach is
preferred from a statistical point of view because the subsampling variance will be lower.
However, the second method may present difficulties if the soil samples are obtained to
considerable depth or by split spoon.  In these situations, it is clearly not reasonable or cost
effective to acquire a core from the entire soil profile.  On the other hand, if a hand-held
core or continuous coring device such as a vibra-corer is being used, then homogenization
of the entire core may be possible. In general, large amounts of material, material that is
difficult to manipulate because of its physical properties, material containing analytes that
will volatilize, or hazardous soil make thorough  mixing more difficult, which may
eventually defeat the positive features associated with homogenization of the entire core.
5.6.3        Random Sampling Across Depth

              A final approach involves randomly sampling a single location within each
core. At first, this approach appears to have many difficulties, but if the interest is in
verifying that  the proportion of soil above a cleanup standard is low, this approach will
work quite well.

              Suppose that an in situ soils stabilization method was used to treat all of the
overburden soils  within a former lagoon. The treatment was previously found to yield
effective and  homogeneous results over depth and  space.  It would  clearly not be
appropriate to  sample at a single depth of, say, 3m. Since depth homogeneity is expected,
it may also not be necessary to evaluate several specific depths by sampling 1-m, 3-m, 7-
m, and  15-m horizons in each boring. Finally and most importantly, it would not be
recommended  to perform compositing because the statistical parameter of interest is the
proportion of soil at the site above the cleanup standard and not the mean concentration.

              In this situation it may be useful to pick a random depth at each location. In
this way, many depths will be represented across the lagoon.  Also, cost may be reduced
                                      5-17

-------
                CHAPTER 5: FIELD SAMPLING PROCEDURES
because at many locations the auger will not have to drill to bedrock because the sample
will be obtained from a random location that, in some samples, will be near the surface.
5.7          Quality  Assurance/Quality  Control  (QA/QC)  in Handling  the
             Sample During and After Collection

             Data resulting from a sampling program can only be evaluated  and
interpreted with confidence when adequate quality assurance methods and procedures have
been incorporated into the design.  An adequate quality assurance program requires
awareness of the sources of error associated with each step of the sampling effort.

             A  full discussion of this topic is beyond the scope  of the document;
however, the implementation of a QA program is important. For additional details, see Soil
Sampling Quality Assurance User's Guide (USEPA, 1984), Brown and Black (1983), and
Garner (1985).
5.8          Summary

             Locating soil samples is accompli shed using a detailed map of the waste site
with a coordinate system to identify sampling locations.  The boundaries of the sample
areas (areas within the site for which separate cleanup verification decisions are to be made)
and strata within the sample areas should be shown on the map. It is not necessary to draw
a grid for the entire waste site, only to identify the actual coordinates selected.

             A random sample of soil units within the sample area or stratum will be
selected by generating a series of random (X,Y) coordinates and identifying the location
associated with these coordinates.

             When selecting the sample coordinates for a systematic  sample, two
common patterns of systematic or grid samples are a square grid and a triangular grid.
Various methods can be used to select a systematic sample; however, the most important
point is that one of the systematic sample locations must be identified randomly.
                                      5-18

-------
                CHAPTER 5: FIELD SAMPLING PROCEDURES
             A separate random or systematic sample is selected for each sample area. In
addition, the extension of these procedures to stratified sampling is straightforward.  Each
stratum is sampled separately.  The sampling approach chosen for one stratum, or sample
area does not have to be used in another stratum.

             Once a horizontal position is  chosen, the method of acquiring samples
across depth must be decided.  Methods for subsampling and sampling across depth should
be executed consistently throughout the site. The methods discussed are:

             •      Depth discrete sampling;
             •      Compositing across depth; and
             •      Random sampling across depth.
                                    5-19

-------

-------
          6.  DETERMINING WHETHER THE  MEAN
    CONCENTRATION OF THE  SITE  IS LESS  THAN A
                       CLEANUP STANDARD
             This chapter describes statistical procedures for determining whether the
mean concentration in the sample area attains the cleanup standard.  Testing whether the
mean attains the cleanup standard is appropriate if the mean (or average) concentration is of
particular interest and if the higher concentrations found in limited areas are not of concern.
If the median concentration or the more extreme concentrations (e.g., the concentration for
which 95 percent of the site is lower and 5 percent of the site is higher) are of interest, then
see Chapter 7 for appropriate statistical techniques.

             The statistical procedures given  in this chapter for deciding if the mean
concentration attains the cleanup standard are called "paramedic" procedures. They usually
require certain assumptions about the underlying distribution of the data. Fortunately, the
procedures perform well even when these assumptions are not strictly true, and thus they
are applicable in  many different  field conditions (see Conover, 1980).

             The following topics-determination of sample size; calculation of the mean,
standard deviation, and confidence interval; and deciding if the sample area attains the
cleanup standard-are discussed for each of the following sample plans in the sections
indicated:

             •      Simple random sampling (section 6.3);
             •      Stratified  random sampling (section 6.4); and
             •      Systematic sampling (section 6.5).
6.1          Notation  Used  in This Chapter

             The following notation is used throughout this chapter:

             Cs     The cleanup standard relevant to the sample area and the contaminant
                    being tested.
                                      6-1

-------
CHAPTER 6:  DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
             }i      The "true" but unknown mean contaminant concentrations across the
                    sample area, the population mean.

             HO    The null hypothesis, which is assumed to be true in the absence of
                    significant contradictory data.  When testing the mean,  the null
                    hypothesis is that the sample area does  not attain the  cleanup
                    standard: HQ: |i ^ Cs.

             a      The desired false positive rate for the statistical  test.  The false
                    positive rate for the statistical procedure is the probability that the
                    sample area will be declared to be clean when it is actually dirty.


             HI    The alternative hypothesis, which is declared to be true only if the
                    null hypothesis  is  shown to  be false based  on  significant
                    contradictory data. When testing the mean, the alternative hypothesis
                    is that the sample area attains the cleanup standard:  HI: |i< Cs.

             [LI    The value of |i under the alternative hypothesis for which a specified
                    false negative rate is to be controlled (\i\ < (i).

             |3      The false negative rate for the statistical procedure is the probability
                    that the sample area will be declared to be dirty when it is  actually
                    clean and the true mean is m. The desired sample size n^ is selected
                    so that the statistical procedure has a false negative rate of (i  at \i\.

             n^    The desired sample size for the statistical calculations.

             n      The final sample size, i.e., the number of data values available for
                    statistical analysis including the concentrations that are below the
                    detection level.

             xj    The contaminant concentration measured  for  soil sample  i,
                    i = 1 to n.  For measurements reported as below detection, x, =
                    the detection limit. See section 2.5.2 for more details.
6.2           Calculating the Mean, Variance, and Standard  Deviation


              For many purposes in this chapter it is necessary to calculate the mean,
variance, or standard deviation for a sample of data. The basic formulas are provided in
Box 6.1 for use in later sections.
                                       6-2

-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                  SITE IS LESS THAN A CLEANUP STANDARD
                                    Box 6.1
              Calculating Sample Mean, Variance, Standard Deviation,
                            and Coefficient of Variation

              If the data are a random sample of n observations (i.e., the sample size
    is n), designate the data as xj, X2...,xj,... to xn.  The sample mean (or average),

    indicated by x, is calculated as:
                                   x==£-                  (6.1)

              The formula for the sample variance, s2, is:
                                       n-1

              The formula for the standard deviation is:
                                             n_
                          s=
                                           n-1

              The formula for the coefficient of variation is:
                                     cv=§                    (6.4)
                                         x
             The standard deviation provides a measure of the variability of the sample
data. In particular, it is used to obtain estimates of standard errors and confidence limits.


             Degrees of freedom, denoted by df, provide  a measure of how much
information the variance or standard deviation is based on. The variance and the standard
deviation calculated above for simple random samples have "n-1 degrees of freedom." The
                                      6-3

-------
 CHAPTER 6:  DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
degrees of freedom are used in calculating confidence intervals and performing hypothesis
tests to determine whether the sample area has attained the cleanup standard.
6.3          Methods for Random Samples

             Methods in this section are applicable when the criterion for deciding
whether the site attains the cleanup standard is based on the mean concentration and the
samples are collected using simple random sampling.  The steps involved in the data
collection and analysis are:

       •      Determine the required sample size (section 6.3.2);
       •      Identify the locations within the site from which the soil samples are to be
             collected and collect the physical samples for analysis (Chapter 5);
       •      Perform appropriate statistical analysis using the procedures described in
             section 6.3.3 and on the basis of the decision rule given in section 6.3.4,
             decide whether the site requires additional cleanup.
6.3.1       Estimating the Variability  of the  Chemical  Concentration
             Measurements
             Before  sample collection, determine the number of samples needed to
achieve the desired confidence in the findings.  The number of soil samples depends on the
anticipated variability of the soil measurements. Therefore, an estimate of the standard
deviation of the underlying contamination levels must be obtained. The true value of the
standard deviation is denoted by the Greek letter sigma, o. Estimation of o is discussed
in the next section.

             To estimate the required sample size, some information about the standard
deviation, a (or equivalently the variance a2), is needed.  Unfortunately, the standard
deviation is usually unknown, and steps must be taken to estimate this quantity for the
purpose of determining sample size. The symbol "A" is used to denote that 6 is an estimate
of a.  In practice, & is either obtained from prior data or by conducting a small preliminary
investigation such as a pilot-scale treatability study. Cochran (1977) discusses aspects of
determining a preliminary value for &.
                                       6-4

-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
6.3.1.1     Use of Data from a  Prior Study to Estimate a

             If there are data on contamination levels for the site under investigation from
a previously selected sample of soil units or a treatability study, this information can be
used to obtain 6\  Note that the characteristics of physical samples used in the previous
study should be roughly  the same as  those planned for the present evaluation. For best
results, the sample from  the prior study should be a simple random  sample.  If not, the
sample should  at  least  be "representative" in the  sense that the  measurements are
distributed evenly across the cleanup area.  In particular, measurements that tend to be
located within a specific subarea would  generally be inappropriate for estimating the
variability across the entire area.

             To obtain & from the existing sample, calculate the variance of the chemical
observations. It is best to have at least 20 observations for the variance calculations. The
sample standard deviation,  s, can  be calculated using equation (6.3) in Box 6.1.  Use the
calculated value of s for 6\
6.3.1.2     Obtain Data to Estimate a After a Remedial Action Pilot

             This approach will be best implemented as part of a pilot scale treatability
study.
       1)     Using the sampling procedures described in Chapter 5, select a preliminary
             (simple random) sample of nj = 20 soil units. Determine the concentrations
             for these 20 units.
       2)     From this preliminary sample, compute the standard deviation, s, of the
             contaminant levels. Using s for &, determine the required sample size, n,
             using equation (6.6) in Box 6.3.
                                      6-5

-------
CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
      3)     If the sample size determined is less than or equal to 20, proceed with the
             statistical analysis as outlined in sections 6.3.2, 6.3.3, and 6.3.4, using the
             preliminary sample as the complete  sample.  Otherwise, select enough
             additional soil samples so that the preliminary sample plus the additional
             samples add up to the required  sample size. In this case, the results for the
             initial sample and the supplement should be combined for the statistical
             analysis.
6.3.1.3     An Alternative Approximation for &


             If there are no existing data to estimate o, and a preliminary study is not

feasible, a crude approximation for & can be obtained. The approximation is based on

speculations and judgments concerning the range within which the soil measurements are

likely to fall.  The approximation is based on virtually no data, so the sample sizes

computed from these approximations may  not satisfy the specified level of precision.

Consequently, it should only be used if no other alternative is available.


             The approximation described in Box 6.2 uses the range of possible  soil
measurements (i.e.,  the largest possible value minus  the smallest value).  The range

provides a measure of the  variability of the data. Moreover, if the frequency distribution

of the soil measurements of interest is approximately bell-shaped, then over 99 percent of

the measurements can be expected to lie within three standard deviations of the mean.
                                    Box 6.2
                        An Alternative Approximation for &


                           An estimate of CT is given by:


                                 & = RANGE/6                    (6.5)

      Where RANGE = the expected spread between the smallest and largest values.
                                      6-6

-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
6.3.2        Formulae for Determining Sample Size


             The equations for determining sample size require the specification of

equations 6.6 and 6.7, given in Box 6.3 and the following quantities:  cleanup standard

(Cs), the mean concentration where the site should be declared clean with a high probability
([LI), the false positive rate (a), the false negative rate (P), and the standard deviation
                                   Box 6.3
                     Formulae for Calculating the Sample Size
                          Needed to Estimate the Mean
                                                                (6.6)
    where zj.p and zi_a are the critical values for the normal distribution with
    probabilities of 1- a and 1 - 13 (Table A.2).

             The sample size may also be written in the following equivalent form:
                          -     -a     L
                   nd =     P 2   «    where T = -    -.       (6.7)


             The term i (Greek letter tau) expresses the difference in units of
    standard deviation.  For convenience, the values of n as computed from this
    formula are given in Table A.6 for selected values of a, 3, and ?.
                                     6-7

-------
CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                SITE IS LESS THAN A CLEANUP STANDARD
            Box 6.4 gives an example of calculating sample size.
                                 Box 6.4
                      Example of Sample Size Calculations

             Suppose it is desirable to verify cleanup when the mean concentration
   is .1 ppm below the cleanup standard of .5 ppm (Cs = .5, fi^ = .3) with a power
   of .80 (i.e., |3 = .20). Also suppose o = .43, a = .05, and 99 percent of the
   sample points will result in analyzable samples, then
                                            •=.465
                              a        .43

             From Table A.6 with P = .20, a = .05, and T = .465, the desired
   sample size is between 25 and 30. Using linear interpolation gives a sample size
   of about 29. From Table A.2,

                         Zj_a=  1.645, Z!_P = 0.842.

             Using formula 6.7,

                      (zl-B + zl-a)2   (.842 +
             and
                                    28.6
             Rounding up, the sample size is 29.
                                    6-8

-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                  SITE IS LESS THAN A CLEANUP STANDARD
             Box 6.5 gives an example of determining sample size for testing the mean
using power curves.
                                    Box 6.5
               Example: Determining Sample Size for Testing the Mean
                             Using the Power Curves

              At a former wood processing plant it is desirable to determine if the
    average concentrations of PAH compounds in the surface soil are below 50 ppm
    (the cleanup standard Cs). The project managers have decided that the dangers
    from long-term exposure can be reasonably controlled if the mean concentration in
    the sample area is less than the cleanup standard. The false positive rate for the test
    is to be at most 5 percent (i.e., a = .05).  The coefficient of variation of the data is
    thought to be about 1.2. After reviewing the power curves in Figure A.2 and the
    approximate sample sizes for random sampling, the managers decide:

              1) While it would be desirable to have a test with power curves similar
    to curves E and F, the samples sizes of more than 100 will cost too much.

              2) Power curves A, B, and C have unacceptably low power when the
    mean concentration is roughly 75 percent of the cleanup standard (i.e., 37 ppm),
    the expected mean based on a few preliminary samples.

              3) Thus the  test should have power similar to that in curve D.

              Based on the specifications above and the table at bottom of the Figure
    A.2, the information needed to calculate the sample size is:

                                    a = .05;
                                  P = .20; and
                            14 = Cs  * .69 =  34.5 ppm.

              These values can be used to determine the sample  size using  the
    equations described earlier.
             If the sample size has been specified in advance, perhaps based on cost
considerations, Figures A.I through A.4 can be used to determine the approximate shape
of the power curve for the associated test. See Box 6.6 for an example.
                                      6-9

-------
CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
                                   Box 6.6
                    Determining the Approximate Power Curve
                           for a Specified Sample Size

             Suppose that after review of the budget and analytical costs, the
    managers had  decided that 40 samples would be collected.  What is the
    approximate shape of the power curve for the associated test assuming a = .05,
    P = .20, and a systematic sample is used?

             Based on previous samples the managers believe that the coefficient of
    variation of the concentration measurement will be around 1.1. Assuming that a
    systematic sample will behave statistically like a random sample (a reasonable
    assumption of a site which has been cleaned up) and looking at the bottom of
    Figure A.2 at the sample sized for testing the mean:

              1) If the cv were 1.0, the power curve for a sample size of 40 would be
    between curves C  (sample size = 34) and curve D (sample size = 65), and closer to
    curve C.

              2) If the cv were 1.5, the power curve for sample size of 40 would be
    between curves A (sample size = 25) and curve B (sample size = 43), and closer to
    curve B.

              3) Since the cv is about 1.2, the power curve for the test will be
    between curves B and C.
6.3.3        Calculating  the  Mean, Standard Deviation,  and Confidence

             Intervals


             This section describes the computational procedures used to calculate the

mean concentration and related quantities necessary to evaluate attainment of the cleanup

standard based on a random sample. For concentrations below the detection limit, as

discussed in section 2.5.2, substitute the detection limit in the calculations below.


             The mean of the sampling data is an estimate of the mean contamination of

the entire sample area, but does not convey information regarding the reliability of the

estimate.  Through the use of a "confidence interval," it is possible to provide a range of

values within which the true mean is located.
                                     6-10

-------
 CHAPTER 6:  DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
             The formula for an upper one-sided lOO(l-oc) percent  confidence
limit around the population mean is presented in Box 6.7.  The one-sided confidence
interval should be used to test whether the site has attained the cleanup standard.  The
corresponding decision rule is given in section 6.3.4.
                                   Box 6.7
                 Computing the Upper One-sided Confidence Limit


                             HUa = x + t^tf—                (6.8)


    where x is the computed mean level of contamination, and s is the corresponding
    standard deviation.  The appropriate value of tj a ft can be obtained from Table
    A.I.
6.3.4        Inference:   Deciding Whether the Site Meets Cleanup Standards


             To determine whether the site meets a specified cleanup standard, use the
upper one-sided confidence limit \i\ja, defined above in equation (6.8). Use the following

rule to decide whether or not the site attains the cleanup standard:


             If |J.ua < Cs, conclude that the area is clean (i.e., }i < Cs).


             If Hua ^ Cs, conclude that the area is dirty (i.e., jj. > Cs).


             Box  6.8 presents an example  of an evaluation of cleanup  standard
attainment.
                                    6-11

-------
 CHAPTER 6:  DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
                                    Box 6.8
               An Example Evaluation of Cleanup Standard Attainment

              From Box 6.4 the required sample size is 29.  Assume for this example
    that all 29 field samples were collected and analyzed. Six values were below the
    detection level; these values were included in the analysis at the detection limit.
    Based on these data, the mean is .29 with a standard deviation of .41.  (Note that
                                     41
    this gives a coefficient of variation of-W = 1.48).


              The upper one-sided 99 percent confidence interval goes to

                    = * + 'l-a.df     = -29 + 2.467      = .478 ppm
              Since 0.478 < 0.5, there is a 99 percent confidence that the mean
    concentration of the sample area attains the cleanup standard of 0.5 ppm.
6.4          Methods  for  Stratified Random Samples


             The following sections discuss methods of obtaining an overall estimate of
the mean contamination from a stratified sample. The steps in data collection and analysis
are:

       •      Determine the required sample sizes for each stratum (Chapter 6.4.1);

       •      Within each stratum, identify the sampling locations (Chapter 5).  Collect
             the physical samples  and send the soil samples to the laboratory for
             analysis;

       •      Perform statistical analysis using the procedures described in section 6.4.2,
             and, on  the basis of the decision rule given in section 6.4.3, decide whether
             the site attains the cleanup standard.


             The calculations for stratified samples require knowledge of the proportion
of the surface area or volume of soil represented in each stratum.  The proportion of the
volume of soil can be  calculated using the formula in Box 6.9.
                                      6-12

-------
CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
                                    Box 6.9
                  Calculating the Proportion of the Volume of Soil
              Suppose there are L strata designated by h = 1, 2, 3, ..., L. Compute
    the volume of soil in stratum h as:
              Vj, =  Surface area of stratum h * Depth of sampling in stratum h
    Then the proportion of the volume in stratum h is:

                                  Wh=--                     (6.9)
                                       h=l
6.4.1       Sample  Size  Determination

             The determination of the appropriate sample size for each  stratum is
complicated.  There are methods (Cochran, 1977 or Kish, 1965) for determining the
"optimum" allocation, but these require considerable advance knowledge about the relative
costs and variability within each strata. Consequently, general guidelines, rather than rigid
rules, are given in this  guidance document to assist in planning the sample  sizes for a
stratified sample. These guidelines are expected to cover most situations likely to occur in
the field. For more complex situations, consult Cochran (1977) and a statistician.

             The formulas for sample size use the following notation, where  h indicates
the stratum number:

       n^    The desired sample size for the statistical calculations in stratum h.
       nj,    The final sample size in stratum h, the number of data values available for
             statistical analysis including the concentrations that are below the detection
             level.
       Wh   Proportion of the volume or area of soil in the sample area that is in stratum
             h.

       &,    The estimated standard deviation of measurements from stratum h.  See
        h
             section 6.3.1 on estimating  & within a strata or sample area.  If only an
             overall estimate, ft, is available, use this for all strata.
                                      6-13

-------
CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
      Ch    The relative cost of collecting, processing, and analyzing a soil sample in
             strata h.  If all strata are assumed to have the same relative cost for an
             additional sample, it will be easiest to use C^ = 1 for all strata.

      L     The number of strata.

      xhj    The  reported concentration of the chemical for the ith sample unit in
             stratum h.

      h     The stratum number.
             After strata are defined, it is necessary to decide how many soil units should

be collected in each stratum.  The recommendations below are based on the following

factors:


             •      The physical size of the stratum;

             •      The cost of sampling and processing a soil unit selected from the
                    stratum; and

             •      The underlying variability of the chemical concentration of the soil
                    units in the stratum.


             The "optimum" sample allocation will produce the most accurate estimate of
the overall mean across strata for a fixed total cost. In Boxes 6.10 and 6.11, n^ will

denote the desired sample size to be selected from stratum h. Thus, for a total of L strata,

the overall desired sample size is n^ = ni
-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                  SITE IS LESS THAN A CLEANUP STANDARD
                                   Box 6.11
              An Example Sample Size Determination for a Stratified Sample

              A site consists of two strata (L = 2). Stratum 1 includes loose sand
     soil, while stratum 2 consists of dense hard clay soil mixed with large rocks, but
     the same kind of contamination is present in both strata. Three meters of soil have
     been excavated from both areas. Stratum 1 comprises 10 percent of the sample
     area (W^ = .10, W2 = .90). The sample and analysis costs are considerably
     different in  the two strata.  The cost of sampling and analysis in stratum 2 is
     estimated to be 10 times that in stratum 1 (C\ = 1, €2 = 10) because of the cost
     associated with extracting a soil core. The estimated standard deviation of the
     measurements,  based on previous sampling, is b\ = 25 in stratum 1, while in
     stratum 2, &2 = 13.1. Using a cleanup standard of 40, a = .01, \LI = 35 and (3 =
     .20, the sample size in each strata can be calculated as follows:

                                       2.326 + .842J2 =
                                          40-  35  J


                              = (.10  * 25 *  VI) + (.90 * 13.1 * VTO") = 39.78


              Using equation (6.10),
              and
                         10 *  25
  nld =   39.78  *.401)    "    ^ =399
                           VI

                         90 *  1^1
n2d  =  39.78 *  .401)   '*U -J3'1  = 59.5
                           V10
              Rounding up, and assuming that all samples will be collected and
    analyzed, the final sample sizes are nlf = 40 and n2f = 60.
             When multiple statistical tests are used, or multiple chemicals tested, use the
field sample size in a stratum that is the largest field sample size for any statistical test or

chemical.  Although this procedure for multiple tests will always provide an adequate
sample size, it may not be the most cost efficient
6.4.2       Calculation of the Mean  and Confidence Intervals


             If the number of values below the detection limit is moderate, procedures

and formulae presented in this chapter and in the following boxes based on the sample
                                     6-15

-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
average are applicable.  Section 2.5.2 discusses the adjustment for values below the
detection limit.  If the proportion of values in the data set that have values below the
detection limit in any stratum is large, the procedures in Chapter 7 for testing proportions
may be preferred.
                                  Box 6.12
              Formula for the Mean Concentration from a Stratified Sample
              The overall mean concentration, xst, should be computed as:
                                          "h
                                                                (6.11)
              or
                                Xst =
                                                                (6.12)
             The equations in Box 6.13 give the formula for the standard error of xst.
The  standard error is required for establishing confidence limits around the actual
population mean and deciding if the site attains the cleanup standard.
                                     6-16

-------
CHAPTER 6:  DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                SITE IS LESS THAN A CLEANUP STANDARD
                                 Box 6.13
              Formula for the Standard Error from a Stratified Sample

             The standard error of xst, denoted by s^ is calculated as follows:



                            ST  =   	                    <6'13'

             where
                                  nh     _ 2
                             s2  = i?i(Xhi'Xh)                  (6.14)

                             Sh        V1
             and
                                      ,
                                x, = 1=1
                                 h   """                    (6.15)
            The approximate degrees  of freedom for the standard error can be
calculated using the formula in Box 6.14. The degrees of freedom should be rounded to
the closest integer.
                                 Box 6.14
              Formula for Degrees of Freedom from a Stratified Sample
                                                              (616)
                                  h=l
                                   6-17

-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
             The mean, standard error, and degrees of freedom are used to estimate an
upper one-sided confidence interval with a confidence of 1-a (see Box 6.15).
                                   Box 6.15
                Formula for the Upper One-sided Confidence Interval
                            from a Stratified Sample

              Compute the upper one-sided confidence limit as:

                                                                (6.17)
    where xst is the  mean level of contamination from Box 6.12, and s*  is the
                                                                   SI
    corresponding standard error from Box 6.13.  The appropriate value of ti_a df can
    be obtained from Table A.I.
              The value M-ua^8 a  100(1 -a)  percent confidence interval for the
    population mean.
6.4.3       Inference: Deciding Whether the Site Meets Cleanup Standards


             The test statistic to be used for testing the hypothesis that the site meets
specified cleanup standards is the upper one-sided confidence M-uct' defined above in

equation (6.17).  Use the following rules to decide whether or not the site attains the

cleanup standard. An example illustrating the procedure is in Box 6.16.


             If (lua < Cs, conclude that the area is clean (i.e., p. < Cs).

             If H-ua - Cs, conclude that the area is dirty (i.e., [0. > Cs).
             If the upper one-sided confidence interval of the sample is below the Cs,
then there is 1-a certainty that the mean of the sample area is below the Cs and the site

attains the Cs.
                                     6-18

-------
CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
                                  Box 6.16
                 An Example Illustrating the Determination of Whether
              the Mean from a Stratified Sample Attains a Cleanup Standard

             Following with the example in Box 6.11, the sample area has two
   strata. Stratum 1 comprises only 10 percent of the total site in terms of surface
   area.  The sample consists of 40 units from stratum 1, and 60 units from stratum
   2. After the soil units were analyzed in the laboratory, it was learned that two of
   the units in stratum 1  were below the detection limit.  Hence, the chemical
   concentration for these two cases was set to the minimum detectable level.

             1) Calculate stratum means:  Suppose that for the 40 data values from
   stratum 1 the average concentration of the chemical under study was computed to
   be KI  = 23 ppm.  Similarly, for the 60 data values from stratum 2, suppose that the
   average concentration was determined to be x2 = 35 ppm.

             2) Calculate stratum variances:  Using equation (6.2) the stratum
   standard deviations are: sj =  18.2 and s2 = 20.5. Note that the 38 observations
   in stratum 1 that were above the detection limit, plus the two observations that
   were set to the minimum detectable level, were used in the calculation of Sj.

             3) Calculate overall mean:  Since 10 percent of the site is contained in
   stratum 1, we have Wj = .10, and W2 = .90.  Thus, from  equation (6.11), the
   overall mean for the entire site is:

              xst = W! Xl + W2x2 = (.1)(23) + (.9X35) = 33.8 ppm.

             4) Calculate standard error: The standard error of the  estimate
   computed from the equation (6.13) is:
                      I^+
-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
6.5          Methods for Systematic Samples

             If systematic sampling is chosen, some changes in statistical methodology
may be considered and are discussed in this section.  One concern is that systematic
sampling should be avoided when the pattern of contamination is likely to have a cyclical or
periodic pattern across the sample area. Such a situation might occur if waste was placed in
trenches, if contamination blew into windows, or if a remediation technology is used such
as vacuum extraction, which creates a regular pattern caused by well induced zones of
influence.  In such a case, a systematic sampling pattern may capture only high (or low)
values of the contaminant and therefore yield biased results.  It is presumed that the
likelihood of this pattern will be known in advance, and be used to create strata and the
need to sample randomly.


6.5.1       Estimating Sample Size

             Systematic sampling can result in an  increase in the precision of the
statistical estimates and a corresponding decrease in the required sample size (Cochran,
1977). Unfortunately, the possible advantages of systematic sampling are difficult to
predict before the sample is collected. The sampling precision of an estimated mean from a
systematic sample depends on the pattern of contamination at the site and how the
systematic sample is constructed.  However, the standard error of a mean based on a
systematic sample will usually  be comparable to or less than the standard error of a mean
based on a random sample of the same size. Therefore, using the sample size formulas for
a random sample when the sample was collected systematically usually will be as or more
protective of human health and the environment.

             Use the procedures in section 6.3.2 to determine the sample size required
for a systematic sample.  If other procedures for calculating sample size for a systematic
sample are considered, a statistician should be consulted.
                                     6-20

-------
 CHAPTER 6:  DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
6.5.2       Concerns Associated  with Estimating the Mean,  Estimating the
             Variance, and Making Inference from a Systematic Sample

             When a systematic sample is obtained, apply the same methods used for a
random sample. As with a simple random sample, the simple average of the sample points
is an unbiased estimate of the population average. Note, however, that the number of
sample points in a systematic sample of an irregularly shaped area may vary from the
targeted sample size. A smaller sample size will produce estimates that have less precision
than larger samples, but will not introduce bias. The loss in precision tends to be negligible
except for small sample sizes.

             In general, an unbiased estimate of the standard error of a mean based on a
systematic sample is not available. In the special case where contamination is distributed
randomly over the sample area, unbiased estimates of the standard error can be constructed.
This situation may be approximately the case after a careful cleanup has been  done where
the cleanup has  effectively  removed the contaminated soil  from all of the  high
contamination areas or the soil is being mixed, fixed, or incinerated.

             Several methods are commonly used to estimate  the standard error of a
mean from a systematic sample (Koop, 1976; Wolter, 1984; Tornqvist, 1963; Yates,
1981). These methods treat the systematic observations as:

             •      A random sample;
             •      A stratified sample; and
             •      A serpentine pattern of observations that employs  a special variance
                    calculation procedure.

             It is suggested that the serpentine pattern be used with overlapping pairs of
points as  the principal method of estimating the standard errors in a systematic design.
However, if the boundaries of the sample area are so irregular as to  make this approach
difficult, the stratification approach is recommended. The random sample estimate should
seldom be used. These approaches are discussed below.
                                     6-21

-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
6.5.2.1     Treating a Systematic Sample as a Random Sample

             The simplest method of estimating the standard error for a systematic
sample is to use the variance formulas in Box 6.1 for a simple random sample (section
6.3). This method is appropriate if the contamination is distributed randomly across the
sample area. If there are gradients of contamination, or if there are substantial contiguous
areas that have higher (or lower) than average contamination, this method can be biased
(Osborne,  1942). In this case, the actual standard error of the mean will, on average, be
smaller than that computed from the simple random sample formulas.  Thus, the sample
estimate will appear to be less precise than it really is and there will  be a tendency to take
more observations than are necessary or to do more cleanup work than is necessary.


6.5.2.2     Treating the Systematic Sample as a Stratified Sample

             An estimate of the standard error that is less subject to bias than the random
sample estimate can be obtained by aggregating adjacent points in the  systematic design into
groups, and treating these groups as  though they were strata (Yates, 1981) as depicted in
Figure 6.1. It should  be noted that this grouping can be done whether or not  the sample
area was previously  stratified.  If  stratification was used,  grouping for purposes of
estimating standard errors would be done within strata (see Box 6.17).

             A commonly used group size  consists of four observations. The groups
need not be the same size, but efficiency is gained if they are  nearly the same size and if
they are small. Points in a group should be adjacent and the groups must cover the sample
area comprehensively. One must not form the groups on the basis of the observed data--
this would add bias. Instead, they should be formed strictly  on the basis of geographic
adjacency  and boundary restrictions without regard to their observed values.  If the sample
locations form a square grid, the recommended grouping will be  four adjacent sample
locations forming a square. (At the edge of the strata or sample area, the clusters of four
points might not form squares due to  irregular boundaries.)

             Although the average contamination measure is computed in the usual
manner as the sum of all observations divided by the number of them, the average may be
                                     6-22

-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
considered as a weighted sum of the group means, where the weights are the number of

observations in the group.
Figure 6.1     An Example of How to Group Sample Points from a Systematic Sample so
             that the Variance and Mean Can Be Calculated Using the Methodology for a
             Stratified Sample
             The tests described in section 6.3 for simple random samples can be adapted

for systematic samples by simply replacing the quantities s/Vn in equation (6.8) with the

expression for the standard error given in equation (6.19).  Box 6.18 gives a formula for

an upper one-sided confidence interval for the true mean contamination when a systematic
sample is treated as a stratified sample.
                                    6-23

-------
CHAPTER 6:  DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                SITE IS LESS THAN A CLEANUP STANDARD
                                  Box 6.17

               Estimating the Mean, the Standard Error of the Mean,
                and Degrees of Freedom When a Systematic Sample
                        Is Treated as a Stratified Sample

                                    L
                              -   1
                                                             (6-18)
                                    h=l
   where h denotes the h* group, L is the total number of groups, nh is the number

   of observations in group h, xh is the mean of the observation in group h, and n is
   the total number of observations in the sample.

             The estimated standard error of the mean, s^, can then be computed as:
                                                         (6.19)
   where sj,2 is the variance of the observations in group h as computed from the
   equations in section 6.2. The degrees of freedom are computed as: df = n-L.
                                  Box 6.18
          Formula for Upper One-sided Confidence Interval for the True Mean
        Contamination When a Systematic Sample Is Treated as a Stratified Sample
             For example, the upper one-sided confidence interval for the true mean
    contamination is:


                                  = * +tl-a,dfSx                 (6.20)
             In this case, the sample area would be declared to be clean if
    less than the cleanup standard; otherwise the sample area would be declared to be
    dirty.
                                     6-24

-------
 CHAPTER 6:  DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
6.5.2.3     Linearization  and  Estimates  from  Differences  Between
             Adjacent  Observations of a Systematic Sample

             Another commonly used method is to linearize the systematic pattern by
forming a serpentine association between each observation and the one preceding it in a
serpentine pattern. Consider the example pattern in Figure 6.2. The numbers represent the
sample points and their location in the sample area.

             The numbers string the pattern into a linear sequence.  The difference
between the observations of an adjacent pair contain a systematic component that represents
the "true" difference between them plus a random component.  The systematic component
represents bias but, since the two members of the pair are adjacent geographically, one
would expect the systematic component of the difference to be small compared to, for
example, comparing point 1 with point 19.
Figure 6.2    Example of a Serpentine Pattern
          Numbers indicate the sequence (i) required for the calculations in Box 6.19.

             To estimate the standard error from a serpentine pattern makes use of
overlapping pairs.  That is, point 1 is compared with point 2, 2 with 3, 3 with 4, and so
on. The method gives a somewhat more precise estimate of the standard error.  The
method is shown in Box 6.19.
                                    6-25

-------
CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
                                  Box 6.19
             Computational Formula for Estimating the Standard Error and
          Degrees of Freedom from Samples Analyzed in a Serpentine Pattern
                        = A/
                            \
(l/2n)  I(xi-xi_i)2/(n-l)         (6.21)
       i=2
    df = 2n/3
             The associated number of degrees of freedom in Box 6.19 is given
approximately by DuMouchel el aL (1973).

             It should be noted that the serpentine pattern can be constructed by moving
from top to bottom, from right to left, or diagonally within the systematic pattern.  The
pattern should be planned prior to sampling. If it is suspected that there will be a gradient
in the data, say from top to bottom, then the serpentine pattern should be formed so that it
follows the contours of the gradient to the extent that it is feasible to do so.


6.6          Using  Composite Samples  When Testing  the  Mean

             "Compositing" refers to the process of physically combining and mixing
several individual soil samples to form a single "composite" sample (see Rohde, 1976 and
1979; Duncan, 1962;  Elder el al, 1980; Gilbert, 1987, Gilbert el al, 1989, and section
5.6.2 of this document). A primary advantage of compositing is that it reduces the number
of lab analyses that must be performed.

             Composite samples can be created using the following procedure:

       •      Collect the samples using a random or systematic sample design, collecting
             n soil samples from the field;
       •      Physically mix randomly selected groups of ten  samples to create n/10 = m
             samples, which are sent to the lab for analysis; and
                                     6-26

-------
 CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                  SITE IS LESS THAN A CLEANUP STANDARD
       •      Perform the statistical analysis on the m lab results to determine if the mean
              attains the cleanup standard.

              In the procedure above, each soil sample sent to the lab was composited
 from 10 original samples; the compositing factor was 10.  The  compositing was
 accomplished by mixing the first 10 randomly selected samples, the second 10 randomly
 selected samples, etc. to get the final m samples to send to the lab.  To specify how
 compositing is done, both the method of selecting the samples that get mixed together and
 the compositing factor must be specified.  In addition, compositing requires that each
 original sample is the same or known physical size in terms of volume or weight and that
 the samples are very well mixed. These criteria may be difficult to achieve. This possible
 advantage will be reduced if the mixing is not complete or uses soil samples of different
 physical sizes. Nevertheless, as mentioned above, the number of lab analyses that must be
 performed may be greatly reduced.

              Other considerations are the decisions related to how best to composite the
 original samples, the  number of soil samples to collect, and the number of soil samples to
 send to the lab.  If the laboratory error is large, compositing may provide little benefit. The
 specification of which samples to combine will be affected by the sample design and the
 variability across the sample area, among other things. For some types of soil or chemicals
 being tested, mixing will affect the laboratory analysis. For example, mixing samples with
 volatile organics may  release contaminants.

              Compositing can be a useful technique if the mean is to be tested, but must
 always be considered and implemented with caution. Compositing should never be used if
 percentiles or proportions are used as the attainment criteria. Other methods of compositing
 are discussed  by Gilbert (1987).  If compositing is considered, consultation  with  a
 statistician is recommended.
6.7         Summary

             The methods in this chapter apply when the cleanup standard is intended to
control the average conditions at the site, not simply the average of the sample. The mean
estimated from a sample must be sufficiently below the cleanup standard to ensure with
confidence that the entire site is below the cleanup standard.
                                     6-27

-------
 CHAPTER 6:  DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                 SITE IS LESS THAN A CLEANUP STANDARD
             Basic formulas are provided to calculate the mean, variance, or standard
deviation for a sample of data. The standard deviation provides a measure of the variability
of the sample data and is used to obtain estimates of standard errors and confidence limits.
These statistics help determine how far the sample mean must be below the cleanup
standard to ensure with reasonable confidence that the site mean is below  the cleanup
standard.

             For a random sampling, the number of soil samples required depends on the
anticipated variability of the soil measurements. To estimate the required sample size, some
information about the standard deviation, a, or the variance CT^, is needed.  Steps to
estimate a  are discussed. Equations for determining sample size require the  following
quantities: cleanup standard (Cs), the mean concentration where the site should be declared
clean with a high probability (m), the false positive rate (a), the false negative rate (P), and
the  standard deviation (a).  The mean of the sampling data is an estimate of the  mean
contamination of the entire sample  area.  The use of an upper "confidence interval"
provides an upper bound on  the true sample area mean.  When a one-sided  100  (1-ot)
percent upper confidence limit of the mean is less than the Cs, the site is judged clean.

             Estimating the mean contamination from a stratified sample requires
considerable advance knowledge about the relative costs and variability within each strata.
Guidelines  and formulae are  given to assist in planning the sample sizes for a stratified
sample and how many soil units should be collected in each stratum. They are also given
for  establishing the standard  error, the  approximate degrees of freedom for the standard
error, and the upper one-sided confidence interval. If the  upper one-sided confidence
interval on the sample mean is below the cleanup standard (Cs), cleanup is verified.

             If systematic  sampling is used, special  methods are required;  these
procedures are discussed and illustrated. To estimate the  standard error for a systematic
sample, formulae used for a simple random sample and  a stratified sample may be applied,
as can be the method of linearizing the systematic pattern  into a serpentine pattern. Two
estimates of the standard error are common  when the points (sampling locations) have been
linearized; these are discussed.
                                      6-28

-------
CHAPTER 6: DETERMINING WHETHER THE MEAN CONCENTRATION OF THE
                SITE IS LESS THAN A CLEANUP STANDARD
            Compositing samples—the act of physically combining and mixing several
individual soil samples to form a single composite sample—is discussed. Its primary
advantage is that it reduces the number of lab analyses that must be performed.
                                 6-29

-------

-------
   7.   DETERMINING WHETHER  A  PROPORTION  OR
        PERCENTILE OF THE SITE IS LESS THAN A
                       CLEANUP STANDARD
             This chapter describes statistical procedures for determining with confidence
whether a specified proportion of the soil is less than a cleanup standard.  The extreme
concentrations at a hazardous waste site are often of primary concern. In this case, an
appropriate statistical test can be based on either a high percentile of the distribution of
chemical measurements over the area, or on a large proportion of the area that has
concentrations less than the cleanup standard.  For example, the methods in this chapter
apply if there is interest in verifying that a large percentage (e.g., 90,95, or 99 percent) of
the soil at the site has concentrations below the cleanup standard.

             Throughout Chapter 7 the statistical evaluations are designed to detect when
a large proportion of the site is less than a cleanup standard.  However, there is another
equivalent way of stating this objective: these evaluations are designed to ensure that no
more than a small proportion of the site is above the cleanup standard.  The numerical
methods in this chapter are designed and presented in the context of the second approach.
Therefore, we will be testing to verify that only a small proportion or percentage of the site
(e.g., 10, 5, or 1 percent) exceeds the cleanup standard.

             Two approaches to testing percentiles  and proportions are discussed in this
chapter
             •      Exact and large sample nonparametric tests for proportions based on
                    the binomial distribution; and
             •      A parametric test for percentiles based on tolerance intervals, which
                    assumes the data have a normal distribution.

             In the nonparametric approach, each soil sample measurement is designated
as either equal to or  above the cleanup standard, Cs  and coded as "1," or below Cs and
coded as "0."  The analysis is based on the  resulting data set of O's and 1's.  The
proportion of the soil (or equivalently, the percentage of the area under investigation) at or
above the cleanup  standard can be estimated from the coded data.  If the proportion of 1's
is high, the site will be declared contaminated.  On the other hand, if the proportion of O's
                                     7-1

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
is sufficiently large, the area is considered to have attained an acceptable level of cleanup.
A test based on proportions works with any concentration distribution and requires only
that the cleanup standard be greater than the analytical method detection limit.  However,
this method has limitations because it does not consider how far above or below the Cs the
data value is, only if it is above or below.

             The second approach for testing percentiles of the  concentration
distribution, which does not require coding the data as above, is based on  estimating a
confidence interval for a percentile of the normal distribution. These intervals are called
tolerance intervals (Guttman,  1970).  The assumption that  the data have a normal
distribution (or that a suitable transformation of the data is approximately  normal) is critical
to this test.  In addition, this method may  be biased if more than  10  percent of the
observations are below the detection limit.

             The following sampling and analysis plans are  discussed in the sections
indicated:

             •      Simple random sampling for proportions (section 7.3);
             •      Stratified random sampling for proportions (section 7.5);  and
             •      Simple random sampling for testing percentiles of a normal
                    distribution (section 7.6).
7.1         Notation Used in  This Chapter

             The following notation is used throughout this chapter:

             Cs    The cleanup standard relevant to the sample area and the contaminant
                    being tested (see section 3.4 for more details).
             P     The  "true"  but  unknown proportion of the  sample area with
                    contaminant concentrations greater than the cleanup standard.
             PQ    The criterion for defining whether the sample area is clean or dirty.
                    According to the attainment objectives, the sample area attains the
                    cleanup  standard  if the proportion of the  sample area with
                    contaminant concentrations greater than the cleanup standard is less
                    than PQ, i.e., the sample area is clean if P < PQ.
                                       7-2

-------
  CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
              HO    The null hypothesis, which is assumed to be true in the absence of
                    significant contradictory data.  When testing proportions, the null
                    hypothesis is that the sample area does not attain the cleanup
                    standard; HQ: P > PQ.

              a     The desired false positive rate for the statistical test to be used. The
                    false positive rate for the statistical procedure is the probability that
                    the sample area will be declared to be clean when it is actually dirty.

              HI    The alternative hypothesis, which is declared to be true only if the
                    null hypothesis  is  shown to be false  based  on  significant
                    contradictory data.  When testing proportions, the  alternative
                    hypothesis is that the sample area attains the cleanup standard; HI: P
                     Cs),
                    thenyj= 1.
7.2          Steps to Correct for Laboratory Error


              All of the procedures for estimating proportions and percentiles assume that

the chemical concentrations can be measured with little or no error. If there is substantial

variability in the measurement process, the corresponding estimates of proportions may be

biased (Mee el aj.., 1986 and Schwartz,  1985).  If an upper percentile (greater then the
                                       7-3

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
median) is being tested, the bias may be conservative. In other words, the sample area may
be cleaner than the statistical test would indicate. This bias will be more important in some

situations than in others:

             •      The measurement error causes  no problems if the median (50th
                    percentile) is tested;

             •      The measurement error is likely to be the greatest problem when the
                    percentile to be tested is between  the 75th and 99th; and

             •      The measurement error is likely to be  the greatest problem if the true
                    proportion of contaminated soil  samples is close to the proportion
                    being tested, i.e., the sample area just attains the cleanup standard.


             There are three possible ways to reduce the bias:

             •      Use a more precise analytical method  that has a smaller measurement
                    error,

             •      Perform multiple laboratory measurements on each soil sample and
                    use the average or median measurement in the statistical analyses
                    (see the example in Box 7.1); or

             •      Perform more cleanup of the sample area than is required to attain
                    the cleanup standard.
Box 7.1
Illustration of Multiple Measurement Procedure for
Reducing Laboratory Error
Soil
unit

1
2
3
Measurement (ppb)
1

<50
75
<50
2

95
105
<50
3

101
102
55
Sample
median
(ppb)
95
102
<50
Coded
result

0
1
0
Detection limit - 50ppb
Cs - 100 ppb
                                       7-4

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
7.3          Methods for Simple Random Samples

             This section describes statistical analysis procedures that apply when the
criterion  for deciding whether the site attains the cleanup standard is based on the
proportion of contaminated soil units and when the soil samples are selected by simple
random sampling.  The basic steps involved in the data collection and analysis are:
      •      Determine the required sample size (section 7.3.1);
      •      Identify the locations within the site from which the soil units are to be
             collected, collect the physical samples, and send sampled material to
             laboratory for analysis (Chapter 5);
      •      Perform appropriate statistical analysis using the procedures described in
             sections 7.3.3 and 7.3.4 and, on the basis of the statistical analysis, decide
             whether the site requires additional cleanup.

             Although the use of random samples is recommended, random sampling
may not be practicable. An alternative is to select a systematic or grid sample using the
procedures described in Chapter 5.  Systematic samples may be easier to collect and will
provide valid estimates of proportions, but may produce a poor estimate of sampling error.
7.3.1       Sample Size Determination

             The sample sizes as computed in Box 7.2 are summarized in Tables A.7
through A.9 for selected values of P0 and PT and for the following values of a and 0:
a =0.01,  0.05, and 0.10, and  (3 = 0.20.  In most cases, Tables A.7 - A.9 will be
adequate for practical application.  However,  for values not in the tables, use equation
(7.1) below.  Notice that the cleanup standard is not required in order to determine the
sample size.
                                     7-5

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
                                    Box 7.2
           Computing the Sample Size When Testing a Proportion or Percentile

              Given  the quantities, PQ, PI, a, and (3, the sample size  can be
    computed from the following formula:
                         ^--^-o
                   nd=t — E - p  . p - )       (7.1)
                                     ro  ri
    where  zj.p and zj. aare  the critical values for the normal  distribution with
    probabilities of 1-oc and l-(3 (Table A.2).
7.3.2       Understanding Sample Size Requirements

             To illustrate the use of the sample size tables, consider the following
scenario (also see Box 7.3). A sample area will be considered clean if less than 20 percent
of the area has concentrations of mercury greater than 1,5 ppm.  That is, PQ = .20 in this
example. The null hypothesis is HQ:? > .20 and specifies that if 20 percent or more of the
sample area has concentrations exceeding 1.5 ppm, the area is still considered dirty and
requires further remedial action.

             Further suppose that the site  manager wants no more  than  a 5 percent
chance of declaring the sample area to be clean when it is dirty (i.e., a = .05).  Moreover,
the site manager wants to be 80 percent certain that if only 10 percent of the area has
concentrations exceeding 1.5 ppm the site will be found clean. That is, for PI = .10, he
wants the false negative rate to be moderately low,  say 20 percent (i.e., (3 = .20). From
Table A.8 (corresponding to values of a = .05 and P = .20), the required sample size for
PQ = .20 and PI  = .10 is rid = 83.

             It is evident from Tables A.7 - A.9 that as the value of PI approaches PQ,
the required sample sizes become larger. For example, if the manager in the above example
wanted the false negative rate to be 20 percent for PI =  .15 (instead of PI = .10), the
required sample size would be 368. Such a large sample size may be impractical for many
waste site investigations. If the cleanup technology is designed to achieve levels that are
only slightly less (Pi) than the cleanup objective (Po), then many samples will be required
                                      7-6

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
to verify attainment of PQ.  If the cleanup is highly effective and PI is well below PQ, then

few samples will be required to verify cleanup.
                                   Box 7.3
                    Example of How to Determine Sample Sizes
              When Evaluating Cleanup Standards Relative to a Proportion

              Soil has been removed from a lagoon bottom that previously contained
    corrosive waste.  The exposed soil will be sampled to determine whether more
    excavation is required.  Wanting  to minimize the possibility of future  health
    effects, the site will be judged in attainment of the cleanup standard if there is 90
    percent confidence (a = . 10) that less than 10 percent (P0 = . 10) of the topsoil has
    concentrations exceeding the cleanup standard.  The expected proportion of
    contaminated soil is low, less than 5 percent The manager wants to be 80 percent
    confident (P = .20) that the sample area will be declared clean if the proportion of
    contaminated soil is less than 2 percent (Pi = 2 percent).

              From Table A.9,  for P0 = 0.10 and PI = 0.02, the required sample
    size is n = 39.

              Using formula (7.1), from Table A.2, zi_a =  1.282 and zt.p = .842
    and:
  fzi
=l
                  nd   — - p  . p
                                    ro   ri
  r.842V.02(.98) + 1 282V.10(.9Q) 12
~l           fm7V5           i
                                  .10 - .02

                                    = 39.4
7.3.3        Estimating the  Proportion  Contaminated and the  Associated
             Standard Error
             This section describes the computational procedures to be used to calculate

the proportion contaminated (see Box 7.4) and related quantities necessary to evaluate

attainment of the cleanup standard.
                                     7-7

-------
 CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
              THE SITE IS LESS THAN A CLEANUP STANDARD
                                   Box 7.4
                     Calculating the Proportion Contaminated
                     and the Standard Error of the Proportion
             Set yj = 1 if the concentration in sample i is greater than the cleanup
    standard and yj = 0 otherwise.  If n = the total number of samples available for
    statistical analysis, the proportion of samples, p, above the cleanup standard can be
    calculated using the following equations:
                                   r =   yi                      (7.2)
                                      i=l
             where yi = 1 if x{ > Cs or yi = 0 if xj < Cs

                                    P = £                       (7.3)

             The standard error, Sp, of the proportion p is
                              Sp = V n n  v'  •                 (7.4)
             These results are used to estimate upper one-sided confidence intervals,
which allow determination of whether the site has attained the prescribed cleanup standard.

             If the sample size is sufficiently large, an approximate confidence interval
may be constructed using the normal approximation (see Box 7.5, section 7.3.4).  If the
sample size is small, an "exact" procedure should be used to calculate the confidence
interval (see Box 7.6, section 7.3.5).
7.3.4       Inference:   Deciding Whether a  Specified Proportion of  the
             Site is  Less than a  Cleanup  Standard Using a  Large  Sample
             Normal  Approximation

             When np > 10 and n(l-p) > 10, the large sample normal approximation can
be used for evaluating the statistical significance of the number of sample values equal to or
                                     7-8

-------
  CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTER OF
              THE SITE IS LESS THAN A CLEANUP STANDARD
above Cs. This condition will generally only be met for tests of percentiles between 10 and

90. If the condition is not met, the exact test should be used.
                                   Box 7.5
              Calculation of the Upper Confidence Limit on a Proportion
                   Using a Large Sample Normal Approximation

              Compute the following:

                                PU = P + zl-a sp                     (7-5)

              If PU < PQ, conclude that the area has attained the cleanup standard.

                   > PQ, conclude that the area has not attained the cleanup standard.
7.3.5       Deciding Whether a Specified  Proportion  of the  Site  is  Less
             Than the Cleanup Standard Using an Exact  Test


             If the normal approximation is not appropriate,  the  "exact" procedure

described below should be used to test whether the proportion meets the cleanup standard.
However,  if the sample size is too small, it may not be possible to construct a useful
decision rule with the stated false positive rate. These instances are indicated in the tables

(Tables A.7 - A.9) used to perform the tests.


             Use the following to perform the exact test:

             •     Given n, a, and P0, determine the "critical value" of the test, ra;n,
                   by referring to Table A. 10. To use this table, a must be .01, .05 or
                   .10, respectively. To determine the critical value, select the column
                   for PQ specified in the attainment objectives.  Reading down the
                   column find the first number greater than the sample size n. Move
                   up one row and read ra:n, the critical value, in the leftmost column.
                   If the number in the first row of the selected column is greater than
                   the sample size, there are not enough data to perform the given test.
                   If the bottom number in the selected column is less than the sample
                   size, use the normal approximation above.

             •     From the sample, determine the number, r, of soil units that have
                   chemical concentrations exceeding Cs. Compare r with ra;n.
                                      7-9

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
             •      If r < ra;n, conclude that the area has attained the cleanup standard.

             •      If r > ra;n, conclude that the area has not attained the cleanup
                    standard.


             For values of n, a, and P0 that are not given in the tables, the critical value

for the "exact" test may be determined directly using the algorithm below or using an
equivalent procedure from Brownlee (1965, p. 148-150) based on the F distribution.


   Step 1    Compute f(0) = (1 - P0)n.

             Step la       If f(0) > a, then set ra;n = 0 and stop. Note that if f(0) >
                           a, a test with the specified false positive rate is not possible;
                           the actual false positive rate would be f(0).

             Step Ib       If f(0) < a, go to Step 2.


   Step 2    Compute


                              f(l) = n (r-V-) f(0)                      (7.6)
                                       1 " ro

                           where f(0) is computed in Step 1.

   Step 3    Next, compare f(0) + f(l) with a.  If f(0) + f(l) >  a, set ra;n = 0, and
             stop.  If f(0) + f(l) < a, define a "temporary" variable, y, and set y = 1.
             Go to Step 4.

   Step 4    For the given value stored in the temporary variable, y, compute f(y) using
             the recursion formula below:
                         f(y) =         ) (Tr) f(y-D-                (7-7)
    Step 5     Compare f(0) + f(l) + ... + f(y) with a. If f(0) + f(l) + ... + f(r) > a, set
              ra;n = y and stop.  If f(0) + f(l) + ... + f(r) < a, increment the temporary
              variable by 1, i.e., set y = y+1, and go to Step 3. Repeat Steps 4 and 5
              until the process stops and ra;n has been determined.

              Box 7.6 gives an example of an inference based on the "exact" test
described above.
                                      7-10

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTTLE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
                                    Box 7.6
                  An Example of Inference Based on the Exact Test

              Assume that only 9 samples collected from 203 sample locations have
    concentrations greater than the cleanup standard, i.e., r = 9, and remember that n
    = 191, a = .05, and P0 = .05.
              Using Table A. 10 read down the column headed by PQ = ,05 and find
    the first number greater than the sample size, in this case 208 in row 6. Go up one
    row and read ra:n from the lefthand column. The value in the left column and fifth
    row is 4 = ra:n.
              Since r > 4, the sample area does not attain the cleanup standard.
7.4         A Simple Exceedance Rule Method for Determining  Whether a
             Site Attains the Cleanup Standard

             One of the most straightforward applications of the methods in this chapter
involves the design of zero or few exceedance rules. To apply this method, simply require
that a number of samples be acquired and that zero or a small number of the concentration
measurements be allowed to exceed the cleanup standard.  This kind of rule is easy to
implement and evaluate once the data are collected; it only requires specification of the
sample size and number of exceedances as indicated in Table 7.1.

             In addition, these rules  also have statistical properties.  For example, the
more samples collected, the more likely that one sample will exceed a cleanup standard.
That is, it is more likely to measure a rare high value with a larger sample. In addition, the
larger the proportion of the site that must have concentrations below the cleanup standard,
the more soil samples that  will be required to document this with certainty. Finally,
because of the  chance of outliers, it may be that the rule  that allows one or more
exceedances would be preferred in order to still have the site judged in attainment of the
cleanup standard. If more exceedances are allowed, more soil samples are required to
maintain the same statistical performance and proportion of the site that is clean.
                                     7-11

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
             Table 7.1 illustrates these tendencies and offers selected sample sizes and
exceedance rules as a function of statistical performance criteria. For example, if there is
interest in: verifying that 99 percent of the site is below a cleanup standard; keeping the
chance of saying the site is clean when it is dirty at 1 percent; and allowing no exceedances
of the cleanup standard, then 459 soil samples would be required. If 459 samples were
obtained and none of them exceeded the cleanup standard, there is 99 percent confidence
that 99 percent of the site is less than the cleanup standard.  If three exceedances were
allowed and the  same statistical performance criteria were required then 1001 soil samples
would be required and 998 of the measurements would have to be less than the cleanup
standard.

             On the other hand, if the statistical performance criteria are relaxed, sample
size requirements decrease.  For example, if there is interest in allowing no exceedances
and a false positive rate of 90 percent that 90 percent of the site is less than the cleanup
standard, then 22 samples would have to be obtained and all results would have to be less
than the cleanup standard.  If three exceedances were permitted and the same statistical
criteria were applied, then 65 samples would be required and 62 of the measurements
would have to be less than the cleanup standard.


7.5          Methods for Stratified  Samples

             In some circumstances it may be useful to establish a stratified sampling
regime as discussed in Chapter 5.  If the waste can be divided into homogeneous subareas,
the precision of an estimated proportion can often be  improved through the  use of a
stratified sample. These homogeneous areas from which separate samples are drawn are
referred to as "strata," and the combined sample from all areas is referred to as a "stratified
sample."

             The statistical procedures discussed here  apply  when the criterion for
deciding whether the site attains the cleanup standard is based on the proportion of
contaminated soil units.  The basic steps involved in the data collection and analysis are:
       •      Determine the required sample sizes for each stratum using the equation in
             section 7.5.1;
                                      7-12

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
       •      Within each stratum, identify the locations within the site from which the
             soil units are to be selected, collect the physical samples, and send sampled
             material to laboratory for analysis (Chapter 5); and
       •      Perform appropriate statistical analysis using the procedures that follow
             (sections 7.5.2 and 7.5.3) and, on the basis of the statistical results, decide
             whether the site has attained the cleanup standard.

             The tests described in this section assume that soil samples within each
stratum are collected randomly.   Although the use of  simple random samples is
recommended, simple random sampling may not always be practicable.  An alternative
method would be to select a systematic (grid) sample; however, this type of sampling
should be approached with caution as described in section 7.3 and Chapter 6.
7.5.1       Sample  Size  Determination

             Determination of the appropriate sample size is complicated in stratified
sampling because there are many ways the sample can be allocated to strata. For example,
if 100 soil units will be sampled, a decision must be made on whether to allocate the sample
equally among strata, in proportion to the relative size of each strata, or according to some
rules. There are methods for determining the "optimum" allocation; however, these require
considerable  advance  knowledge  about  the underlying variability  of each strata.
Consequently, the equations below are general guidelines to assist in planning the sample
sizes for a stratified sample. These guidelines will cover many field situations.  For more
complex situations, a text such as Cochran (1977) should be consulted.

             The formulas for sample  size use the following notation, where h indicates
the stratum number:
             h      As a subscript, indicates a value for a stratum within the sample area
                    rather than for the entire sample area.
             njxj    The desired sample size for the statistical calculations.
             nh     The final sample size, the number of data values available for
                    statistical analysis including the concentrations that are below the
                    detection level.
             Wh    Proportion of the volume of soil  in the sample area which is in
                    stratum h
                                     7-13

-------
 CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
              THE SITE IS LESS THAN A CLEANUP STANDARD
                   Cost of collecting, processing, and analyzing one additional soil
                   sample, on a relative scale.

                   The number of strata.

                   The  scored concentration data, where  yhi = 1 if the measured
                   concentration is greater than the cleanup standard and 0 otherwise.
Table 7.1     Selected information from Tables A.7 - A.9 that can be used to determine
             the sample sizes required for zero or few exceedance rules associated with
             various levels of statistical performance and degrees of cleanup
Chance of Saying the
Site is Clean When
It is Dirty
(Certainty)
Proportion of
the Site That
Is Clean
    Sample Size Requirements
    Under Various Numbers of
    Allowed Exceedances of the
    Cleanup Standard
False Positive Rate, Alpha
       (1 - Alpha)
      1-PO
Number of Allowed Exceedances
  0135
       .01

       (.99)
     .99

     .95

     .90
 459   662   1001  1307

  90   130    198   259

  44    64     97   127
        .05

      (.95)
     .99

     .95

     .90
 299   473    773  1049

  59    93    153   208

  29    46     76   103
        .10

       (.90)
     .99

     .95

     .90
 230    388   667   926

  45     77   132   184

  22     38    65    91
                                     7-14

-------
  CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
             Once the sample area has been divided into strata, it is necessary to decide
how many soil units should be collected in each stratum. The equations below will provide
an "optimal" sample size for each  stratum provided that the following information is
available:

             •      The physical size of the stratum;

             •      The cost of sampling and processing a soil unit selected from the
                    stratum;

             •      The underlying proportion of the soil units in the stratum that are
                    contaminated, i.e., have chemical concentrations exceeding the
                    specified cutoff,  Cs; and

             •      The overall desired accuracy of the test.


             An optimum sample allocation to each stratum will produce the most
accurate measure of the proportion of soil contaminated across strata in the entire sample
area for a fixed total cost. In what follows, n^ will denote the corresponding sample size to

be selected from stratum h.  Thus, the total sample  size n, is calculated as follows:  n =
n1+n2+...+nL.


             Although the sample size equations assume that the quantities C^ and P^ are

known, reasonable assumptions can be used, following the rules below (see Box 7.7):

       •      If the relative sampling costs, Cn, are not known or all strata are assumed to
             have the same cost for an additional sample, set Cn =1 for all strata;

       •      If data are not available to provide an estimate of P^ in some strata, set PI, =
             P0 for those strata.
                                     7-15

-------
  CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
              THE SITE IS LESS THAN A CLEANUP STANDARD
             The expected number of contaminated samples in stratum h is Ph* n^. It is
recommended that the expected number of contaminated samples in each stratum be at least
5 for calculation of reliable confidence intervals. Occasionally this may require increasing
the sample size in one or more strata.
                                   Box 7.7
                     Computing the Sample Size for Stratum h
              Given Ch, PH. and WH, the sample size for stratum h should be
    computed as:

                                  W  A/~Tr *•!  l'a	^-f  * ——
                                    hVSj   \P0-P1  j
7.5.2       Calculation of Basic  Statistics

             This section describes the computational procedures to be used to calculate
the quantities necessary to evaluate attainment of the cleanup standard on the basis of the
overall proportion of contaminated samples. Box 7.8 gives sample size calculations for
stratified sampling.

             Use the formula below in Box 7.9 for calculating an overall proportion of
exceedance from a stratified sample. Note that the overall sample proportion, denoted by
pst, is simply a weighted average of the individual stratum means.
                                     7-16

-------
CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
             THE SITE IS LESS THAN A CLEANUP STANDARD
                                 Box 7.8
                 Sample Size Calculations for Stratified Sampling

            At a site with heavy metal contamination, the sample area has been
  divided into two strata, one consisting  of high elevation areas,  another of low
  elevation areas which received most of the historical runoff. The strata are the
  same volume (W^ = .5, W2 = .5) The expected proportion of contaminated soil is
  5 percent on the higher ground and 10 percent in the lower area (Pi = .05, P2 =
  .10).  Due to difficult access and low trafficability in the lower area, the cost of
  sampling is twice  what it is on the high ground (Cj = 1,  €2 = 2).  EPA has
  decided that less than 10 percent of the soil can have concentrations over the
  cleanup standard (with a confidence of  90 percent, a  = .10).  The site manager
  must be able to conclude that the site is clean with a confidence of 80 percent (P =
  .20) at an overall contamination proportion of 4 percent.

            To determine the sample size, the site manager first determines:

                       zla  =  1.282,  ZI_P  =  .842

  from Appendix A.  Then, following equation (7.8):


                      -a+zl-Bl2 _ J1.282+.84212 _
                        -P     ~l -10 - .04  J  -
                         1
                           = {(.5 * VT) + (.5 * VI)} = 1.207
                         J

                                                   f W. ]
                  nhd =  Phd - Ph) * 1-207  * 1,253 * -
            and

               nld  = .05(1 - .05) * 1.207 * 1,253 * -^ = 35.9
                                                  VI


                n2d  = .10(1 - .10) * 1.207 * 1,253 * ~ = 48.1
                                                  V2

            Rounding up, the samples sizes of the strata are:

                               = 36, and n2f = 48
                                   7-17

-------
CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
            THE SITE IS LESS THAN A CLEANUP STANDARD
                                 Box 7.9
              Calculating an Overall Proportion of Exceedances and
            the Standard Error of the Proportion From a Stratified Sample
            Ph = the sample proportion of units in stratum h that have chemical
  concentrations exceeding Cs.
            The estimated overall proportion of soil units that have chemical
   concentrations exceeding Cs is given by the formula below:
                                   IWhPh                  (7.10)
                                   h=l
            Use equation (7.11) to estimate the standard error of p ,.  The
                                                                SI
   standard error is required for constructing an approximate decision rule and also
   for establishing confidence limits around the actual population proportion.
                                   7-18

-------
  CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
              THE SITE IS LESS THAN A CLEANUP STANDARD
             In Box 7.10, use the equation (7.12) to compute the upper limit of the
one-sided  confidence interval.
                                  Box 7.10
                    Calculating the Upper Limit of the One-sided
                Confidence Interval on an Estimate of the Proportion

                             PUa  = Pst + zl-a Spst    (7-12)

    where pst is the computed overall proportion of contaminated units, and  Sp  is the
                                                                  *st
    corresponding standard error. The value of z\.a can be obtained from Table A.2.
The value pUa designates an upper 100(1-a) percent one-sided confidence limit for the

population proportion.
7.5.3        Inference:  Deciding Whether the Site Meets Cleanup Standards


             The upper  one-sided  confidence  limit, P\ja, is  used for testing the
hypothesis that 1 - PQ of the site attains the specified cleanup standard.  Use the following

rules to decide whether or not the site attains the cleanup standard:


             If pua < PO> conclude that the site meets the cleanup standard.


             If puct ^ PO, conclude that the site does not meet the cleanup standard.
                                    7-19

-------
  CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
              THE SITE IS LESS THAN A CLEANUP STANDARD
             See Box 7.11 for an example of an inference for proportions using stratified
sampling.
                                  Box 7.11
                Inference for Proportions Using Stratified Sampling

             Following the example in Box 7.8, all 434 samples from stratum 2
    were collected; however, of the 324 samples in stratum 1, four were lost due to a
    lab error, leaving 320 samples for the analysis.  The proportion of samples
    collected in each strata that had concentrations greater than or equal to the cleanup
    standard are: .0531 in strata 1 (the higher ground) and .0922 in strata 2.

             Using equation (7.10)
                      L
                 Pst = XWh Ph = -5 * -0531 + -5 * .0922 = .0727
                      h=l

             Using equation (7.11)
                            =A/LWh2
                               X h=l
Ph(l - Ph)
             _ / .25 * .0531(1 - .0531)  . .25 * .0922(1 - .0922)   ___.
             ~  \          320          +           454-.0094

             Using equation (7.12)

             PUa = Pst + zl-a Sp = -O727 + 1-282 * .0094 = .0848

             Since  .0848 is  less than P0  (.10),  based  on the  proportion of
    contaminated samples, the sample area attains the cleanup standard.
7.6          Testing Percentiles  from  a Normal  or Lognormal  Population
             Using Tolerance  Intervals


             Tolerance  intervals  assume  that the distribution of concentration
measurements follows a normal distribution. Tolerance interval techniques are sensitive to
the assumption that the data follow a normal distribution.  This procedure is not robust to
departures from the normality assumption.
                                    7-20

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
             If it is suspected that the data do not approximately follow a normal
distribution, then either:
                    •      Do not use the tolerance interval procedure and instead use
                          the nonparametric procedures described in section 7.4; or
                    •      Transform the data so that the transformed data more nearly
                          approximate a normal distribution.

An approach that may be used to evaluate the assumption that the data follows a normal
distribution is discussed in section 7.6.2. If the data are not normal and a transformation is
being used then the transformation should be applied in the following manner. First,
transform both the data and the cleanup standard.  Then calculate the upper confidence limit
on the percentile estimate of the transformed data. Compare the transformed upper limit
with the transformed cleanup standard.  Do not reverse; transform the upper confidence
limit on the percentile for comparison with the untransformed cleanup standard.  If
stratified random sampling is used then consult Mee (1989).
7.6.1        Sample Size Determination

             To determine the required sample size, the following terms need to be
defined, PQ, Pr a, p.  Once these terms have been established, the following are obtained
from Table A.2 and the equation in Box 7.12 is used to estimate the sample sizes:

             zl p    the upper p-percentage point of a z distribution;

             zi-a    *e UPP61" tt-percentage point of a z distribution;
             z1 p    the upper P0-percentage point of a z distribution; and

             z j p    the upper Pj -percentage point of a z distribution.
                                     7-21

-------
 CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTTLE OF
              THE SITE IS LESS THAN A CLEANUP STANDARD
                                  Box 7.12
              Calculating the Sample Size Requirements for Tolerance Intervals
                               (Guttman, 1970)
                                   1-P0   zl-p
                                                          (7,3,
             This sample size equation (7.13) requires smaller sample sizes than the
corresponding formula in section 7.3.1. This happens because the tolerance intervals gain
efficiency over the other methods in this chapter from the assumption that the data follow a
normal distribution.

             If the normal distribution is not followed, even after transformation, the
procedure in this  section is inappropriate.  However, distributional form will not be
evaluated until after the sample is collected and the data analyzed. At this point it may be
decided to use the nonparametric procedures presented earlier in this chapter, but the
sample size may not be sufficient to ensure the desired false negative rate and, therefore,
may not be as sensitive as required.

             Two example sample size calculations for tolerance intervals are shown in
Box 7.13.  The reduction in the required sample size between the nonparametric test and
the tolerance interval test can be compared.  The  comparable sample  sizes for the
nonparametric test are 1990 samples for example #1 and 315 samples for example #2. In
both examples, the tolerance interval method requires fewer samples, provided that it can
be reasonably concluded that the data follow a normal distribution.
                                     7-22

-------
  CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
              THE SITE IS LESS THAN A CLEANUP STANDARD
7.6.2       Testing the Assumption of Normality

            The statistical tests used for evaluating whether or not the data follow a
specified distribution are called goodness-of-fit tests.  There are many different tests and
references demonstrating the evaluation of normality (e.g., Conover, 1980; D'Agostino,
1970; Filliben, 1975; Mage, 1982; and Shapiro and Wilk, 1965). If a choice is available,
the Shapiro-Wilk or the Kolmogorov-Smirnov test with Lilliefors critical values are
suggested.  For easy application, Geary's test described by D'Agostino (1970) can be
used.
                                  Box 7.13
             Calculating Sample Size for Tolerance Intervals—Two Examples
             Following are two examples of the computation required to calculate
    the sample size when testing percentiles using confidence intervals.
             Example #1           PQ=.010   Zj p = z99Q= 2.326
                                  Pj=.005   Zj p = z955 = 2.576
                                   a=.05    z   = z= 1.645
                                                = z 800 = °-842
                      f.842 + 1.64512  f7 48712
                  nd=|2.326-2.576J "^S   = 98'96
Example #2           PQ=. 10    zl p =2^=1 .282
                                              p
                                  P^.05    zlpi = z95= 1.645
                                   a=.05    zla =z95= 1.645
                                   P=.05    z   =z  = 1.645
         1.645 + 1.64515
         1.282- 1.645 J
                                         3.29
                                               = 82.14 = 82
                                    7-23

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
7.6.3       Inference: Deciding Whether the Site Meets  Cleanup Standards
             Using  Tolerance  Limits

             The  test of significance will be performed by estimating the  upper
confidence interval on the point below which at least (1-P0)*100 percent of the data falls:
the [(l-P0)*100]th percentile. For example, the concentration measurement associated with
PO = .05 is the value below which  at least 95 percent of the data falls. The concentration
measurement associated with PQ = .05 will be calculated from  the sample mean  and
standard deviation, x and s, as well as  the constant k. The  constant, k, necessary for
finding the upper tolerance limit, Tu is found using values of a, PQ, n, and T in Table A.3.
For values of k not shown in Table A.3, see Guttman (1970).  With these three quantities
an estimated upper tolerance limit will be calculated for the desired percentile using the
equation in Box 7.14.
                                   Box 7.14
                       Calculating the Upper Tolerance Limit

                                 Tu  = x + ks.                        (7.14)
              If Tu is greater than the cleanup standard, then it is concluded that the
    site fails to meet the cleanup standard.
             Box 7.15 presents data and calculations that illustrate use of tolerance
intervals to test for percentiles with lognormal data.
                                     7-24

-------
CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
             THE SITE IS LESS THAN A CLEANUP STANDARD
                                 Box 7.15
      Tolerance Intervals: Testing for the 95th Percentile with Lognormal Data

            The following data were collected to determine if the 95th percentile of
  the concentrations was below the cleanup  standard of  100 ppm (with a false
  positive rate of 1 percent). The data is assumed to follow a lognormal distribution,
  therefore logarithm of the data (the  transformed data) are analyzed.  In the
  following,  x refers to the original data and y refers to the transformed data.
  Because the log of the data is used, the upper confidence interval on the 95th
  percentile of the  data must be compared to the log of the cleanup standard
  (ln(100)=4.605).  Twenty samples were obtained.
X
34
79
38
62
6
14
20
31
42
36
57
24
57
188
26
45
46
83
25
33
Total
ln(x)=y
3.526
4.369
3.638
4.127
1.792
2.639
2.996
3.434
3.738
3.584
4.043
3.178
4.043
5.236
3.258
3.807
3.829
4.419
3.219
3.497
72.372
V2
12.433
19.088
13.235
17.032
3.211
6.964
8.976
11.792
13.973
12.845
16.346
10.100
16.346
27.416
10.615
14.493
14.661
19.528
10.362
12.229
271.645
  (6.3):
            Using the logarithms as the data to analyze, the sample mean is:
                             _  72.372  _,1Q
                             y =^-=3.619

            The standard deviation, s, can be calculated using equations (6.2) and
 2_  271.645-20(3.619)     .„
s —          , a          = .jil
                                             s = V3TT = 0.715
            For a sample size of 20, a = .01 and P0 = 5 percent, k =  2.808 (from
  Table A.5). Finally, Tu can be calculated using equation (7.14):

                      TU = 3.619 + 2.808(.715) = 5.627
            Since  5.627  is greater than 4.605 (the cleanup standard in logged
  units), the sample area does not attain the cleanup standard.
                                   7-25

-------
  CHAPTER 7:  DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
7.7          Summary

             These methods can apply to the 50th percentile or median as an alternative to
the mean or to a high percentile such as the 90th, 95th, or 99th. High percentile criteria
apply when the clean up standard is viewed as a value that should be rarely exceeded at the
site.  Similar to testing the mean, the proportion of soil samples above the cleanup standard
must be sufficiently low to ensure with confidence that the proportion of soil at the site
meets the established percentile.

             Two approaches to testing whether proportions or percentiles of the soil at a
site are less than the cleanup standard are discussed:

             •      Exact and large sample nonparametric tests for proportions based on
                    the binomial distribution; and
             •      A parametric test for percentiles based on tolerance intervals, which
                    assumes the data have a normal distribution.

             The first approach, or test, works with any concentration distribution and
requires only that the cleanup standard(s) be greater than the analytical method detection
limit.  For testing proportions, simple  random and stratified  random sampling  are
discussed.

             All of the procedures discussed assume that the chemical concentrations can
be measured with little or no error; variability in measurement may bias the corresponding
estimates of proportions. Ways to reduce the potential bias are discussed.

             For simple random samples, the basic steps involved and that are discussed
include the following:
             •      Determine the required sample size;
             •      Identify locations within  the site from which soil units are to be
                    collected, collect the samples, and send them to the laboratory; and
             •      Perform the statistical procedures described in this chapter, and then
                    decide whether the site needs additional cleanup.
                                      7-26

-------
  CHAPTER 7: DETERMINING WHETHER A PROPORTION OR PERCENTILE OF
               THE SITE IS LESS THAN A CLEANUP STANDARD
             The implementation of simple exceedance rules in the statistical plan design
requires that a certain number of  samples be acquired and that none or a few of the
concentration measurements be allowed to  exceed the cleanup standard.  The more
exceedances  allowed, the more soil samples that need  to be collected to maintain the
statistical performance and proportion of the site that is clean. Sample sizes and exceedance
rules as a function of statistical performance criteria are presented in the chapter.

             If stratified sampling is chosen,  the basic steps involved include the
following:
             •      Determine the required sample sizes for each stratum;
             •      Within each  stratum, identify the locations within the site from
                    which the soil units are to be collected, collect the samples, and send
                    them to the laboratory; and
                    Perform the statistical procedures described in this chapter, and then
                    decide whether the site needs additional cleanup.

             The use of tolerance intervals, which is  discussed next in this chapter,
assumes that the distribution of concentration measurements follows a normal distribution.
Techniques for using tolerance intervals, including the transformation of lognormal data to
a normal distribution,  are included with two examples of sample size calculation and other
relevant equations.
                                     7-27

-------

-------
    8.  TESTING PERCENTILES AND PROPORTIONS
                USING SEQUENTIAL  SAMPLING
             This chapter discusses sequential sampling  as a  method for testing
percentiles. With sequential sampling, a statistical test is performed after each sample or
small batch of samples is collected and analyzed. The statistical test determines whether an
additional sample should be collected or whether the sample area is judged to have or have
not attained the cleanup standard.

             Chapters 6 and 7 dealt with statistical tests that are based on samples of a
predetermined  size.   Fixed sample size  methods will sometimes require  that an
unnecessarily large sample size be used in order to meet the stated precision requirements.
This can be avoided by using a sequential procedure.  Sequential procedures terminate
when enough evidence is obtained to either accept or reject the null hypothesis, and thus,
sequential tests can respond quickly to very clean or very contaminated sites. Sequential
procedures will also yield a lower sample size on the average than the fixed sample size
procedure even when the true level of P is not greatly different from PQ.

             Decisions based on sequential sampling methods will be particularly useful
in conjunction with the "rapid turnaround" analytical methodologies that are being used
more often at Superfund  sites. Devices that measure volatile soil gases, H-NU's,  ion
specific probes, or onsite scanning laboratories can  be used much more rapidly and
extensively than conventional intensive  laboratory extraction, identification, and
quantification methods. Without rapid turnaround and the potential for additional sampling
within a day or two, sequential methods are not useful because of the cost to remobilize a
sampling team and the time  required for laboratory processing.  Nevertheless, "rapid
turnaround" technology is typically less accurate than conventional methods and therefore,
despite the larger sample sizes that are possible, should be applied in an orderly and
thoughtful manner.

             References on sequential analysis include: Armitage (1947), Wetherill
(1975), Siegmund (1985), Sirjaev (1973), and Wald (1973).
                                      8-1

-------
CHAPTER 8:  TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
                                   SAMPLING
8.1           Notation Used in This Chapter


              The following notation is used throughout this chapter:

              Cs    The cleanup standard relevant to the sample area and the contaminant
                    being tested (see section 3.4 for more details).

              P     The "true" but unknown proportion of the sample area with
                    contaminant concentrations greater then the cleanup standard.

              PQ    The criterion for defining whether the sample area is clean or dirty.
                    According to the attainment objectives, the sample area attains the
                    cleanup standard if the proportion of the sample area with
                    contaminant concentrations greater than the cleanup standard is less
                    than PQ, i.e., the sample area is clean if P < PQ.

              HO    The null hypothesis, which is assumed to be true in the absence of
                    significant contradictory data. When testing proportions, the null
                    hypothesis is that the sample area does not attain the cleanup
                    standard; HQ:  P > PQ-

              a     The desired false positive rate for the statistical test to be used. The
                    false positive rate for the statistical procedure is the probability that
                    the sample area will be declared to be clean when it is actually dirty.

              HI    The alternative hypothesis, which is declared to be true only if the
                    null hypothesis is shown to be false based on significant
                    contradictory data. When testing proportions, the alternative
                    hypothesis is that the sample area attains the cleanup standard; HI: P
                    
-------
CHAPTER 8: TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
                                   SAMPLING
8.2          Description of  the Sequential  Procedure

              In the sequential testing procedures developed by Wald (1973), sampling is
performed by analyzing one soil unit at a time until enough data have been collected to
either reject the null hypothesis in favor of the alternative hypothesis, or accept the null
hypothesis.1 The expected sample  size,  using this sequential procedure, will  be
approximately 30 to 60 percent lower than the corresponding fixed sample size test with the
same a, (3, PQ, and PI. The sequential procedure will be especially helpful in situations
where contamination at the site is very high or very low. In these situations the sequential
procedure will quickly accumulate enough evidence to conclude that the site either fails to
meet or meets the cleanup standard.  However, it must be emphasized that the actual sample
size of the sequential procedure for a given site could be larger than for the fixed sample
size methods (see section 8.3).

              Wald's  sequential procedure consists of forming an acceptance and rejection
region for the cumulative number of contaminated soil units relative to the total number of
soil units evaluated.  Figure 8.1 shows graphically how the procedure operates. The
horizontal axis, denoted by n, represents the number of soil units evaluated. The vertical
axis represents  the cumulative number of contaminated  soil units after n soil unit
evaluations.  The two lines in the graph establish the boundaries of the acceptance and
rejection regions for the test. The intersection of these lines, CA and CB, with the vertical
axis and their slope, are important parameters of this sequential procedure.

              The sampled soil units are evaluated one at a time, and after each evaluation,
the cumulative number or sum of contaminated units  (i.e., soil units with concentrations
exceeding the cleanup standard, Cs) is determined.  If the  cumulative sum crosses the
topmost line into the acceptance region, the hypothesis  of contamination is accepted.  If the
cumulative sum  stays  low and enters the rejection region  below the second (lowermost)
line, it is concluded  that the  site is  not contaminated (i.e.,  the  null  hypothesis of
contamination is rejected).  Otherwise,  the process continues;  that  is, another soil unit is
evaluated, and the new cumulative sum is compared with the boundary values to determine
whether to accept or reject the null hypothesis  or to  continue evaluating  soil units.  In
lpThe procedure in Wald's book is for a test of PI > PQ. In the present situation this has been reversed. To
 adapt the sequential procedure to this situation, the roles of a and P were reversed. The corresponding
 acceptance and rejection regions of the graphs were also reversed.

                                       8-3

-------
CHAPTER 8: TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
                                  SAMPLING
Figure 8.1 the process terminates after 22 soil units have been evaluated, at which time the
null hypothesis that the area is contaminated is accepted.

             Note that several soil samples can be collected and analyzed at the beginning
of the sequential process, since some minimum number of results must be available before
a decision can be reached
Figure 8.1    Graphic Example of Sequential Testing
        Cumulative
          Sum of
       Concentrations
       Exceeding the
         Cleanup
         Standard
        9  •
        7  •  site is dirty - accept
         5
         3
         1
        -ID
                 -XX XX
             xxxx
           XX
          x   continue sampling
x xxx xx
                         xx
                              site is clean - reject
                   'B
                                         n
                              Number of Soil Units Evaluated
8.3
Sampling  Considerations in Sequential Testing
              It may be impractical to randomly collect a soil unit, chemically analyze the
soil unit, and then decide whether or not to acquire the next unit. Instead, multiple soil
units can be selected initially using the simple random sampling procedures described in
section 5.2. The sampled soil units can then  be chemically analyzed and each result
evaluated individually in random order, until the sequential procedure terminates. It may
also be possible, provided that the holding times or other analytical criteria are not violated,
to chemically analyze samples one at a time.

              In situations where contaminant concentrations at the site are marginally
different from the cleanup standard, the  sequential procedure can be expected to require
more samples until the sample size approaches the sample size required for the fixed sample
procedure. However, this is only an expectation, so in some situations where the actual
                                       8-4

-------
CHAPTER 8:  TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
                                   SAMPLING
contamination is close to the cleanup standard, the sequential procedure can require a
substantially larger sample than the fixed sample procedure.  In this situation, a cutoff rule
is suggested.  If the sequential procedure requires  a sample  size twice the sample size
requked for the fixed procedure, then the sequential sampling should be stopped and a
decision made on the data collected up to that point. Procedures for accommodating this
situation are discussed in Box 8.2.

              Also, as with all of the procedures in  this manual, the site is assumed to be
at steady state during sampling.  During the sequence of sampling the soil concentrations
should not be changing.  Sequential sampling and analysis does not imply that changes
over  time are  being evaluated  or that the progress of cleanup  is being  monitored.
Sequential sampling is performed during steady state conditions, only to reduce the sample
size required for a decision.
8.4          Computational Aspects of Sequential  Testing

              As was the case for the fixed sample tests described in earlier chapters, the
following quantities must be defined  to implement the sequential testing procedure:  Cs,
PO, PI, a, and p.  Box 8.1 describes the method for establishing the acceptance and
rejection boundaries described in Figure 8.1.

              Denote the Qth percentile of chemical concentrations by XQ. To test
whether XQ > Cs or greater (i.e., the site fails to meet the cleanup standard) against the
hypothesis that XQ < Cs (the site meets cleanup standards), set PQ = 1 - Q, and set the
maximum allowable error rate for falsely rejecting that the true percentile is Cs (i.e., false
positive rate) to a.  If the Qth percentile is really less than Cs (indicating that fewer than PQ
of the area is contaminated), specify the minimum value of this percentile, PI < PO, that
should be detected with at least a probability of 1 - (3.

              To test whether the Qth percentile is equal to Cs, the sequential procedure is
formatted by calculating the sequential  procedure  acceptance  and rejection criteria as
described in Box 8.1.  Then follow the steps in Box 8.2 to decide whether the site attains
the cleanup standard.
                                       8-5

-------
CHAPTER 8: TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
                                 SAMPLING
                                   Box 8.1
              Defining Acceptance and Rejection Criteria for the Sequential
                              Tests of Proportions

              Let ln(x) denote the natural logarithm of x. Given a, (3, P0, and P1?
    compute:


              1)   B = ln(— ) and A = ln(-^);



              2)   *i = (\~_ p")   and  R2=pY5

              3)   Use these values computed in (1) and (2) to determine the slope
                  of the two lines defining the rejection and acceptance regions,
                   and the points at which the two lines cross the vertical axis,


              4)   CA = -4-  and  CB -- !U.
              5)  Finally, compute the desired sample size for the corresponding
                  fixed sample size procedure,
                  f                 l-u
                = 1	c	p  . p	J
nd~l—	p  . p
                rO  rl
                                      8-6

-------
CHAPTER 8: TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
                                  SAMPLING
8.5         Inference: Deciding Whether the Site  Meets Cleanup Standards
                                   Box 8.2
                Deciding When the Site Attains the Cleanup Standard

               1)     Calculate the sequential procedure acceptance and rejection
                     criteria described in Box 8.1.

               2)     After each evaluation calculate the cumulative number of soil
                     units that exceed the cleanup standard, Cs:
                                                                  (8.2)
                     where yj = 1 if the i-th  sample was  above the cleanup
                     standard, and yi = 0 otherwise; and where n is the number of
                     soil units evaluated up to  this point.  Compare the current
                     value of k against the current critical value to decide whether to
                     accept or reject the null hypothesis or to continue sampling.

               3)    Starting with n = 1, if k > nM + CA, then stop evaluating
                     samples and accept HO: P > PQ. Conclude that the site is dirty
                     and requires additional cleanup.

               4)    If k < nM + CB, then stop evaluating samples and reject HO in
                     favor of P < PI. Conclude that the site is clean.

               5)    If neither of the two conditions above is met, continue
                     sampling and evaluation.

               6)    If the number of soil units  that has currently been evaluated
                     exceeds 2.0*nf, stop the sampling and:


                     accept HQ: P > PO if k > nM + °A2+CB Or
                     accept HI: P < PI if k < nM +  CA^" °B.
Rule 6 provides an approximate test and will have only a small effect on the actual levels of
a and p (see Wald, 1973).
                                      8-7

-------
CHAPTER 8: TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
                                  SAMPLING
             If the conclusion in step 3 is reached, this means that the cumulative sum
has exceeded the line with intercept CA in Figure 8.1 and the site is judged contaminated.
However, if the conclusion in step 4 is made, then the cumulative sum has fallen below the
line with intercept CB in Figure 8.1 and the site is found to be clean.  Notice that the
intercept values depend on the error rates (a and (3), the proportion that is being tested
(PO), and the proportion where the false negative error rate is estimated (Pi).  The slope of
these lines is determined strictly by PO and PI.

             Box 8.3 presents an example application of sequential testing.


8.6          Grouping Samples in Sequential Analysis

             Under the random sampling  approach discussed in section 7.3, a large
number of soil units are selected from the site at one time, and the laboratory analysis is
conducted on each unit, one at a time.  In many situations it will be more efficient for the
laboratory to analyze the soil units in small batches or groups rather than one at a time.  The
sequential procedure can be modified easily to account for this type of laboratory analysis.

             The quantities Cs, PO, PI, ot, and |3 are defined in exactly the same way as
for stratified sampling. Similarly, the stopping rules are also defined in exactly the same
way. The only modification to the previously discussed procedures is in the calculation of
k. Previously,  after each soil unit was analyzed, k was calculated  as the cumulative
number of soil units that exceeded the cleanup standard, Cs.  To modify k to take into
account the grouped nature of the data, k should be computed as the cumulative number of
soil units that exceed Cs after each batch has been analyzed. This minor modification is
illustrated in Box 8.4 for groups of five, using the example of Box 8.3. In the example,
after 4 groups of 5 or a total of 20 soil units, k = 4 > nM  + CA = 3.0324, so sampling is
terminated and the site is considered contaminated.
                                       8-8

-------
CHAPTER 8: TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
                                 SAMPLING
                                   Box 8.3
                        An Example of Sequential Testing

             Assume that for the chemical under investigation, the following values
    have been specified in the objectives worksheet: a = .05, P = .10,  PO = .05, and
    P! = .02. In this case, the quantities necessary to construct the acceptance and
    rejection regions are:

                 B = ln(-) = -2.8904, and A = InpJ) = 2.2513;
             Rl=||=.9694, and R2=:§=2.5; M =            = .0328;
                     -2 8904                         2 2513
                                       - and  CA -             = 2'3764-
             Below is  a  sequence of outcomes that might  be observed for a
    particular chemical. Note that the values of the boundary limits use the values of
    M, CA, and CB, computed above. In the table, k = the cumulative number of soil
    units that are found to  have excessive levels of the contaminant. The process
    terminates after the 18th soil unit has been analyzed. Prior to the 18th observation,
    the value of k falls  between the computed values of nM+CA  and nM+CB.
    However, with the 18th soil unit, k = 3 > nM+CA = 2.9668, and hence the null
    hypothesis is accepted, i.e., the site fails to meet the cleanup standard.

     Soil    Sample
     unit    outcome         k      nM+CA    nM+CB      Decision
      1         0           0      2.4092     -3.0182      continue
      20           0      2.4420     -2.9854      continue
      30           0      2.4748     -2.9526      continue
      40           0      2.5076     -2.9198      continue
      50           0      2.5404     -2.8870      continue
      60           0      2.5732     -2.8542      continue
      70           0      2.6060     -2.8214      continue
      80           0      2.6388     -2.7886      continue
      90           0      2.6716     -2.7558      continue
     10         0           0      2.7044     -2.7230      continue
     11         1            1      2.7372     -2.6902      continue
     12         0           1      2.7700     -2.6574      continue
     13         0           1      2.8028     -2.6246      continue
     14         0           1      2.8356     -2.5918      continue
     15         0           1      2.8684     -2.5590      continue
     16         1            2      2.9012     -2.5262      continue
     17         0           2      2.9340     -2.4934      continue
     18         1            3      2.9668     -2.4606       accept
                                     8-9

-------
CHAPTER 8: TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
                                SAMPLING



Box 8.4


Example of Sequential Test Using Grouped Samples
Example
Soil
unit
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
using the data

Group
1
1
1
1
1
2
2
2
2
2
3
3
3
3
3
4
4
4
4
4
of Box 8.3
Sample
outcome
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
1
0
1
0
1
after grouping soil units

k nM + CA
2.4092
2.4420
2.4748
2.5076
0 2.5404
2.5732
2.6060
2.6388
2.6716
0 2.7044
2.7372
2.7700
2.8028
2.8356
1 2.8684
2.9012
2.9340
2.9668
2.9996
4 3.0324
into groups

nM + Ce
-3.0182
-2.9854
-2.9526
-2.9198
-2.8870
-2.8542
-2.8214
-2.7886
-2.7558
-2.7230
-2.6902
-2.6574
-2.6246
-2.5918
-2.5590
-2.5262
-2.4934
-2.4606
-2.4278
-2.3950
of 5.

Decision




continue




continue




continue




accept
8.7
Summary
             Sequential sampling means that a statistical test is performed after each
sample or small batch of samples is collected and analyzed.  Sequential testing does not
imply that a time dynamic phenomenon is being monitored. Volume 2, which discusses
ground water, considers sampling and  analysis over time.  Sequential sampling is
performed during steady state conditions and is used only to reduce the sample size
required for a decision.

             Sequential sampling procedures terminate when  enough evidence is
obtained to either accept or reject the null hypothesis.  Thus, sequential tests can respond
                                    8-10

-------
CHAPTER 8: TESTING PERCENTILES AND PROPORTIONS USING SEQUENTIAL
                                 SAMPLING
quickly to very clean or very contaminated sites and in these cases require far less sampling
than the conventional methods discussed in Chapter 7. In situations where contaminant
concentrations at the  site are only marginally different from the cleanup standard, the
sequential procedure can be expected to require more samples until the sample size
approaches the sample size required for the fixed sample procedure.

             The procedure and some computational aspects of sequential testing are
discussed. Sequential sampling and testing are treated separately  from the discussions of
other similar evaluation methods because of the distinct differences in sampling approach.
However,  the chapter makes comparisons with Chapter  7 procedures for sample size
determination.
                                     8-11

-------

-------
                9.   SEARCHING FOR  HOT SPOTS
              As suggested by Barth el M- (1989), it may be desirable to verify cleanups
by documenting that no hot spots could be identified provided that a sampling plan was
used that had an acceptably large probability of finding hot spots. This chapter discusses
how to conduct a valid sampling program to search for hot spots and the conclusions that
can be drawn regarding the presence or absence of hot spots. In general, the methods in
this chapter are presented so they are easy to understand and apply.

              This chapter first describes the literature that discusses methods for locating
hot spots.  This will provide the interested reader with an avenue into discussions regarding
specific applications and details. A simple approach, useful under two different sampling
designs, is summarized. This enables application of selected basic methods without having
to obtain and study the literature.
9.1          Selected  Literature  that Describes Methods for  Locating Hot
              Spots

              Table 9.1 lists several references regarding hot spots and their identification.
Gilbert (1987) offers a general overview of the hot spot searching technique, including
example applications of the simplest methods as  well as more advanced application.
Zirschky and Gilbert (1984) offer applications of these methods at hazardous waste sites.
9.2          Sampling and Analysis  Required to Search for Hot Spots

9.2.1        Basic  Concepts

              The term hot spot is used frequently in discussions regarding the sampling
of hazardous waste sites, yet there is no universal definition of what constitutes a hot spot.
The methods in this chapter model hot spots as localized elliptical areas with concentrations
in excess of the cleanup standard.  Hot spots are generally small relative to the area being
sampled. The hot spot must either be considered a volume defined by the projection of the
surface area through the soil zone that will be sampled or a discrete horizon within the soil
                                       9-1

-------
                  CHAPTER 9:  SEARCHING FOR HOT SPOTS
zone that will be sampled. When a sampl    aken and the concentration of a chemical

exceeds the cleanup standard for that chemk    .t is concluded that the sampling position in
the field was located within a hot spot.
Table 9.1     Selected references regarding the methodologies for identifying hot spots at
             waste sites
       Gilbert, R.O. (1982)               Some Statistical Aspects of Finding Hot
                                        Spots and Buried Radioactivity

       Gilbert, R.O. (1987)               Statistical Methods for Environmental
                                        Pollution Monitoring

       Parkhurst, D.F. (1984)             Optimal Sampling Geometry for Hazardous
                                        Waste Sites

       Singer, D.A. (1972)               Elipgrid: A Fortran IV Program for
                                        Calculating the Probability of Success in
                                        Locating Elliptical Targets with Square,
                                        Rectangular, and Hexagonal Grids

       Singer, D.A. (1975)               Relative Efficiencies of Square and
                                        Triangular Grids in the Search for Elliptically
                                        Shaped Resource Targets

       USEPA   (1985)                   Verification of PCB Spill Cleanup by
                                        Sampling and Analysis

       Zirschky, J. and                   Detecting Hot Spots at Hazardous Waste
       Gilbert, R.O. (1984)               Sites
             Hot spot location techniques involve systematic sampling from a grid of

sampling points arranged in a particular pattern. If a systematic sample is taken and none

of the samples yield concentrations in excess of the cleanup standard, then no hot spots

were found and the site is judged clean.  However, what does this mean in terms of the

chances of contaminant residuals remaining at the site? Since all of the soil could not be

sampled, hot spots could still be present. An important question is: What level of certainty

is there that no hot spots exist at the site?  The answer to this question requires that several

other questions be answered. For example:
                                       9-2

-------
                   CHAPTER 9: SEARCHING FOR HOT SPOTS
              •      What shape hot spot is of concern: circular, fat-elliptical, skinny-
                    elliptical?

              •      What is the length of the longest axis of the hot spot: 1 cm, 10m,
                    or 100 m?
              •      What sampling pattern was used: square, triangular?
              •      What was the distance between sampling points in the grid: 0.1 m,
                    1 m, 100 m?

              If these questions are answered; a sampling plan implemented; and no hot
spots are found, it is possible to conclude with an associated level of confidence that no hot
spots of a certain size are present.  In  general, there is a smaller chance of detecting hot
spots and less confidence in conclusions when:

              •      Hot spot sizes of interest become smaller,
              •      Hot spots are likely to be narrow;
              •      A square rather than a triangular grid is used; and
              •      The spacing between grid points is increased.

Figure 9.1 illustrates a sampling grid with hot spots of various sizes and shapes.  Hot spots
B and D were "hit" with sampling points and hot spots A anc C were missed.

              If one of the samples results in concentrations in excess of the applicable
cleanup standard, a hot spot has been identified.  The conclusion is that the site is not clean.
The normal, reasonable action will be to continue remediation in the areas identified as hot
spots. However, once these locations are remediated, another systematic sample, over the
entire site, with a new random start must be taken in order to conclude with confidence that
no hot  spots of a specified size and shape  are present at  the  site.  Because of this
requirement it may be advisable, after identifying the presence of a single hot spot, to
continue less formal searching followed by treatment throughout the entire sample area.
                                       9-3

-------
                  CHAPTER 9: SEARCHING FOR HOT SPOTS
9.2.2
Choice of a Sampling Plan
             The sampling plan requires no calculations. Instead all the information is
obtained from tables.  Figure 9.2 describes the grid spacing definition for two grid
configurations and how to calculate the parameter for defining the ellipse shape.

             The sampling plan for hot spot detection can be approached in three ways.
The  three factors listed in Table 9.2 control the  performance of a hot spot detection
sampling episode.  Two  of these factors are chosen and fixed.  The third factor is
determined by the choice  of the  first two factors.  Table A. 11  includes information that
allows choice of two factors while providing the resulting third parameter.
Figure 9.1    A Square Grid of Systematically Located Grid Points with Circular and
             Elliptical Hot Spots Superimposed
                                       9-4

-------
                  CHAPTER 9: SEARCHING FOR HOT SPOTS
Figure 9.2    Grid Spacing and Ellipse Shape Definitions for the Hot Spot Search Table
             in Appendix A (Table A. 11)
                   Square
Trianglar
                     G
                           • Sample point location
                           G Grid spacing
             Ellipse Shape
                                  	Long axis
                                      Short axis
                            Length of the long axis = L
                            Length of the short axis = S
                            S/L = Ellipse Shape (ES)
                                     9-5

-------
                  CHAPTER 9: SEARCHING FOR HOT SPOTS
Table 9.2     Factors controlling the design of a hot spot search sampling plan
GRID PATTERN
       Spacing between sample points.
       Geometry of the sample point locations.
HOT SPOT SHAPE
       The length of the long axis of the hot spot.
FALSE POSITIVE RATE
       An acceptable false positive probability; concluding that no hot spots are present
       when there is at least one present.
             Three examples are offered that describe the approaches to sample plan
design.  First, suppose that the size of the hot spot is known or assumed. The shape and
size of the hot spots that are being searched for are elliptical with a long axis of L = 5 m and
a short axis of S = 2 m.  Therefore, the ellipse shape, ES = S/L = 2/5 = 0.4. In addition,
the sampling team decided that they could accept no more than a 10 percent chance of
missing a hot spot if a hot spot was present the false positive rate. A triangular grid pattern
was chosen because the probability of detection was better with an elliptical shaped hot spot
and the sampling team had experience laying out a triangular coordinate system.  The
triangular grid pattern in Table A.I 1 is entered for a value of ES = 0.4 across the top and a
false positive rate of a = .10 or less within  the table.  This corresponds to an L/G value of
0.9, since L = 5, and 0.9 = 5/G, G = 5.55,  or a grid spacing in a triangular pattern of 5.6
m. The density of the grid spacing must be  evaluated with respect to the size of the sample
area.

             Once the grid  spacing density has been determined it is  important  to
estimate for the sample area how many samples would be required given sampling intervals
of 5.6 m on a triangular grid as specified in Figure 9.2.  The following method in Box 9.1
can be used to approximate the sample size necessary when area and grid interval are
known.
                                      9-6

-------
                  CHAPTER 9:  SEARCHING FOR HOT SPOTS
                                   Box 9.1
                         Approximating the Sample Size
                     When Area and Grid Interval Are Known
                                  n = A/G2
             Where:    n = total number of samples required
                       A = size of the area to be sampled (in the same units of
                       measures as G)
                       G = grid spacing as defined in Figure 9.2
             For example, suppose that a lagoon will be sampled that is 45 m by 73 m.
This is a 3285-m2 lagoon. The number of samples required is:

                           3285 m2 / (5.6 m)2 =  104

             On the other hand, a lagoon that is  17 m by 20 m or 340-m2 would require
the following number of samples:

                             340m2/(5.6m)2 =11

             If the size of the  area is relatively small, then the level of confidence
described above may be affordable and acceptable. However, if the area is large and the
number of samples required excessive,  alternatives are available.

             For example, a second approach can be considered that limits the  samples
from the 3285-m2 lagoon.  Suppose that no more than 40 samples are available because of
cost, time, or logistics. The minimum grid spacing is estimated to be:

                              3285m2/G2<40
                                  G>9.1m
                                     9-7

-------
                  CHAPTER 9:  SEARCHING FOR HOT SPOTS
             The question becomes: what probability statement can be made with a 9.1 m
grid spacing searching for the same size hot spot. Review of Table A. 11 indicates that if
L/G = 5/9.1 = .55, and ES = S/L = 2/5 = 0.4 then .33 < a < .63.  Reference to Gilbert
(1987) indicates that a -.55.  This means first that the cost has been reduced by taking 64
fewer samples from the 3285-sq. m lagoon. This was accomplished by increasing the grid
spacing from 5.6 m to 9.1 m. However, the sampling cost reduction increases the chance
of missing contamination.  Specifically, the chance of missing a hot spot and concluding
that the site is clean when a hot spot with an ES of 0.4 and a long axis of 5 m is really
present increases from 10 percent to 55 percent when the sample size is reduced from 104
to 40. If this chance is unacceptably high, there is a third approach.

             The third approach involves fixing the false positive rate, fixing the sample
size or grid spacing, and searching for hot spots that are larger or have a different shape.
Suppose it could be safely assumed that the hot spot of concern was not as elliptically
shaped or as skinny as the ellipse with an ES = 0.4. Instead, the ES = L = 4/5 = 0.8. The
long axis remained at 5 m, but the  short axis doubled from 2 m to 4 m.  For the grid
spacing of G = 5.6 m, the L/G = 5/5.6 = 0.9. From Table A.ll it is clear that the false
positive rate is low, a = .01.  A willingness to search for a larger sized or fatter shaped hot
spot improves the performance of the hot spot search technique from a 10 percent false
positive rate to a less than 1 percent false positive rate with no increase in sample intensity
above 104 samples.
9.2.3        Analysis  Plan

              The analysis is straightforward. Establish a grid of sampling points as
described in Chapter 5 with density and pattern determined using the methods in section
9.2.2 and Figure 9.2.  If one of the chemical measurement results exceeds the cleanup
standard then conclude that a hot spot has been found and the completion of remediation
can not be verified. If none of the samples exceeds the cleanup standard, assume that the
site is clean and conclude with the level of confidence associated with the sampling plan
that it is unlikely a hot spot exists at the site.
                                       9-8

-------
                  CHAPTER 9:  SEARCHING FOR HOT SPOTS
9.3          Summary

             Hot spots are generally defined as relatively small, localized, elliptical areas
with contaminant concentrations in excess of the cleanup standard.  Samples that are taken
and found to exceed the cleanup standard are defined as being located within a hot spot.

             Locating hot spots involves systematic sampling from a grid of sampling
points arranged in a specific pattern. Several questions must be answered to conclude with
a level of confidence that no hot spots of a certain size are present:
             •      What size hot spot is of concern?
             •      What sampling pattern was used?
             •      What was the distance between sampling points in the grid?

             The sampling plan for hot  spot detection is guided by the dimensions and
shape of the grid pattern, the hot spot shape of interest, and the false positive rate. The
information needed is contained in Table A.ll.  Three illustrative examples present
sampling plans for these cases:

             •      The  size of the hot  spot  and false positive rate are known or
                    assumed, and the grid spacing/sample size is determined;
             •      Sample size/grid spacing and ellipse shape  are fixed, and the false
                    positive rate is determined;
             •      The false positive rate and sample size or grid spacing are fixed, and
                    hot spot size is determined.
                                      9-9

-------

-------
   10.   THE USE  OF GEOSTATISTICAL TECHNIQUES
 FOR EVALUATING THE ATTAINMENT OF CLEANUP
                             STANDARDS
             The science of gee-statistics involves the analysis of spatially correlated data.

There are several features of geostatistics that are important to any potential user.

             •     Geostatistical methods provide a powerful and attractive method for
                   mapping  spatial data.   Geostatistical methods provide  for
                   interpolation between existing data points that have been collected in
                   a spatial array and a method for estimating the precision of the
                   interpolation.

             •     Geostatistical methods are complicated mathematically, and the
                   procedures  required to  contour an area cannot be practically
                   implemented by hand and calculator.

             •     New users of geostatistics will need to devote time to understanding
                   the basic approach, concepts and the unique vocabulary associated
                   with geostatistical methods.

             •     To help explore applications, PC-based geostatistical computer
                   software is now readily available to the EPA community (USEPA,
                   1988). However, some preliminary study should be completed, and
                   then the software can be used as an educational and exploratory tool
                   to better understand how geostatistical methods perform.

             This chapter:

             •     Explains fundamental concepts regarding geostatistical methods;

             •     Offers a point of departure into the literature that will provide more
                   details;

             •     Discusses which cleanup scenarios can benefit the most from a
                   geostatistical evaluation;

             •     Describes which geostatistical methods are most appropriate for
                   evaluating the completion of cleanup; and

             •     Lists software available for implementing  geostatistical methods.
                                     10-1

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                THE ATTAINMENT OF CLEANUP STANDARDS
10.1        Background

10.1.1      What Is Geostatistics and How Does  It Operate?

             Many view the science of geostatistics in a broad context as the use of
statistical methods applied to the geographic and geological sciences.  Others refer to
geostatistics as a science that strictly applies to the family of methods that enable the
analysis, evaluation, or characterization of spatially correlated data.  Regardless, kriging
and variogram modeling are primary tools of geostatistical analysis.

             In simple terms, a geostatistical analysis can be viewed as a two-step
process. First, a model is developed that predicts the spatial relationship between a location
where a concentration will be estimated and the existing data obtained from sample points
which are various distances away from the location.  Existing data points nearer to the
location will tend to be closely related and have a large influence on the estimate, and points
far away will tend to be less related and, therefore, impose less influence.  This relationship
function, which describes how influential nearby existing data will be, is modeled and
called a variogram or semi-variogram.

             Figure 10.1 illustrates the general form of a standard or typical variogram
model.  The X or horizontal axis  measures the distance between sample points.  The
vertical or Y axis measures the degree of relationship between points. When there is little
distance between points it is expected that there will be little variability between points. As
the distance  between points increases,  the  difference or variability  between points
increases. The form of this relationship depends on what the variogram modeler knows
about characteristics of the site and the data, and what assumptions are reasonable to make
regarding spatial relationships at the site.

              The second step of the geostatistical analysis is kriging. This involves
estimating chemical concentrations for each point or block in the area of concern. For each
point to be estimated, the surrounding points  provide a weighted contribution to the
estimate. The weightings  are determined by using the variogram model, the location of the
point that is being estimated, and the proximity of other nearby data values, enabling
chemical concentration estimation for locations within the sample area that were not
sampled and therein lies the true value of a geostatistical analysis. In addition to estimates
                                      10-2

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                THE ATTAINMENT OF CLEANUP STANDARDS
of the concentration, kriging allows estimation of the precision associated with the estimate.
If the surrounding data are highly variable, or if the closest data points are relatively far
away, the precision may be low.


Figure  10.1   An Example of an Empirical Variogram and a Spherical Variogram Model
           2.5
             2 -
 Square of the
 Difference   1.5 -
 Between Points
 at Distance h
 Apart         1
           0.5 -
                  •     •
0    10    20    30    40    50    60
                          Distance h
                                                        70    80    90   100
             Kriging provides concentration and associated precision estimates across the
site at all possible points or blocks within the site.  The concentration and precision
estimates can then be graphically contoured across the site. Maps, plotting concentration
isopleths, are the final product. In addition, a precision map that provides isopleths of the
kriging variance or some function of the kriging variance is generated.  These sorts of maps
are illustrated in Flatman and Yfantis (1984) and USEPA (1987b).

             As  a slightly  more technical conclusion to this section, consider the
following discussion of kriging and variogram modeling. Kriging is an interpolation
method based on a weighted moving average where the weights are assigned to samples in
a way that minimizes the variance associated with interpolated estimates. The estimation
variance is computed as  a function of the spatial relationship model known  as the
                                      10-3

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                THE ATTAINMENT OF CLEANUP STANDARDS
variogram, the location of the sampling points relative to each other, and to the location
being estimated (USEPA, 1988).
10.1.2      Introductory Geostatistical  References

             The discussion in section 10.1.1 is intended to provide a simple notion of
how kriging operates.  The next level of understanding requires that the reader consult
specialized literature and a practicing geostatistician.  Several general discussions of
geostatistics are available and are listed in Table 10.1.  In addition to the references in Table
10.1, there is a wide  range of refereed journal literature supporting the theory and
application of geostatistics.  Finally, the  EPA's Environmental Monitoring Systems
Laboratory in Las Vegas, Nevada (EMSL-LV), includes a group of researchers specializing
in the application of geostatistical methods to environmental monitoring problems.  The
group is responsible for the development of the GEOEAS software referenced in Tables
10.1, 10.3, and Box 10.1. In addition, the EMSL-LV has produced refereed literature and
funded university researchers. The researchers operating under cooperative agreement with
the EMSL-LV have produced a series of reports that also provide insights regarding
application of geostatistical methods to environmental problems.
10.2        Soils Remediation Technology and  the Use of  Geostatistical
             Methods
             As recognized in Chapter  1, there  are a variety of soils remediation
methods. Geostatistical methods have many applications, and are especially useful during
remedial investigations where a primary objective is to characterize the extent  of
contamination.   Geostatistical techniques, particularly specialized kriging techniques
referenced in section 10.3, will also be useful for evaluating certain soils remediation
efforts.

             This  section provides guidance  that will help in deciding whether
geostatistical methods are most appropriate for  use under  different  types of  soils
remediation methods. The reader should note that in cases where geostatistical approaches
are not necessarily called for if they are used then the geostatistical approaches will give the
same result as the classical approaches  used throughout the document.  The choice of
                                      10-4

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                THE ATTAINMENT OF CLEANUP STANDARDS
whether to use geostatistical data analysis and evaluation methods depends on the physical
arrangement of the cleanup system, its mode of operation, and the effect that the

remediation technology will have on the soils environment
Table 10.1    Selected introductory and advanced references that introduce and discuss
             geostatistical concepts
INTRODUCTORY

Clark, I.
(1979)

Davis, J.C.
(1986)

USEPA
(1987a)

USEPA
(1987b)
USEPA
(1988)

ADVANCED

Journel, A.G. and
Huijbregts, C.H.
(1978)

David, M.
(1984)
Verly, G.
(1984)
            Practical Geostatistics
            Statistical and Data Analysis in Geology
            Data Quality Objectives for Remedial
            Response Activities: Development Process

            Data Quality Objectives for Remedial
            Response Activities: Example
            Scenario RI/FS Activities at a Site with
            Contaminated Soils and Ground Water

            GEOEAS (Geostatistical Environmental
            Assessment Software) User's Guide
             Mining Geostatistics
             Geostatistical Ore Reserve Estimation
             Geostatistics for Natural Resources
             Characterization
10.2.1
Removal
             Soils remediation may involve either permanent or temporary removal of

soils.  Soils may be permanently transported away from the site.  However, soils may be
                                    10-5

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                THE ATTAINMENT OF CLEANUP STANDARDS
temporarily removed to undergo treatment and then  returned.  In these  situations,
geostatistical methods may be useful for efficiently directing the removal.

             For example, although a single three-dimensional geostatistical study or a
series of two-dimensional geostatistical studies at various depth horizons would have been
preferred during the site characterization phase, this may not have been done.  Therefore,
during removal, as the surface material is skimmed off and new layers are exposed, the
areas of greatest concentration may change.  This changing condition with depth could be
characterized via a geostatistical study.  However, there are practical requirements in this
situation.  In order to be most successful and efficient onsite rapid chemical analysis and
geostatistical data analysis must take place.

             A geostatistical analysis will permit the estimation of concentrations between
the sampled points and allow prediction of which areas should and should not be removed.
As horizons are reached that are below the cleanup standard, they can be avoided.  The
sampling program and data analysis have the ability to operate in a useful and constructive
way that will help direct the cleanup effort and minimize costs.  Indicator and probability
kriging, discussed below, are ideal candidates for evaluating areas that are above and below
cleanup standards.
10.2.2      Treatment Involving Homogenization

             Many soils remediation technologies homogenize the soils media.  This
occurs during soils fixation or chemical modification when soil mixers are used to blend
materials with the soil media. Sampling this type of process could occur at a discharge
point of the mixing apparatus. In this instance, samples may be taken, placed in canisters,
and allowed to solidify or undergo the chemical reaction.  After an established period of
time, the media in the canisters can be extracted and the  leachate concentrations tested
relative to the appropriate cleanup standard.  Samples may also be acquired onsite after the
mixing equipment such as banks of steam injection augers, has passed over each location
that has been pre-selected for sampling to test attainment of the cleanup standard.

             Regardless of how the sampling is conducted, from a statistical perspective,
there are several anticipated results. First, there should be reduction in the variability of
                                      10-6

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                 THE ATTAINMENT OF CLEANUP STANDARDS
chemical contaminants across the site. One way of viewing the effect of treatment is that it
has reduced the magnitude of the large values in the distribution of values at the site. This
can be thought of as "bringing in" the upper tail of the distribution such that the distribution
becomes less lognormal-like and more bell-shaped or normal-like. In practical terms, this
is the same as reducing the variance.  In short, the site should be more homogenous, and
there should be a more random behavior of contaminants across the site.  Finally, the
degree of spatial relationship will be reduced because of the homogeneity.  That is, a point
1m away from a point of concern should be just as similar as a point 50m away.

             Because of these anticipated  results, geostatistical applications are less
useful when remediation results in a homogenization. First, it is likely that the spatial
correlation has been grossly disturbed by the treatment process. Also, sampling may occur
at a discharge point or in association with the operation of a mixing device, rather than in a
spatial framework. If the treatment technology is operating as anticipated, the effectiveness
will be high; the extractable concentrations will be low relative to the cleanup standard and
will have  a  small variance.  Under this scenario, a sampling and analysis program  as
discussed  in Chapters 4-9 can be implemented with a minimum of samples to verify the
effectiveness of treatment rather than require an elaborate geostatistical study.


10.2.3      Flushing

             There is a family of soils remediation techniques that can be thought of as
flushing methods. They rely on surface manifolds attached to extraction wells on one end
and to suction pumps on  the  other end.  These systems can be designed to remove
infiltrated water, artificial liquids, or air. In either case, the liquid or air is the media used
to transport the contaminants. The liquid can flush out soluble contaminants, and the air
can flush out volatile contaminants. Often extraction systems have to contend with both air
and liquid.

             A system of extraction wells, screened at appropriate depths, are installed
across the  contaminated area. Each of the wells is linked by a manifold or piping system,
which is connected to a pump system that provides the vacuum for withdrawal.  The
dynamics of removal  differ depending on many factors including the makeup of the soils
media, the degree of infiltration, the  surrounding ground water system, the type  of
                                      10-7

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                THE ATTAINMENT OF CLEANUP STANDARDS
contaminants, and the media that is being extracted. Regardless of these factors, there is a
tendency with these systems to create zones of influence around each well. Depending on
how long the system has operated and many other factors, the zone of influence will have
much higher or lower concentrations than the surrounding area. The site will then tend to
have a series of zones of influence across the site. Some of the zones will overlap; others
will be irregular in shape because of irregularities in the soils media or the turning on and
off of banks of wells in the system.

             Geostatistical methods are generally not practical for characterizing sites that
have been remediated using flushing technologies because of the highly complex structure
associated with the many overlapping zones of influence around each of the extraction
wells that are distributed across the site.  Although it may be technically possible to
geostatistically model this structure, many samples would be required to provide sufficient
resolution of the many complex gradients across the site.

             However, it may be that by the time verification sampling is conducted the
zones of influence are not likely to be apparent and the site is anticipated to  be uniformly
below the relevant cleanup standard. If extraction has been completed to this point and
there is interest in characterizing the concentration profile across the site, a  geostatistical
study may be warranted. However, the main objective at this stage will normally not be to
characterize the extent of the remaining contaminants that have concentrations below the
cleanup standard, but  instead to simply document that  the site  has met its cleanup
objectives.
10.3        Geostatistical Methods  that Are  Most Useful for Verifying the
             Completion of Cleanup

             As previously described, there are many methods of variogram modeling
and many approaches to kriging. Each technique requires different assumptions or has
advantages in a particular application.  The traditional forms of kriging allow estimates of
central tendency and variance throughout an area.  These forms, which include simple,
ordinary, and universal kriging, require different assumptions regarding the model used to
make the kriging estimates. These types of kriging methods can be used to describe the
                                      10-8

-------
CHAPTER 10:  THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                 THE ATTAINMENT OF CLEANUP STANDARDS
extent of contamination remaining and the precision associated with the concentration
estimates. In this way, the traditional forms of kriging are useful for cleanup verification.

              In addition to the more common methods of kriging described above, there
are several forms of nonparametric kriging, such as indicator and probability kriging, that
have been developed relatively recently and are directly useful for evaluating attainment of
cleanup standards.  These types of kriging are the best forms of kriging for demonstrating
that a particular area is less than a cleanup standard, and unlike the conventional forms of
kriging, these forms are distribution-free.

              Indicator kriging  operates basically by  kriging data that have been
transformed into zeros and ones. For each measurement, the value is transformed to a zero
if the measurement was less than or equal to the cleanup standard, and transformed to a one
if the measurement was greater than the cleanup standard. The data set of zeros and ones is
then used to produce kriging estimates of the probability of exceeding the cleanup standard
across the site.  It then becomes possible to produce a map that contours the probabilities of
having concentrations in excess of the cleanup standard.  Extensions of indicator kriging to
probability kriging allow the development of false positive and false negative error maps.
That is, probability kriging can be used to estimate where there is a chance that an area that
appears to be clean is  actually dirty and where there is a chance that areas that might be
indicated dirty are actually  clean.  Figures 10.2, 10.3,  and 10.4 were adapted from the
probability kriging  study of a lead smelter (Flatman el M-, 1985).

              Although these forms of kriging are directly  applicable to the cleanup
verification problem, they are relatively new methods. Nonparametric and Baysian kriging
are currently an active area of research.  Understanding and application of these kriging
methods will require a substantial investment of time and study. Table 10.2 offers some
initial references.
10.4        Implementation  of Geostatistical  Methods

             As mentioned in  the introduction to this chapter, kriging cannot be
conveniently or practically implemented without a computer and the appropriate software.
Even with the appropriate software, it will take an interested individual a considerable
                                      10-9

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                THE ATTAINMENT OF CLEANUP STANDARDS
investment of time to understand the jargon and mathematics associated with geostatistical
methods.
Table 10.2    Introductory references for indicator, probability, and nonparametric global
             estimation kriging
Buxton, B.E.              Geostatistical Construction of Confidence
(1985)                    Intervals for Global Reserve Estimation

Isaaks, E.H.               Risk Qualified Mappings for Hazardous
(1984)                    Waste Sites: A Case Study in Distribution
                          Free Geostatistics

Journel, A.G.              Nonparametric Estimation of Spatial
(1983)                    Distributions

Sullivan, J.                Conditional Recovery Estimation Through
(1984)                    Probability Kriging
             In many cases, it is best to recognize the power and utility of a geostatistical
study and acquire, or at least have available, the expertise of a geostatistician.  An

alternative is to obtain a first-level familiarity with the methodology and then use a

softwarepackage along with example data sets to explore the practical dynamics and effects
of different modeling decisions.


             The EMSL-LV has recently produced the first version of a geostatistical
software package that provides a convenient environment for exploring the application of

geostatistical methods to hazardous waste site sampling problems (USEPA, 1988).  The

software operates on a PC and is provided in an executable form. It is entirely in the public

domain and can be obtained using the information in Box 10.1.


             The software does not support indicator and probability kriging at this point;

however, as the software undergoes development, it is anticipated that these will be added.


             There are other geostatistical software packages  available in the public

domain that can be purchased.  Table 10.3 lists some examples  and sources of software.
                                     10-10

-------
    CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                 THE ATTAINMENT OF CLEANUP STANDARDS
 Figure 10.2  Contour Map of the Probability in
            Percent of Finding the Value of 1,000
            ppm or a Larger Value
 10000
 7500
 sooo
 2500
                            Figure 10.4  Contour  Map of the Probability in
                                       Percent of a False Negative  in the
                                       Remedial Action Areas and the 1,000
                                       ppm Contour Line
          2500
5000
7500
                                  10000
 Figure 10.3  Contour Map of the Probability in
            Percent of a False  Positive  in the
            Remedial Action Areas and the 1,000
            ppm Contour Line
10OOO
 750O
 5000
 2500
                                            N
                                            I
                                               10000
                                               7500
                              5000
                                               2500
                                                                  l
                                                        2500
                                                                5000
                                                                        7500     1000O
         2500     5000     7500     10000
                                          10-11

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
              THE ATTAINMENT OF CLEANUP STANDARDS
                               Box 10.1
            Steps for Obtaining Geostatistical Software from EMSL-LV

            The software:

               •   Operates on a PC;

               •   Is provided in an executable form;

               •   Is entirely in the public domain; and

               •   Can be obtained by writing to:

                        Evan Englund (GEO-EAS)
                         USEPA, EMSL-LV, BAD
                             P.O. Box 93478
                        Las Vegas, NV 89193-3478

            PLEASE, YOU MUST DO THE FOLLOWING TO OBTAIN
                          THE SOFTWARE!:

              1)  PRE-FORMAT ALL DISKETTES.

             2)  SEND ENOUGH DISKETTES FOR 3 MEGABYTES
                 OF STORAGE AS FOLLOWS:

                       TYPE                    NUMBER
            5 1/4"               1.2MB              3
            51/4"               360KB              9
            3 1/2"               1.44MB              3
            3 1/2"               722KB              6
10.5        Summary


            Geostatistical methods provide a method for mapping spatial data that

enables both interpolation between existing data points and a method for estimating the

precision of the interpolation.


            Geostatistical applications normally involve a two-step process.  First, a

spatial correlation model is  developed that predicts how much spatial relationship exists

among sample points various distances apart
                                 10-12

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                THE ATTAINMENT OF CLEANUP STANDARDS
Table 10.3    Selected geostatistical software
             Program
      Geo-EAS SYSTEM
      (Geostatistical Environ-
         mental Assessment
         Software)
      USGS Statpac Programs
       TOXIPAC
       GEOBASE and GEORES
                Source
See Box 10.1
COGS (Computer Oriented Geological Society)
P.O. Box 1317
Denver, Colorado 80201-1317
Geostat Systems International, Inc.
P.O.Box 1193
Golden, CO  80402
GEOMATH
4860 Ward Road
Wheat Ridge, CO 80033
             The second step, kriging, involves estimating chemical concentrations for
locations within the sample area that were not sampled. For each point to be estimated, the
surrounding points provide a weighted contribution to the estimate based on the variogram
model, the location of the point being estimated, and the proximity of other nearby data
values.  Kriging  also allows estimation of the precision associated  with the estimated
chemical concentrations. Maps that plot concentration isopleths are the final product of the
geostatistical analysis.

             Geostatistical  methods have  many applications in soil remediation
technology, especially when the extent of contamination needs to be characterized. This
chapter includes guidance to help decide whether geostatistical data analysis and evaluation
methods are appropriate for use with three types  of soils remediation activities:  removal,
treatment involving renamed homogenization, and flushing.

             Of the many methods of variogram modeling and many approaches  to
kriging, each requires different assumptions or has advantages in certain applications. The
                                     10-13

-------
CHAPTER 10: THE USE OF GEOSTATISTICAL TECHNIQUES FOR EVALUATING
                THE ATTAINMENT OF CLEANUP STANDARDS
traditional forms of kriging, including simple, ordinary, and universal, are primarily useful
for characterization but may also  be  used for cleanup verification. Nonparametric,
indicator, and probability kriging are the  best forms for demonstrating probabilistically that
an area is less than a cleanup standard and, unlike the traditional forms, are distribution-
free.

             Geostatistical techniques  referred to in the chapter will need in-depth study
by the intended user before being applied.  References are provided to help familiarize the
interested reader.  Because  kriging  cannot be conveniently or practically implemented
without a computer  and the appropriate software, a first-level familiarity with the
methodology along with use of a software package is a practical way of exploring example
applications and data sets. EPA has developed the first version of a geostatistical software
for the novice, available by following instructions at the end of this chapter.
                                      10-14

-------
                          BIBLIOGRAPHY
Armitage, P. 1947. "Some Sequential Tests of Student's Hypothesis." Journal of the
         Royal Statistical Society.  Series B, Vol. 9 : 250-263.

Barich, J. Memo to B. Johnson on April 21, 1988.

Barnett, V., and T. Lewis. 1978. "Outliers in Statistical Data." John Wiley & Sons,
          Bath, England.

Barth, D. S., Mason, B. J., Starks, T. H., and K. W. Brown.  1989.  Soil
          Sampling Quality Assurance User's Guide, Second Edition, USEPA
          Cooperative Agreement No. CR 814701, Environmental Monitoring Systems
          Laboratory, Las Vegas, Nevada.

Bavinck, H. F.,  Roels, J. M., and J. J. Vegter. 1988. The Importance of
          Measurement Procedures in Curative and Preventive Soil Protection."  In K.
          Wolf, W. J. van den Brink, L. F. J. Colon (eds.) Contaminated Soil '88, 125-
          133, Kluwer Academic Publishers.

Beckett, P. H. T., and R. Webster. 1979. "Soil Variability: A Review."  Soils and
          Fertilizers. Vol. 34, No.  1: 1-15.

Benson, R. C., Glaccum, R. A., and M. R. Noel.  1984.  Geophysical
          Techniques for Sensing Buried Wastes and Waste Migration. Miami, Fla.:
          Technos, [June 1984].

Bisgaard, S., and W. G. Hunter. 1986. Report No. 7, Studies in Quality
          Improvement: Designing Environmental Regulations, Center for Quality and
          Productivity Improvement, University of Wisconsin-Madison,  [February
          1986].

Bishop, T. A.  1985. Statistical View of Detection and Determination Limits in Chemical
          Analyses Prepared for the Committee on Applications of Statistical Techniques
          in Chemical Problems.  Columbus, Ohio: Battelle Columbus Laboratories,
          [January 5, 1982].

Box, G. E. P., Hunter  W. G., and J. S.  Hunter. 1987. Statistics For
          Experimenters. New York: John Wiley.

Bross, I. D. 1985. "Why Proof of Safety is Much More  Difficult Than  Proof of
          Hazard." Biometrics.  Vol. 41: 785-793.

Brown, K. W., and S. C. Black. 1983. "Quality Assurance and Quality Control
          Data Validation Procedures Used for the Love Canal and Dallas Lead Soil
          Monitoring Programs." Environmental Monitoring and Assessment.  Vol. 3:
          113-122.
                                    Bro-i

-------
                               BIBLIOGRAPHY
Brown, G. H., and N. I. Fisher. 1972. "Subsampling a Mixture of Sampled
          Material." Technometrics, Vol. 14, No. 3: 663-668.

Brown, M. B., and R. A. Wolfe. 1983. "Estimation of the Variance of Percentile
          Estimates."  Computational Statistics and Data Analysis. Vol. 1: 167-174.

Brownlee, K. A. 1965. Statistical Theory and Methodology in Science and
          Engineering, 2nd. New York: John Wiley & Sons, Inc.

Burgess, T. M., and R. Webster. 1980a.  "Optimal Interpolation and Isarithmic
          Mapping of Soil Properties: I. The Semi-Variogram and Punctual Kriging."
          Journal of Soil Science, Vol. 31: 315-331.

Burgess, T. M., and R. Webster. 1980b.  "Optimal  Interpolation and Isarithmic
          Mapping of Soil Properties: II. Block Kriging." Journal of Soil Science.  Vol.
          31: 333-341.

Buxton, B. E.  1985. Geostatistical Construction of Confidence Intevals for Global
          Reserve  Estimation, Unpublished Master's Thesis, Stanford University.

Campbell, J. B.  1979. "Spatial Variability of Soils."  Annals of the Association of
          American Geographers.  Vol. 69, No. 4: 544-556.

Casey, D., Nemetz, P. N., and D.  Uyeno. 1985. "Efficient Search Procedures for
          Extreme Pollutant Values." Environmental Monitoring and Assessment.  Vol.
          5: 165-176.

Clark, I. 1979. Practical Geostatistics. New York: Elsevier Science.

Clayton, C. A., Hines, J. W.,  and P. D. Elkins. 1987. Detection Limits  with
          Specified Assurance Probabilities. Analytical Chemistry. Vol. 59, No. 20:
          2506-2514.

Cline, M. G. 1944. "Principles of Soil Sampling."  Soil Science.  Vol. 58: 275-288.

Cochran, W. 1977. Sampling Techniques.  New York: John Wiley.

Cohen, A. C. 1961. "Tables for Maximum Likelihood Estimates: Singly Truncated and
          Singly Censored Samples." Technometrics. Vol. 3, No. 4: 535-541.

Conover, W. J. 1980. Practical Nonparametric Statistics. New York: John Wiley.

D'Agostino, R. B.  1970.  "A Simple Portable Test of Normality: Geary's Test
          Revisited." Psychological Bulletin.  Vol.. 74, No. 2: 138-140.

David, M. 1984. Geostatistical Ore Reserve Estimation.

Davis, J. C. 1986.  Statistical Data Analysis in Geology.  John Wiley and Sons, New
          York.
                                     BIB-2

-------
                              BIBLIOGRAPHY
Doctor, P. G., and R. O. Gilbert. 1978. "Two Studies in Variability for Soil
         Concentrations: With Aliquot Size and with Distance." Selected Environmental
         Plutonium Reports of the NAEG. Richland, Wash.: Battelle Pacific Northwest
         Laboratory, [June 1978].

Dowd, P. A. 1982. "Lognormal Kriging - The General Case." Mathematical Geology.
         Vol. 14, No. 5: 475-499.

DuMouchel, W. H., Govindarajulu,  Z., and  E. Rothman. 1973. "Note on
         Estimating the Variance of the Sample Mean in Stratified Sampling." The
         Canadian Journal of Statistics. Vol. 1, No. 2: 267-274.

Duncan, A. J. 1962. "Bulk Sampling: Problems and Lines of Attack." Technometrics.
         Vol. 4, No. 2: 319-344.

Elder, R. S., Thompson, W.  O., and R.  H. Myers.  1980. "Properties of
         Composite Sampling Procedures." Technometrics. Vol. 22, No. 2:  179-186.

Esposito, M. P.,  McArdle, J.  L., Crone, A.  H., Greber, J. S.,  Clark, R.,
         Brown,  S., Hallowell, J.  B., Langhant, A.,  and  C. D.
         McCandlish. 1985. Guide for Decontaminating Buildings, Structures, and
         Equipment at Superfund Sites. Columbus, Ohio: Battelle Columbus
         Laboratories, [March 1985].

Exner, J. H., Keffer, W. D., Gilbert, R. O.,  and R. R.  Kinnison.  1985.
         "Sampling Strategy for Remedial Action at Hazardous Waste Sites: Cleanup of
         Soil Contaminated by Tetrachlorodibenzo-P-Dioxin." Hazardous Waste and
         Hazardous Materials. Vol. 2, No. 4: 503-521.

Filliben, J. J. 1975.  "Probability Plot Correlation Coefficient Test for Normality."
         Technometrics. Vol. 17, No.  1:  111-117.

Finney, D. J. 1948.  "Random and Systematic Sampling in Timber Surveys." Forestry.
         Vol. 22: 64-99.

Flatman, G. T., Brown, K. T., and J. W. Mullins. 1985.  "Probabilistic Spatial
         Contouring of the Plume Around a Lead Smelter." The 6th National
         Conference on Management of Uncontrolled Hazardous Waste Sites,
         Superfund '85 Proceedings, Hazardous Material Control Research Institute:
         442-448.

Flatman, G. T., and A. A. Yfantis. 1984.  "Geostatistical Strategy for Soil
         Sampling: The Survey and the Census." Environmental Monitoring and
         Assessment. Vol. 4: 335-349.

Ford, P., and P. Turina.  1985. Characterization of Hazardous Waste Sites—A
         Methods Manual, Volume I—Site Investigations.  Las Vegas, Nevada: EPA
         Environmental Monitoring Systems Laboratory, [April 1985].

Garner, F. C. 1985. Comprehensive Scheme for Auditing Contract Laboratory Data
         [interim report]. Las Vegas, Nevada: Lockheed-EMSCO.
                                    BIB-3

-------
                               BIBLIOGRAPHY
Gastwirth, J. L., and H. Rubin. 1971. "Effect of Dependence on the Level of Some
          One-Sample Tests." Journal of the American Statistical Association. Vol.
          66: 816-820.

Gilbert, R. O. 1982. Tran-Stat: Statistics for Environmental Studies, No. 19, Some
          Statistical Aspects of Finding Hot Spots and Buried Radioactivity. Richland,
          Wash.: Pacific Northwest Laboratory.

Gilbert, R. O. 1983. Tran-Stat: Statistics for Environmental Studies, No. 24, Field
          Sampling Designs, Simple Random and Stratified Random Sampling.
          Richland, Wash.: Battelle Pacific Northwest Labs.

Gilbert, R. O. 1984. Tran-Stat: Statistics for Environmental Studies, No. 26, Field
          Sampling Designs: Systematic Sampling.  Richland, Wa.: Battelle Memorial
          Institute, Pacific Northwest Laboratory.

Gilbert, R. O. 1987. Statistical Methods for Environmental Pollution Monitoring. New
          York: Van Nostrand Reinhold.

Gilbert, R. O., and P. G. Doctor. 1985. "Determining the Number  and Size of Soil
          Aliquots for Assessing Particulate  Contaminant Concentrations." Journal of
          Environmental Quality. Vol. 14: 286-292.

Gilbert, R. O., and R. R. Kinnison. 1981. "Statistical Methods for Estimating the
          Mean and Variance from Radionuclide Data Sets Containing Negative,
          Unreported or Less-Than Values." Health Physics.  Vol. 40: 377-390.

Gilbert,  R. O., Miller, M.  L, and H.  R.  Meyer.  1989. "On the Design of a
          Sampling Plan to Verify Compliance with EPA Standards for Radium-226 in
          Soil at Uranium Mill Tailings Remedial-Action Sites." ASA/EPA Conference
          Proceedings on Interpretation of Environmental Data. IV Compliance
          Sampling, October 5-6,1987 (EPA-230-03-047).

Gilliom, R. J., and D. R. Helsel. 1986.  "Estimation of Distributional Parameters
          for Censored Trace Level Water Quality Data, 1. Estimation Techniques."
          Water Resources Research. Vol.  22, No. 2: 135-146.

Gleit, A. 1985. "Estimation for Small Normal Data Sets with Detection Limits."
          Environmental Science and Technology. Vol. 19, No. 12: 1201-1206.

Grubbs, F. E. 1969. "Procedures for Detecting Outlying Observations in Samples."
          Technometrics. Vol. 11, No. 1: 1-21.

Guttman, I. 1970. Statistical Tolerance Regions: Classical and Bayesian. No. 26
          Griffin's Statistical Monographs and Courses. Alan Stuart, ed, Darien, Conn.:
          Hafner Publishing.

Hammond, L. C., Pritchett, W. L., and V. Chew. 1958. "Soil Sampling in
          Relation to Soil Heterogeneity." Soil Science Society of America Proceedings.
          Vol. 22: 548-552.
                                     BIB-4

-------
                              BIBLIOGRAPHY
Hansen, M., Hurwitz, W, and W. Madow.  1953. Sample Survey Methods and
          Theory, Volume 1.  New York: John Wiley.

Hazardous Materials Control Research Institute. 1985. 6th National Conference
          on Management of Uncontrolled Hazardous Waste Sites.  Washington, D.C.:
          HMCRI, [November 4-6, 1985].

Hazardous Materials Control Research Institute. 1986. 7th National Conference
          on Management of Uncontrolled Hazardous Waste Sites.  Washington, D.C.:
          HMCRI, [December 1-3, 1986].

Helsel, D. R., and R. J. Gilliom. 1986. "Estimation of Distributional Parameters for
          Censored Trace Level Water Quality Data, 2. Verification and Applications."
          Water Resources Research. Vol. 22, No. 2: 147-155.

Hoaglin, D. C., Mosteller, F., and J. W. Tukey. 1983. Understanding Robust
          and Exploratory Data Analysis. New York: John Wiley.

Hughes, J. P., and D. P. Lettenmaier. 1981. "Data Requirements for Kriging:
          Estimation and Network Design." Water Resources Research. Vol. 17, No. 6:
          1641-1650.

Isaaks, E. H.  1984. Risk Qualified Mappings for Hazardous Waste Sites: A Case Study
          in Distribution Free Geostatistics. Master's Thesis.  Stanford University.

Journel, A. G. 1983. "Nonparametric estimation of spatial distributions."
          Mathematical Geology. Vol. 15, No. 3:  445-468.

Journel, A. G. 1984. Indicator Approach to Toxic Chemical Sites [EPA Project No.
          CR-811235-02-0-1984]. Stanford, Ca.: Stanford University.

Journel, A. G., and C. J. Huijbregts. 1978. Mining Geostatistics. New York:
          Academic Press, London.

Kish, L.  1965. Survey Sampling.  New York: John Wiley.

Koop, J.  C. 1976. Systematic Sampling of Two-Dimensional Surfaces and Related
          Problems. Research Triangle Park, N.C.: Research Triangle Institute, [May
          1976].

Krige, D. G., and E. J. Magri. 1982. "Geostatistical Case Studies of the Advantages
          of Lognormal-De Wijsian Kriging With Mean for a Base Metal Mine and a Gold
          Mine." Mathematical Geology. Vol. 14, No. 6:547-555.

Land, C. E. 1971. "Confidence Intervals for Linear Functions of the Normal Mean and
          Variance." Annals of Mathematical Statistics. Vol. 42, No.4: 1187-1205.

Land, C. E. 1975. Tables of Confidence for Linear Functions of the Normal Mean and
          Variance. Selected Tables in Mathematical Statistics,  Vol. III. pp.385-419.
          Providence, R.I.:  American Mathematical Society.
                                    BIB-5

-------
                              BIBLIOGRAPHY
Legg, D. E.,  Shufran, K. A., and K. V. Yeargan. 1985. "Evaluation of Two
          Sampling Methods for Their Influence on the Population Statistics of Alfalfa
          Weevil (Coleoptera: Curculionidae) Larva Infestations in Alfalfa." Journal of
          Economic Entomology.  Vol. 78, No. 6: 1468-1474.

Lehmann, E.  L. 1975. Nonparametrics: Statistical Methods Based on Ranks.  San
          Francisco: Holden-Day.

Madow, W. G., and L. H. Madow. 1944. "On the Theory of Systematic
          Sampling." Annals of Mathematical Statistics. Vol. 15: 1-24.

Mage, D. T. 1982. "Objective Graphical Method for Testing Normal Distributional
          Assumptions Using Probability Plots." American  Statistician. Vol. 36, No. 2:
          116-120.

Mason, B. J.  1983. Preparation of Soil Sampling Protocol: Techniques and Strategies.
          Las Vegas, Nevada: Environmental Research Center, UNLV, [August 1983].

McBratney, A. B., and R. Webster. 1981. "Design of Optimal Sampling Schemes
          for Local Estimation and Mapping of Regionalized Variables - II. Program and
          Examples." Computers and Geosciences.  Vol. 7, No. 4: 335-365.

McBratney, A. B. Webster, R., and T. M.  Burgess. 1981. "Design  of Optimal
          Sampling Schemes for Local Estimation and Mapping of Regionalized Variables
          -1. Theory and Method." Computers and Geosciences.  Vol. 7, No. 4: 331-
          334.

Mee, R. W. 1984. "Tolerance Limits and Bounds for Proportions Based on Data
          Subject to Measurement Error."  Journal of Quality Technology. Vol. 16, No.
          2: 74-80.

Mee, R. W. 1989. "Normal Distribution Tolerance Limits for Stratified Random
          Samples." Technometrics. Vol. 31, No. 1: 99-105.

Mee, R. W.,  Owen, D. B., and  J. C. Shyu. 1986. "Confidence  Bounds  for
          Misclassification Probabilities Based on Data Subject to Measurement Error."
          Journal of Quality Technology. Vol. 18, No. 1: 29-40.

Mendenhall,  W., and L. Ott. 1980. Understanding Statistics. N. Scituate, Mass.:
          Duxbury Press.

Milne, A. 1959. "Centric Systematic Area-Sample Treated as a Random Sample."
          Biometrics. Vol. 15, No. 2: 270-297.

Natrella, M.  1963. Experimental Statistics. Washington, D.C.: U.S. Department of
          Commerce, National Bureau of Standards.

Neter, J., Wasserman, W., and G. Whitmore. 1982. Applied Statistics. Boston:
          Allyn-Bacon.
                                    BIB-6

-------
                              BIBLIOGRAPHY
Osborne, J. G. 1942. "Sampling Errors of Systematic and Random Surveys of Cover-
         Type Areas." Journal of the American Statistical Association. Vol. 37: 256-
         264.

Owen, D. B. 1963. Factors for One-Sided Tolerance Limits and for Variables Sampling
         Plans.  Albuquerque, N.M.: Sandia Corporation.

Palley, M. N., and L. G. Horwitz. 1961. "Properties of Some Random and
         Systematic Point Sampling Estimators." Forest Science, Vol. 77: 52-65.

Parkhurst, D. F. 1984. "Optimal Sampling Geometry for Hazardous Waste Sites."
         Environmental Science and Technology. Vol. 18, No. 7: 521-523.

Patel, J. K. 1986. "Tolerance Limits- A Review." Communications 5.   vies: Theory
         and Methods. Vol. 15, No. 9: 2719-2762.

Peshkova, N. V. 1970. "Comparison of Results of Determinations of  e Density of
         Plant Populations by Systematic and by Restricted Random Sampling
         Procedures." Ecology (Ekologiya). Vol.2: 180-181.

Phillip, G. M, and D. F. Watson.  1986. "Matheronian Geostatistics - Quo Vadis?"
         Mathematical Geology. Vol. 18, No. 1: 93-117.

Puente, C. E., and R. L. Bras. 1986. "Disjunctive Kriging, Universal Kriging, or
         No Kriging: Small Sample Results with Simulated Fields." Mathematical
         Geology, Vol. 18, No. 3: 287-305.

Rao,  P. V., Rao, P. S.  C.,  Davidson, J. M., and  L.  C. Hammond.  1979.
         "Use of Goodness-of-Fit Tests for Characterizing the Spatial Variability of Soil
         Properties." Soil Scientist Society  of America Journal. Vol. 43, No. 2: 274-
         278.

Rao, R. G. S., and F. T. Ulaby. 1977. "Optimal Spatial Sampling Techniques for
         Ground Truth Data in Microwave Remote Sensing of Soil Moisture." Remote
         Sensing of Environment.  Vol. 6, No. 4: 289-301.

Rendu, J. M. 1979. "Normal and Lognormal  Estimation." Mathematical Geology.  Vol.
          11, No. 4:407-422.

Ripley, B. D. 1981. Spatial Statistics.  New  York: John Wiley.

Roels, J. M., and P. J. Jonker. 1983. "Probability Sampling Techniques for
          Estimating Soil Erosion." Soil Science Society of America Journal.  Vol. 47:
          1224-1228.

Rohde, C. A. 1976. "Composite Sampling."  Biometrics. Vol. 32: 273-282.

Rohde, C. A. 1979. Batch, Bulk, and Composite Sampling. Sampling Biological
          Populations, pp. 365-367. Edited  by R.M. Cormack.  Fairland, Md.:
          International Cooperative Pub. House.
                                    BIB-7

-------
                              BIBLIOGRAPHY
Schaeffer,  D. J., Derster, H. W., and D. G. Janardan.  1980.  Grab versus
         composite sampling: A primer for managers and engineers. /. Environ. Mgmt.
         Vol.4: 157-163.

Schaeffer, D.J.,  and D. G. Janardan. 1978. Theoretical comparison of grab and
         composite sampling programs. BiomJ.  Vol. 20: 215-227.

Scheaffer,  R. L., Mendenhall, W., and L. Ott. 1979. Elementary Survey
         Sampling, Second Edition.  Boston: Duxbury Press.

Schmid, C. F. 1983. Statistical Graphics: Design Principles and Practices, John Wiley
         and Sons, Inc.

Schwartz, J. E. ;   5. "Neglected Problem of Measurement Error in Categorical Data."
         Socioi  cal Methods and Research. Vol. 13, No. 4: 435-466.

Schweitzer, G.  H., and J. A. Santolucito, editors.  1984.  Environmental
         Sampling for Hazardous Wastes. ACS Symposium Series 267. Washington,
         D.C.: American Chemical Society.

Shapiro, S. S., and M. B. Wilk. 1965. "Analysis of Variance Test for Normality
         (Complete Samples)."  Biometrika.  Vol. 52: 591-611.

Sharpe, K.  1970. "Robustness of Normal Tolerance Intervals." Biometrika.  Vol. 57,
         No. 1: 71-78.

Sigmund, D.  1985.  Sequential Analysis: Tests and Confidence Intervals. New York:
         Springer-Verlag.

Singer, D. A. 1972. "Elipgrid. A Fortran IV Program for Calculating the Probability of
         Success in Locating Elliptical Targets With Square, Rectangular, and
         Hexagonal Grids." Geological Programs. Vol. 4: 1-16.

Singer, D. A. 1975. Relative Efficiencies of Square and Triangular Grids in the Search
         for Elliptically Shaped Resource Targets. Jour Research U. S. Geological
         Survey.  Vol. 3,  No. 2: 163-167.

Sirjaev, A.  N. 1973. Statistical Sequential Analysis. Providence, R.I.: American
         Mathematical Society.

Size, W. B., editor. 1987. Use and Abuse of Statistical Methods in the Earth Sciences.
         New York: Oxford University Press.

Snedecor, G. W., and W. G. Cochran. 1980. Statistical Methods. Seventh Edition
         Ames  Iowa:  The Iowa State Press.

Sokal, R. R., and F. J. Rohlf. 1973. Introduction to Biostatistics.  San Francisco:
         W. H. Freeman.

Sokal, R. R., and F. J. Rohlf. 1981. Biometry: The Principles and Practice of
         Statistics in Biological Research. Second Edition. New York: W. H. Freeman.
                                    Bffi-8

-------
                              BIBLIOGRAPHY
Starks, T. H. 1986. "Determination of Support in Soil Sampling." Mathematical
          Geology. Vol. 18, No.6: 529-537.

Stellingwerf, D. A., and S. Lwin. 1985. "Stratified Sampling Compared with Two-
          Phase Stratified Cluster Sampling for Timber Volume Estimation." Netherlands
          Journal of Agricultural Science. Vol. 33, No. 2: 151-160.

Sullivan, J. 1984. Conditional Recovery Estimation Through Probability Kriging.
          Doctoral Thesis, Stanford University.

Swinger, D. A.  1972. Elipgrid, A Fortran IV Program for Calculating the Probability of
          Success in Locating Elliptical Targets with Square, Rectangular, and Hexogonal
          Grids. Geostatistical Programs. Vol. 4:1-16.

Switzer, P. 1983. When Will a Pollutant Standard Be Exceeded: Model Prediction and
          Uncertainty.  Technical Report No. 67. New Canaan, Ct.: SIMS, [January
          1983].

Taylor, J. A.,  Friend, G.  R., and M. L. Dudzinski. 1984. "Influence of
          Sampling Strategy on the Relationships Between Fauna and Vegetation
          Structure, Plant Lifeform and Floristics." Australian Journal of Ecology. Vol.
          9: 281-287.

Thomas, V. W., and R. R. Kinnison. 1983.  Recommended Sampling Strategies
          for Spatial Evaluation of Windblown Contamination Around Uranium Tailings
          Piles. Richland, Wa.: Battelle Pacific Northwest Labs, [October 1983],

Tornqvist, L. 1963. "Theory of Replicated Systematic Cluster Sampling With Random
          Start." Review of the International Statistical Institute. Vol. 31, No. 1: 11-23.

Tukey, J. W. 1977. Exploratory Data Analysis.  Reading, Mass.:  Addison-Wesley.

U.S. Environmental Protection Agency. 1983. Preparation of  Soil  Sampling
          Protocol: Techniques and Strategies,  Washington D.C., August 1983 (EPA
          600/4-83-020).

U.S. Environmental Protection Agency. 1984. Soil Sampling Quality Assurance
          User's Guide, First Edition, Washington, D.C., May 1984 (EPA 600/4-84-
          043).

U.S. Environmental Protection  Agency. 1985. Verification of  PCB Spill Cleanup
          by Sampling and Analysis, Washington D.C., August 1985 (EPA 560/5-85-
          026).

U.S. Environmental Protection Agency. 1986a.  A Primer on  Kriging.
          Washington, D.C.

U.S. Environmental Protection Agency. 1986b. Guidance Document for Cleanup
          of Surface Impoundment Sites,  Washington D.C., June 1986 (OSWER
          Directive 9380.0-6).
                                    BIB-9

-------
                              BIBLIOGRAPHY
U.S. Environmental Protection Agency. 1986c. Guidance on Remedial Actions for
         Contaminated Ground Water at Superfund Sites [draft] Washington D.C.,
         October 1986.

U.S. Environmental  Protection Agency. 1986d. Superfund Public Health
         Evaluation Manual, Washington D.C.: EPA [October 1986].

U.S. Environmental Protection Agency. 1987a. Data Quality Objectives For
         Remedial Response Activities: Development Process, Washington D.C.,
         March 1987 (EPA 540/G-87/003).

U.S. Environmental Protection Agency. 1987b.  Data Quality Objectives For
         Remedial Response Activities: Example Senarip RI/FS Activities at a Site with
         Contaminated Soils and Ground Water, Washington D.C., March 1987 (EPA
         540/G-87/004).

U.S. Environmental Protection Agency. 1987c. EPA Journal, The New
         Superfund: Protecting People and the Environment,  Washington D.C., Vol.
         13, No. 1.

U.S. Environmental  Protection Agency. 1987d. Surface Impoundment Clean
         Closure Guidance Manual [draft], Washington D.C., March 1987, (OSWER
         Directive 9476.0-8.C).

U.S. Environmental  Protection Agency. 1987e.  Test Methods for Evaluating Solid
         Waste,  Washington D.C., November 1987 (SW-846).

U.S. Environmental  Protection Agency. 1988. GEOEAS  (Geostatistical
         Environmental Assessment Software User's Guide, EMSL-Las Vegas, NV,
         (EPA 600/4-88/033).

Verhagen, E. J. H. 1988. "Review of Thermal and Extraction Soil Treatment Plants in
         the Netherlands," National Institute of Public Health and Environmental
         Protection (RTVM) Laboratory for Waste and Emission Research (LAE), In K.
         Wolf, W. J. vanden Brink, F. J. Colon (eds.), Contaminated Soil '88, 797-
         808, Kluwer Academic Publishers.

Verly, G.,  David, M.,  Journel, A. G., and A.  Marachal, editors.  1984.
         Geostatistics for Natural Resources Characterization, Parts 1 and 2, D. Reidel
         Publishing Company.

Wald, A. 1973. Sequential Analysis.  New York: Dover Publications.

Watson, G. S.  1984. "Smoothing and Interpolation by Kriging and with  Splines."
         Mathematical Geology. Vol. 16, No. 6: 601-615.

Webster, R., and T. M. Burgess. 1980. "Optimal Interpolation and Isarithmic
         Mapping of Soil Properties: in. Changing Drift and Universal Kriging."
         Journal of Soil Science. Vol. 31: 505-524.

Wesolowsky, G. O. 1976. Multiple Regression and Analysis of Variance.  New York:
         John Wiley.
                                   BIB-10

-------
                              BIBLIOGRAPHY
Wetherill, G. B. 1975. Sequential Methods in Statistics.  New York: Halsted Press.

Willits, N. H. 1982. Compliance Testing for Statistical Air Quality Standards.
          Technical Report No. 64. New Canaan, Ct: SIMS, [October 1982].

Wolter, K. M. 1984. "Investigation of Some Estimators of Variance for Systematic
          Sampling." Journal of the American Statistical Association.  Vol. 79, No. 388:
          781-790.

Yakowitz, S. J., and F. Szidarovsky. 1985.  "A Comparison of Kriging with
          Nonparametric Regression Methods." Journal of Multivariate Analysis. Vol.
          16: 21-53.

Yates, F. 1981. Sampling Methods for Censuses and Surveys, Fourth Edition.  New
          York: Oxford University Press.

Yfantis, E.  A., Flatman, G. T., and J. V. Behar. 1987. "Efficiency of Kriging
          Estimation for Square, Triangular, and Hexogonal Grids." Journal of
          Mathematical Geology. Vol. 19, No. 3: 183-205.

Zirschky, J., and R. O. Gilbert. 1984. "Detecting Hot Spots at Hazardous Waste
          Sites." Chemical Engineering. Vol.91: 97-100.
                                    BIB-11

-------

-------
            APPENDIX A:   STATISTICAL  TABLES
Table A. 1     Table of t for selected alpha and degrees of freedom
Use alpha to determine which column to use based on the desired parameter, ti-
-------
                     APPENDIX A:  STATISTICAL TABLES


Table A.2     Table of z for selected alpha or beta


Use alpha or beta to determine which row to read. Obtain the z value from the zi-
-------
                     APPENDIX A: STATISTICAL TABLES
Table A.3     Table of k for selected alpha, PQ, and sample size where alpha = 0.10
             (i.e., 10%)


Use alpha to determine which table to read. The k for use in a tolerance interval test is at
the intersection of the column with the specified PO and the row with the sample size, n.
n
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
35
40
50
70
100
200
500
infinity
0.25
5.842
2.603
1.972
1.698
1.540
1.435
1.360
1.302
1.257
1.219
1.188
1.162
1.139
1.119
1.101
1.085
1.071
1.058
1.046
1.035
1.025
1.016
1.007
1.000
0.992
0.985
0.979
0.973
0.967
0.942
0.923
0.894
0.857
0.825
0.779
0.740
0.674
PO
0.1
10.253
4.258
3.188
2.742
2.494
2.333
2.219
2.133
2.066
2.011
1.966
1.928
1.895
1.867
1.842
1.819
1.800
1.782
1.765
1.750
1.737
1.724
1.712
1.702
1.691
1.682
1.673
1.665
1.657
1.624
1.598
1.559
1.511
1.470
1.411
1.362
1.282
0.05
13.090
5.311
3.957
3.400
3.092
2.894
2.754
2.650
2.568
2.503
2.448
2.402
2.363
2.329
2.299
2.272
2.249
2.227
2.208
2.190
2.174
2.159
2.145
2.132
2.120
2.109
2.099
2.089
2.080
2.041
2.010
1.965
1.909
1.861
1.793
1.736
1.645
0.010
18.500
7.340
5.438
4.666
4.243
3.972
3.783
3.641
3.532
3.443
3.371
3.309
3.257
3.212
3.172
3.137
3.105
3.077
3.052
3.028
3.007
2.987
2.969
2.952
2.937
2.922
2.909
2.896
2.884
2.833
2.793
2.735
2.662
2.601
2.514
2.442
2.326
                                      A-3

-------
                    APPENDIX A:  STATISTICAL TABLES
Table A.4    Table of k for selected alpha, PQ, and sample size where alpha =  0.05
            (i.e., 5%)
n
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
35
40
50
70
100
200
500
infinity
0.25
11.763
3.806
2.618
2.150
1.895
1.732
1.618
1.532
1.465
1.411
1.366
1.328
1.296
1.268
1.243
1.220
1.201
1.183
1.166
1.152
1.138
1.125
1.114
1.103
1.093
1.083
1.075
1.066
1.058
1.025
0.999
0.960
0.911
0.870
0.809
0.758
0.674
PO
0.1
20.581
6.155
4.162
3.407
3.006
2.755
2.582
2.454
2.355
2.275
2.210
2.155
2.109
2.068
2.033
2.002
1.974
1.949
1.926
1.905
1.886
1.869
1.853
1.838
1.824
1.811
1.799
1.788
1. 777
1.732
1.697
1.646
1.581
1.527
1.450
1.385
1.282
0.05
26.260
7.656
5.144
4.203
3.708
3.399
3.187
3.031
2.911
2.815
2.736
2.671
2.614
2.566
2.524
2.486
2.453
2.423
2.396
2.371
2.349
2.328
2.309
2.292
2.275
2.260
2.246
2.232
2.220
2.167
2.125
2.065
1.990
1.927
1.837
1.763
1.645
0010
37.094
10.553
7.042
5.741
5.062
4.642
4.354
4.143
3.981
3.852
3.747
3.659
3.585
3.520
3.464
3.414
3.370
3.331
3.295
3.263
3.233
3.206
3.181
3.158
3.136
3.116
3.098
3.080
3.064
2.995
2.941
2.862
2.765
2.684
2.570
2.475
2.326
                                   A-4

-------
                    APPENDIX A: STATISTICAL TABLES
Table A.5    Table of k for selected alpha, PQ, and sample size where alpha = 0.01 (i.e.,
            1%)

     n                                PO

2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
35
40
50
70
100
200
500
infinity
0.25
58.939
8.728
4.715
3.454
2.848
2.491
2.253
2.083
1.954
1.853
1.771
1.703
1.645
1.595
1.552
1.514
1.481
1.450
1.423
1.399
1.376
1.355
1.336
1.319
1.303
1.287
1.273
1.260
1.247
1.195
1.154
1.094
1.020
0.957
0.868
0.794
0.674
0.1
103.029
13.995
7.380
5.362
4.411
3.859
3.497
3.240
3.048
2.898
2.777
2.677
2.593
2.521
2.459
2.405
2.357
2.314
2.276
2.241
2.209
2.180
2.154
2.129
2.105
2.085
2.065
2.047
2.030
1.957
1.902
1.821
1.722
1.639
1.524
1.430
1.282
0.05
131.426
17.370
9.083
6.578
5.406
4.728
4.258
3.972
3.738
3.556
3.410
3.290
3.189
3.102
3.028
2.963
2.905
2.854
2.808
2.766
2.729
2.694
2.662
2.633
2.606
2.581
2.558
2.536
2.515
2.430
2.364
2.269
2.153
2.056
1.923
1.814
1.645
0.010
185.617
23.896
12.387
8.939
7.335
6.412
5.812
5.389
5.074
4.829
4.633
4.472
4.337
4.222
4.123
4.037
3.960
3.892
3.832
3.777
3.727
3.681
3.640
3.601
3.566
3.533
3.502
3.473
3.447
3.334
3.249
3.125
2.974
2.850
2.679
2.540
2.326
                                   A-5

-------
                     APPENDIX A: STATISTICAL TABLES
Table A.6    Sample sizes required for detecting a scaled difference tau of the mean from
             the cleanup standard for selected values of alpha and beta*


1
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0.50
0.55
0.60
0.65
0.70
0.75
0.80
0.85
0.90
0.95
1.00
P = 0.20
a
0.10
1,798
449
200
112
72
50
37
28
22
18
15
12
11
9
8
7
6
6
5
4
0.05
2,470
618
274
154
99
69
50
39
30
25
20
17
15
13
11
10
9
8
7
6
0.01
4,020
1,005
447
251
161
112
82
63
50
40
33
28
24
21
18
16
14
12
11
10
3 = 0.10
a
0.10
2,621
655
291
164
105
73
53
41
32
26
22
18
16
13
12
10
9
8
7
7
0.05
3,422
856
380
214
137
95
70
53
42
34
28
24
20
17
15
13
12
11
9
9
0.01
5,213
1,303
579
326
209
145
106
81
64
52
43
36
31
27
23
20
18
16
14
13
*See section 6.1 and Box 6.3 for definitions of alpha (a), beta ((3), and tau (T).
                                      A-6

-------
                     APPENDIX A: STATISTICAL TABLES
Table A.7     Sample size required for test for proportions with a = .01 and (3 = .20, for
             selected values of PO and PI
PO
0.005
0.010
0.020
0.030
0.040
0.050
0.060
0.070
0.080
0.090
0.100
0.002
4,519
1,131
407
241
169
129
103
86
73
64
56
0.005
3,383
659
333
217
158
124
101
85
73
64
Value of P under the alternative hypothesis, PI
0.010 0.020 0.030 0.040 0.050 0.060
1,676
577
323
218
162
127
104
88
75

2,649
823
434
281
202
156
125
104


3,593
1,058
538
340
240
182
144



4,515
1,287
639
396
276
207




5,416
1,509
737
451
311





6,295
1,726
833
504
0.070 0.080






7,155
1,938 7,994
925 2,145
0.090








8,813
                    Value of P under the alternative hypothesis, PI
PO
0.050
0.100
0.150
0.200
0.250
0.300
0.350
0.400
0.450
0.500
0.010
218
75
43
28
21
16
12
10
8
6
0.020
434
104
53
34
24
18
14
11
9
7
0.050

311
103
55
35
25
19
14
11
9
0.100


469
140
71
43
30
22
16
13
0.150



606
171
83
50
33
24
17
0.200




723
197
93
54
36
25
0.250





819
217
101
58
37
0.300






894
233
106
60
0.350 0.400







950
243 986
109 248
0.450









1,001
                                      A-7

-------
                      APPENDIX A: STATISTICAL TABLES
Table A.8     Sample size required for test for proportions with a = .05 and P = .20, for
              selected values of PO and PI
                    Value of P under the alternative hypothesis, PI
PO
0.005
0.010
0.020
0.030
0.040
0.050
0.060
0.070
0.080
0.090
0.100
0.002
2,623
633
222
129
90
68
55
45
38
33
29
0.005

1,990
373
185
119
86
67
54
45
39
34
0.010


986
332
183
122
90
70
57
48
41
0.020



1,588
485
252
162
116
88
71
58
0.030




2,171
630
317
198
139
105
83
0.040





2,741
772
380
234
162
120
0.050






3,297
910
441
268
183
0.060







3,840
1,044
500
301
0.070 0.080








4,371
1,175 4,889
557 1,303
0.090










5,394
                    Value of P under the alternative hypothesis, PI
PO
0.050
0.100
0.150
0.200
0.250
0.300
0.350
0.400
0.450
0.500
0.010
122
41
23
15
11
8
7
5
4
3
0.020
252
58
29
19
13
10
7
6
5
4
0.050

183
59
31
20
14
10
8
6
5
0.100


282
83
41
25
17
12
9
7
0.150



368
103
49
29
20
14
10
0.200




440
119
56
33
21
15
0.250





500
132
61
35
23
0.300






548
142
64
37
0.350







583
149
67
0.400 0.450








606
153 616
                                       A-8

-------
                      APPENDIX A:  STATISTICAL TABLES
Table A.9     Sample size required for test for proportions with a = .10 and (3 = .20, for
              selected values of PQ and PI
                     Value of P under the alternative hypothesis, PI
PO
0.005
0.010
0.020
0.030
0.040
0.050
0.060
0.070
0.080
0.090
0.100
0.002
1,822
426
145
84
58
44
35
29
24
21
19
0.005

1,398
254
124
79
57
44
35
29
25
22
0.010


693
229
125
82
60
47
38
32
27
0.020



1,133
341
175
111
79
60
48
39
0.030




1,559
447
223
138
97
72
57
0.040





1,975
551
269
164
113
84
0.050






2,381
652
314
189
129
0.060







2,778
750
357
214
0.070








3,166
846
399
0.080 0.090









3,544
940 3,913
                     Value of P under the alternative hypothesis, PI
PO
0.050
0.100
0.150
0.200
0.250
0.300
0.350
0.400
0.450
0.500
0.010
82
27
15
10
7
5
4
3
3
2
0.020
175
39
20
12
9
6
5
4
3
2
0.050

129
41
22
14
10
7
5
4
3
0.100


202
59
29
18
12
9
6
5
0.150



265
73
35
21
14
10
7
0.200




318
85
40
23
15
11
0.250





363
95
44
25
16
0.300






398
103
47
26
0.350







424
108
48
0.400 0.450








441
111 449
                                        A-9

-------
                     APPENDIX A: STATISTICAL TABLES
Table A. 10    Tables for determining critical values for the exact binomial test, with a =
             0.01, 0.05, and 0.10

To determine the critical value, select the column for PQ specified in the attainment
objectives, reading down the column finding the first number greater than the sample size,
n, move up one row, read ra:n. the critical value, in the leftmost column.





Alpha =.01
PQ, Proportion of contaminated soil units under
rft;n
0
1
2
3
4
5
6
7
8
9
10
0,01
459
662
838
1001
1157
1307
1453
1596
1736
1874
2010

0,02
228
330
418
499
577
652
725
796
866
935
1003

0.03
152
219
277
332
383
433
482
529
576
622
667

0.04
113
164
207
248
287
324
360
396
431
465
499

0.05
90
130
165
198
229
259
288
316
344
371
398

0.06 0.07
75 64
108 92
137 117
164 140
190 162
215 184
239 204
263 225
286 244
309 264
331 283
Alpha = .05
PQ, Proportion of contaminated soil units under
r«:n
0
1
2
3
4
5
6
7
8
9
10
0.01
299
473
628
773
913
1049
1182
1312
1441
1568
1693

0,02
149
236
313
386
456
523
590
655
719
782
845

0.03
99
157
208
257
303
348
392
436
478
521
562

0.04
74
117
156
192
227
261
294
326
358
390
421

0.05
59
93
124
153
181
208
234
260
286
311
336

0.06 0.07
49 42
78 66
103 88
127 109
150 129
173 148
195 167
217 185
238 203
259 221
280 239
Alpha =.10





the null hypothesis
0.08
56
81
102
122
142
160
178
196
213
230
247

the null
0.08
36
58
77
95
112
129
146
162
178
193
209

PQ, Proportion of contaminated soil units under the null
ra;n
0
1
2
3
4
5
6
7
8
9
10
0,01
230
388
531
667
798
926
1051
1175
1297
1418
1538
0,02
114
194
265
333
398
462
525
587
648
708
768
0.03
76
129
176
221
265
308
349
390
431
471
511
0.04
57
%
132
166
198
230
262
292
323
353
383
0.05
45
77
105
132
158
184
209
234
258
282
306
0.06 0.07
38 32
64 55
88 75
110 94
132 113
153 131
174 149
194 166
215 184
235 201
255 218
0.08
28
48
65
82
98
114
130
145
160
175
190
0.09
49
71
91
109
126
142
158
174
189
204
219

0.10
44
64
81
97
113
127
142
156
170
183
197

0.11
40
58
74
88
102
116
129
141
154
166
178

0,12
37
53
67
81
93
106
118
129
141
152
163

hypothesis
0.09
32
51
68
84
100
115
129
143
158
172
185

0.10
29
46
61
76
89
103
116
129
142
154
167

0.11
26
42
56
69
81
93
105
117
128
140
151

0.12
24
38
51
63
74
85
96
107
117
128
138

hypothesis
0.09
25
42
58
73
87
101
115
129
142
156
169
0.10
22
38
52
65
78
91
104
116
128
140
152
0.11
20
34
47
59
71
83
94
105
116
127
138
0.12
19
31
43
54
65
76
86
96
106
116
126
                                     A-10

-------
                     APPENDIX A: STATISTICAL TABLES
Table A. 11    The false positive rates associated with hot spot searches as a function of
             grid spacing and hot spot shape
                                False Positive Rates
                                                      ES
Triangular Grid Pattern






Square Grid Pattern





L/G
0.1
0.3
0.5
0.7
0.9
1.0
0.1
0.3
0.5
0.7
0.9
1.0
1.0
.95
.66
.08
.00
.00
.00
.97
.72
.21
.00
.00
.00
.80
.96
.74
.27
.00
.00
.00
.97
.77
.38
.02
.00
.00
.60
.97
.80
.44
.08
.00
.00
.98
.80
.54
.16
.00
.00
.40
.98
.86
.63
.33
.10
.04
.98
.88
.69
.42
.17
.08
.20
.98
.93
.82
.65
.47
.37
.98
.94
.85
.70
.53
.44
.10
.99
.96
.91
.83
.72
.66
.99
.97
.92
.85
.76
.70
Source: These tables were extracted from the graphs in Gilbert (1987).
                                     A-ll

-------
                      APPENDIX A: STATISTICAL TABLES
Figure A. 1    Power Curves for a = 1 %
           0.9-.
           0.8-.
Probability  0.7 •.
of Deciding
  the Site
Attains the  0-5 • •
 Cleanup
  Standard
                                                        Cs or Po
                                                                            A

                                                                            B

                                                                            r*
                                                                        •— F
                       True parameter as a fraction of Cs or
Parameters for the
Power Curves
a =
P =
Hl =
Pl-
A J
.01
.20
.19*Cs
.19*P0
B
.01
.20
.36*Cs
.36*P0
Powei
C
.01
.20
.53*Cs
.53*P0
r Curve:
D
.01
.20
.65*Cs
.65*P0
E
.01
.20
.75*Cs
.75*P0
F
.01
.20
.81*Cs
.81*P0
 Approximate sample sizes for simple random sampling for testing the parameters indicated
                                                 Power Curve:
Parameters being tested |     A    j    B     |     C    I     D    I     E    I     F
Mean
with cv(data) = .5
with cv(data) = 1
withcv(data) = 1.5
Proportions
P0 = 10%
Non-parametric test
Tolerance Intervals
P0 = 20%
Non-parametric test
Tolerance Intervals

4
16
35


101
16

46
12

7
25
56


179
38

81
26

12
46
103


356
89

161
60

21
82
185


670
184

301
122

41
161
362


1353
399

607
261

70
279
626


2384
728

1066
473
     Note:  a = saying the site is clean when dirty, (3 = saying the site is dirty when clean, 1-(J = saying
     the site is clean when clean.
                                        A-12

-------
                      APPENDIX A:  STATISTICAL TABLES
     Figure A.2      Power Curves for a = 5%
                 1 -
               0.9 ..
               0.8 ..
     Probability  0.7 • .
    of Deciding  0.6 ..
                                     V   Cs or Po
      the Site
    Attains the
      Cleanup
      Standard
0.5 ..
0.4 ..
0.3 ••
0.2 -.
0.1 ••
 0
                                                             A

                                                             B
                                                                         «-«  F
Parameters for the
Power Curves
        True parameter as a fraction of Cs or PQ.
                                  Power Curve:
        I     A    |    B     |     C    |    D
a =
P =
Hi —
P ^
.05
.20
.25*Cs
.25*P0
.05
.20
.43*Cs
.43*P0
.05
.20
.57*Cs
.57*P0
.05
.20
.69*Cs
.69*P0
.05
.20
.77*Cs
.77*P0
.05
.20
.84*Cs
.84*P0
 Approximate sample sizes for simple random sampling for testing the parameters indicated
                                                 Power Curve:
Parameters being tested I     A    |    B     I     C    I    D    I     E    |     F
Mean
with cv(data) = .5
with cv(data) = 1
withcv(data) =1.5
Proportions
P0 = 10%
Non-parametric test
Tolerance Intervals
P0 = 20%
Non-parametric test
Tolerance Intervals

4
11
25


70
14

32
10

5
20
43


136
33

62
23

9
34
76


257
69

116
47

17
65
145


520
151

234
100

30
117
264


975
296

438
193

61
242
544


2065
649

925
420
Note:  a = saying the site is clean when dirty, (i = saying the site is dirty when clean, l-(3 = saying the site
      is clean when clean.
                                        A-13

-------
                      APPENDIX A:  STATISTICAL TABLES
Figure A.3    Power Curves for a = 10%
    Probability
    of Deciding
     the Site
    Attains the
     Cleanup
     Standard
Parameters for the
Power Curves
 True parameter as a fraction of Cs or PQ.
                          Power Curve:
I     A    |    B     I    C     |    D
                                                                           A

                                                                           B
                                                                        '-*  F
a =
P =
Hi =
PI =
.10
.20
.30*Cs
.30*P0
.10
.20
.46*Cs
.46*P0
.10
.20
.60*Cs
.60*P0
.10
.20
.71*Cs
.71*P0
.10
.20
.79*Cs
.79*P0
.10
.20
.85*Cs
.85*P0
 Approximate sample sizes for simple random sampling for testing the parameters indicated
                                                Power Curve:
Parameters being tested I    A    I    B     I     C    I    D    I    E    I     F
Mean
with cv(data) = .5
with cv(data) = 1
withcv(data) =1.5
Proportions
P0 = 10%
Non-parametric test
Tolerance Intervals
P0 = 20%
Non-parametric test
Tolerance Intervals

3
10
21


57
13

26
9

4
16
35


108
28

50
19

8
29
64


214
60

97
40

14
54
121


430
129

194
85

26
103
231


849
264

382
172

51
201
452


1706
544

764
351
Note: a = saying the site is clean when dirty, P = saying the site is dirty when clean, 1-(J = saying the site
      is clean when clean.
                                       A-14

-------
                      APPENDIX A:  STATISTICAL TABLES
Figure A.4    Power Curves for a = 25%
                                                        CsorPo
    Probability
    of Deciding
     the Site
    Attains the
     Cleanup
     Standard
— A

— B



    Q
                                                                        ~-  F
                       True parameter as a fraction of Cs or
Parameters for the
Power Curves
a =
P =
M| =
P —
A
.25
.20
.19*Cs
.19*P0
B
.25
.20
.40*Cs
.40*P0
Powe
C
.25
.20
.54*Cs
.54*P0
r Curve:
D
.25
.20
.76*Cs
.76*P0
E |
.25
.20
.83*Cs
.83*P0
F
.25
.20
.87*Cs
.87*P0
 Approximate sample sizes for simple random sampling for testing the parameters indicated
                                                Power Curve:
Parameters being tested I     A    I   B     I     C    I     D    I     E    I     F
Mean
with cv(data) = .5
with cv(data) = 1
with cv(data) =1.5
Proportions
P0 = 10%
Non-parametric test
Tolerance Intervals
P0 = 20%
Non-parametric test
Tolerance Intervals

2
7
15


38
11

18
8

3
11
25


73
22

34
15

5
20
45


147
46

67
30

10
40
90


315
100

142
66

20
80
179


654
212

294
138

34
136
306


1143
375

513
242
Note: a = saying the site is clean when dirty, p = saying the site is dirty when clean, 1-|3 = saying the site
     is clean when clean.
                                       A-15

-------

-------
          APPENDIX B:  EXAMPLE  WORKSHEETS

          The worksheets in this appendix have been completed to serve as an example in
understanding the forms and making the necessary calculations.

          The numbers and situations represented on the worksheets are hypothetical.
The example situation consists of a waste site that is divided into two sample areas. The
first uses random sampling  to test the mean and proportion of contaminated soil for two
chemicals. The second uses stratified sampling to test the mean and proportion of
contaminated soil for one chemical. In this example, the different chemicals, labeled only
Chemical #1, #2, and #3, are tested in the different sample areas; in most applications, the
same chemicals will be tested in all or most of the sample areas. Two statistical parameters
are tested  for two chemicals to show how to complete the worksheet under a variety of
conditions.

          The following figures show the 1) the parameters being tested, 2) a hypothetical
map of the site, and 3) the sequence in which the worksheets are completed.  The
worksheets for sample area #2 follow those for sample area #1 in this appendix.

          In actual  use,  these  worksheets  would be accompanied by additional
documentation such as maps, background material, justification of different choices, field
notes, and copies of the results as reported by the laboratory.
                                         B-l

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
Figure B. 1    Example Worksheets: Parameters to Test in Each Sample Area and Map of the Site
                                             Waste Site
                                        Old XYZ Disposal Site
              a
              a.

              a
           Sample area #1
    Field used for storing batteries
         Random sampling
  Sample area #2
 Old Lagoon Area
Stratified sampling
             e
             I
Mean concentration
  of Chemical #1
                           Proportion of soil
                           contaminated with
                              Chemical #1
                                      Mean concentration
                                        of Chemical #3
             a>
            6
                     Proportion of soil
                     contaminated with
                        Chemical #2
                                                                    Proportion of  soil
                                                                    contaminated with
                                                                      Chemical #3
                                           Waste Site

                             Sample area#l
                             Field used for
                             storing batteries
                                          Stratum #1
                                        Center of Lagoon
                                                ^  Stratum #2 Edge of Lagoon
                                                ^                 .......
                                         B-2

-------
                      APPENDIX B: EXAMPLE WORKSHEETS
Figure B.2    Example Worksheets: Sequence in Which the Worksheets Are Completed
                                 Worksheet 1
                              Define sample areas
             Worksheet 2
            Sample area #1
           Random Sample
          Attainment Objectives
                 I
              Worksheet 3
             Sample Design
            and Analysis Plan
   Worksheet 4
  Sample Size for
  testing the mean
    Worksheet 6
    Chemical #1
   Data Sheet and
    Calculations
    Worksheet 7
    Analysis and
     Inference
   Worksheet 5
  Sample Size for
testing proportion
  Worksheet 6
  Chemical #2
 Data Sheet and
  Calculations
   Worksheet 7
   Analysis and
     Inference
                                  Worksheet 2
                                Sample area #2
                               Stratified Sample
                               Attainment Objectives
                                      I
                                  Worksheet 3
                                 Sample Design
                                and Analysis Plan
           Worksheet 8
           Define Strata
Worksheet 9
Chemical #3
Sample Sizes
for the mean
  Worksheet 10
  Chemical #3
  Sample Sizes
for proportions
          Worksheet 11
        Field Sample Sizes
                                          Worksheet 12
                                           Stratum #1
                                          Chemical #3
                                          Data Sheet and
                                           Calculations
                                         Worksheet 12
                                          Stratum #2
                                         Chemical #3
                                         Data Sheet and
                                          Calculations
                                         Worksheet 13
                                         Chemical #3
                                         Analysis and
                                         Inference for
                                           the mean
                                        Worksheet 14
                                        Chemical #3
                                        Analysis and
                                        Inference for
                                        proportions
                                       B-3

-------
                       APPENDIX B: EXAMPLE WORKSHEETS



                         WORKSHEET 1  Sample Areas

See Section 3.1 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
           SITE:   Former XYZ Disposal Site
Sample
 Area
Number      Describe the sample areas and the reasons for treating each area separately.
   g
    1
Field used for storing batteries
         Old Lagoon Area
Use the Sample Area Number (g) to refer on other sheets to the sample areas described above.
Attach a map showing the sample areas within the waste site.

Date Completed: EXAMPLE                  Completed by      EXAMPLE

Use additional sheets if necessary.                                       Page	of.


Continue to WORKSHEET 2
                                        B-4

-------
                        APPENDIX B: EXAMPLE WORKSHEETS



                       WORKSHEET  2  Attainment Objectives

 See Section 3.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
            SITE:
       Former XYZ Disposal Site
                   NUMBER(g) AND DESCRIPTION [1]
  SAMPLE AREA:	1. Field used for storing batteries
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.

Sample Collection Procedures to be used (attach separate sheet if necessary):
For example: .5 liter scoop of soil from the top 5 cm of soil, etc.
 Probability of mistakenly declaring the site clean = a =
                                    .05
 Chemical
to be tested
 Number
    j
Chemical
 Name
 Cleanup
 Standard
(with units)
    Cs
                                                  Parameter to test:
 Mean
Yes/No
Proportion
    PO
1
2



Chemical #1
Chemical #2



20
2



Yes
No



25%
50%



Secondary Objectives/ Other purposes for which the data is to be collected:
Use the Chemical Number (j) to refer on other sheets to the chemical described above.
Attach documentation describing the lab analysis procedure for each chemical.

Date Completed:  EXAMPLE                   Completed by       EXAMPLE

Use additional sheets if necessary.                                        Page	of

Continue to WORKSHEET 3
                                          B-5

-------
                       APPENDIX B: EXAMPLE WORKSHEETS

              WORKSHEET  3  Sampling Design  and Analysis Plan
See Chapter 4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
           SITE:
Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1]
 SAMPLE AREA:	1. Field used for storing batteries
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
Sample Design:
                               Simple Random Sample
                               Systematic Random Sample
                               Stratified Sample
Chemical Comments on the Prob of Type n error Alternate Parameter value
to be tested Sample Design and Chance of concluding the for the specified P
Number [2] Analysis Plan site is dirty when it is clean Mean Proportion
j p m P!
1
2








.20
.20



15




5%
20%



Date Completed:  EXAMPLE
Use additional sheets if necessary.
                       Completed by
EXAMPLE
                                              Page
              of
Continue to WORKSHEET 4 for random or systematic sampling and WORKSHEET 8 for stratified sampling.
                                         B-6

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
    WORKSHEET  4  Sample Size for Testing the Mean  Using Simple Random
                                      Sampling

See Section 6.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
If the mean concentration is not to be tested for this chemical, continue to WORKSHEET 5	
            SITE:
      Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [ 1 ]
  SAMPLE AREA:        	1. Field used for storing batteries
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.

                                                                  From z -Table, Appendix A
 Probability of mistakenly declaring the site clean [2] = a
                                                    zl-a =
Chemical
Number   [3]
   [2]
  From
  z table
Appendix A

   Zl.p
                                Calculate:
[2]


Cs
[3]
                                                           A=
                                                                 Cs-m
                                                                              nj=.
1




.20




.842




20




15




49




4.042




Column Maximum, Max n; =
12.12




12.12
                              Fraction of samples expected to be analyzable = R =
                                                              Max n[
                                                                  T, '' = B =
                                                                  R
                     B rounded up  = Sample Size for  Testing Means = nf =
Date Completed:  EXAMPLE

Use additional sheets if necessary.

Continue to WORKSHEET 5
                             Completed by
                                   EXAMPLE
                                                     Page
                                                 of
                                          B-7

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
  WORKSHEET 5  Sample Size for Testing Proportions Using Simple  Random
                                      Sampling

See Section 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
If the mean concentration is not to be tested for this chemical, continue to WORKSHEET 6	
            SITE:
       Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1]
 SAMPLE AREA:	1. Field used for storing batteries
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
                                                     	  From z -Table, Appendix A
  Probability of mistakenly declaring the site clean [2] = a

                                        Calculate:
                                        .05
Chemical
Number   [3]
   [2]

    j      P
From
z table
                           [2]


                           PO
[3]


Pi
I
2



.20
.20



.842
.842



.25
.50



.05
.20



.712
.823



.184
.337



Column Maximum, Max HJ =
20.06
14.93



20.06
                               Fraction of samples expected to be collectible = B =
                                                              Max nj  _ p _
                                                                  B        ~~
                      C rounded up = Sample  Size for Testing Proportions =
Date Completed:  EXAMPLE

Use additional sheets if necessary.

Continue to WORKSHEET 6
                               Completed by
                                                                  EXAMPLE
                                                      Page
                                                                                of
                                          B-8

-------
                       APPENDIX B: EXAMPLE WORKSHEETS


 WORKSHEET  6  Data Calculations for a Simple Random Sample, by Chemical

See Section 7.3 or 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1]
 SAMPLE AREA:                        1. Field used for storing batteries
     CHEMICAL:
                  NUMBER(j) AND DESCRIPTION [2]
                       1. Chemical #1
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
              Maximum Sample Size from Worksheets 4 and 5 = Sample Size =
                                                 Cleanup standard[2] = Cs
                                                Method Detection Limit: =

   Concentration used when it is reported as less than the method detection limit =
                                                      22
                                                      20
                       Was the    Reported                   Is \[ Greater
                       Sample   Concentration  Concentration   than Cs?
Sample     Sample   Collectible?      If        Corrected for     l = Yes
Number       ID        0 = No    Collectible    Detection Limit    0 = No
   i                    l = Yes                      x;            y
                                                     (Xi)2
1
2
3
4
5
6
7
8
9
10
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
Total from previous page
Column Totals:
1
1
1
1
1
1
1
0
1
1
14.7
17.7
22.8
2.9
35.5
28.6
4.9
#N/A
5.2
17.2
14.7
17.7
22.8
4
35.5
28.6
4.9
0
5.2
17.2





A 9

B 150.6
0
0
1
0
1
1
0
0
0
0



C 3
216.09
313.29
519.84
16
1260.25
817.96
24.01
0
27.04
295.84



D 3490.32
                        A = n
Date Completed:  EXAMPLE
Use additional sheets if necessary.
                       Completed by
                                        C = r
EXAMPLE
                                              Page
              of
Complete WORKSHEET 6 for other chemicals or continue to WORKSHEET 7
                                         B-9

-------
                       APPENDIX B: EXAMPLE WORKSHEETS



  WORKSHEET  6  Data Calculations for a Simple Random Sample, by  Chemical

See Section 6.3 or 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
           SITE:
           Former XYZ Disposal Site
 SAMPLE AREA:
                  NUMBER(g) AND DESCRIPTION [ 1 ]
                            1. Field used for storing batteries
     CHEMICAL:
                  NUMBER(j) AND DESCRIPTION [2]
                                  1. Chemical #1
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
             Maximum Sample Size from Worksheets 4 and 5 = Sample Size =
                                                Cleanup standard[2] = Cs

                                               Method Detection Limit: =

          Concentration used when it is reported as less than the detection limit =
                                                                 22
                                                                 20
Sample
Number
           Was the    Reported
           Sample  Concentration
Sample   Collectible?      If
  ID        0 = No    Collectible
            l = Yes
              Is Xj Greater
 Concentration   than Cs?
 Corrected for     1 = Yes
Detection Limit    0 = No
1
11
12
13
14
15
16
17
18
19
20
2243
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
Total from previous page
Column Totals:
1
1
1
1
1
1
1
1
1
1
1
14.7
10.9
7.7
12.4
15.2
14.9
10.2
17.4
11.6
12.4
19.1
14.7
10.9
7.7
12.4
15.2
14.9
10.2
17.4
11.6
12.4
19.1
0
0
0
0
0
0
0
0
0
0
0

9


A 19

150.6

B 282.4
3
216.09
118.81
59.29
153.76
231.04
222.01
104.04
302.76
134.56
153.76
364.81

3490.3

C 3
D 5335.2
                        A = n
    B =
Date Completed:  EXAMPLE

Use additional sheets if necessary.
                                  Completed by
                                                   C = r
                   EXAMPLE
                                                         Page
                                 of
Complete WORKSHEET 6 for other chemicals or continue to WORKSHEET 7
                                        B-10

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
 WORKSHEET  6  Data Calculations for a Simple Random Sample,  by Chemical

See Section 6.3 or 7.3 in "Statistical Methods for Evaluating the Attainment of Superfund Cleanup Standards",
Volume 1                                                  	  	
           SITE:
Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1]
 SAMPLE AREA:                        1. Field used for storing batteries
     CHEMICAL:
                  NUMBER(J) AND DESCRIPTION [2]
                       1. Chemical #1
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
             Maximum Sample Size from Worksheets 4 and 5 = Sample Size =
                                                Cleanup standard [2] = Cs
                                                Method Detection Limit: =
         Concentration used when it is reported as less than the detection limit =
                       Was the     Reported
                       Sample   Concentration
                                     Is Xj Greater
                       Concentration   than Cs?
                                                      22
                                                      20
Sample Sample Collectible? If Corrected for l = Yes
Number ID 0 = No Collectible Detection Limit 0 = No
i l = Yes Xj yj (xO2
21
22








2263
2264








Total from previous page
Column Totals:
1
1








8.9
16.5








8.9
16.5









19

282.4

A 21

B 307.8
0
0









3
79.21
272.25









5335.2

C 3
D 5686.6
                        A = n
                                        C = r
Date Completed:  EXAMPLE
Use additional sheets if necessary.
                       Completed by
EXAMPLE
                                              Page
              of
Complete WORKSHEET 6 for other chemicals or continue to WORKSHEET 7
                                        B-ll

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
      WORKSHEET  7  Inference for Simple Random Samples by Chemical

Sec Section 6.3 or 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
                       Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1]
 SAMPLE AREA:	1. Field used for storing batteries
     CHEMICAL:
                  NUMBER(j) AND DESCRIPTION [2]
                                              1. Chemical #1
                                                                           .05
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.

Testing the Mean                                 [2]     a =
                                                     [2]
                             Number of Collectible Samples [6] = n =

              Total of the concentration measurements [6] = V xj = B =

                                   Total for x^ [6] =
                                                   .     B   _
                                     Mean concentration = — = x =
                      Standard Deviation of the Data = 'V ——j— = s =

                             Degrees of Freedom for s = n - 1 = df =
                     Standard Error for the Mean concentration = -7=- =
                                                           Vn
                                            _          §
         Upper One Sided Confidence Interval = x + ti_a>(jf -r=-

         If \i\j(,< Cs then circle Clean, otherwise  circle Dirty:
        __ Based on the mean concentration, the sample area is:
=
=
=
=
-
1
1 «
DC""
20
21
307.8
5687
14.66
7.67
20
1.73
1.67
17.54

Clean Dirty
Testing Percentiles                              [2]    P0 =
                                                    [4 or 5] zi.a =
        Number of Samples with Concentration Greater than Cs [6] = r =

                       Proportion of Contaminated Samples = — =  p =
                                                                            .25
                   Standard Error for the Proportion = "V   n   = sp =
                           Test Statistic = p + zj.a 'V ^  „  =
          If UL < PQ then circle Clean, otherwise circle Dirty:
  Based on the proportion of contaminated samples, the sample area is:
                                                                           1.645
                                                                             3
                                                                           .143
                                                                           .0764
                                                                            .298
                                                                  Clean    Dirty
Date Completed:  EXAMPLE

Complete WORKSHEET 7 for other chemicals
                                              Completed by
EXAMPLE
    Page_
                                                                               of
                                         B-12

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
 WORKSHEET  6  Data Calculations for a Simple Random Sample, by Chemical

See Section 6.3 or 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
           Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [ 1 ]
 SAMPLE AREA:                        1. Field used for storing batteries
     CHEMICAL:
                  NUMBER(j) AND DESCRIPTION [2]
                                  2. Chemical #2
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
             Maximum Sample Size from Worksheets 4 and 5 = Sample Size =
                                                Cleanup standard [2] = Cs
                                                Method Detection Limit: =
         Concentration used when it is reported as less than the detection limit =
                                                                  22
                                                                  20
                                                                  1.2
                                                                  1.2
Sample
Number
           Was the    Reported
           Sample  Concentration
Sample    Collectible?      If
  ID        0 = No     Collectible
            l = Yes
              Is Xj Greater
 Concentration    than Cs?
 Corrected for     l = Yes
Detection Limit    0 = No
                   Xi
                                                                            (Xi)2
1
2
3
4
5
6
7
8
9
10










1
1
1
1
1
1
1
0
1
1
1.2
2.1
0.9
0.1
0.5
0.3
0.3
#N/A
1.9
8.3
1.2
2.1
1.2
1.2
1.2
1.2
1.2
0
1.9
8.3
0
1
0
0
0
0
0
0
0
1










Total from previous page
Column Totals:


B
19.5



C 2
D
                        A = n
Date Completed:  EXAMPLE
Use additional sheets if necessary.
                                   Completed by
                                                    C = r
                   EXAMPLE
                                                          Page
                                 of
Complete WORKSHEET 6 for other chemicals or continue to WORKSHEET 7
                                         B-13

-------
                       APPENDIX B: EXAMPLE WORKSHEETS



 WORKSHEET 6  Data  Calculations for a Simple Random Sample, by Chemical

See Section 6.3 or 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
           Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [Ij
 SAMPLE AREA:	1. Field used for storing batteries
     CHEMICAL:
                  NUMBER(j) AND DESCRIPTION [2]
                                  2. Chemical #2
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
             Maximum Sample Size from Worksheets 4 and 5 = Sample Size =
                                               Cleanup standard [2] = Cs

                                               Method Detection Limit: =

         Concentration used when it is reported as less than the detection limit =
                                                                 22
                                                                 20
                                                                 1.2
                                                                 1.2
Sample
Number
           Was the    Reported
           Sample  Concentration
Sample   Collectible?      If
  ID        0 = No    Collectible
            l=Yes
              Is X{ Greater
 Concentration   than Cs?
 Corrected for     l = Yes
Detection Limit    0 = No
                   Xi
                                                                           (Xi)2
11
12
13
14
15
16
17
18
19
20










1
1
1
1
1
1
1
1
1
1
0.5
0.7
2.2
0.7
1.7
2.3
0.3
3.7
0.1
5.6
1.2
1.2
2.2
1.2
1.7
2.3
1.2
3.7
1.2
5.6
0
0
1
0
0
1
0
1
0
1










Total from previous page
Column Totals:
          A   19    I
19.5

B 41
2

C 6


D
                        A = n
                                      B
      •I"
Date Completed:  EXAMPLE

Use additional sheets if necessary.
                                  Completed by
C = r
                   EXAMPLE
                                                         Page
                                 of
Complete WORKSHEET 6 for other chemicals or continue to WORKSHEET 7
                                        B-14

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
 WORKSHEET  6  Data Calculations for a Simple Random Sample, by  Chemical
See Section 6.3 or 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards." Volume 1 _
           SITE:
           Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION 11]
 SAMPLE AREA:                       1. Field used for storing batteries
     CHEMICAL:
                  NUMBER(j) AND DESCRIPTION [2]
                                  2. Chemical #2
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
             Maximum Sample Size from Worksheets 4 and 5 = Sample Size =
                                                Cleanup standard[2] = Cs
                                               Method Detection Limit: =
          Concentration used when it is reported as less than the detection limit =
Sample
Number
           Was the    Reported                  Is xj Greater
           Sample  Concentration  Concentration   than Cs?
Sample   Collectible?      If       Corrected for     l = Yes
  ID        0 = No    Collectible    Detection Limit    0 = No
            1 = Yes                      x;           y.
                                                                 22
                                                                 20
21
22


















Total from previous page
Column Totals:
1
1








1.3
1.8








1.3
1.8









19


A 21

41
0
0









6

B 44.1
C 6













D
A = n B = Yx: C = r D = Y(xi)2
Date Completed:  EXAMPLE
Use additional sheets if necessary.
                                  Completed by       EXAMPLE
                                                         Page	of.
Complete WORKSHEET 6 for other chemicals or continue to WORKSHEET 7
                                        B-15

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
       WORKSHEET 7   Inference for Simple Random Samples by Chemical

See Section 6.3 or 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
  Former XYZ Disposal Site
  SAMPLE AREA:
                  NUMBER(g) AND DESCRIPTION [1J
                   1. Field used for storing batteries
     CHEMICAL:
                  NUMBERQ) AND DESCRIPTION [2]
                         2. Chemical #2
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.


Testing  the Mean                                [2]

                                                     [2]

                             Number of Collectible Samples [6] = n =

              Total of the concentration measurements [6] =

                                    Total for Xi2 [6] =
                                                        B   _
                                     Mean concentration = — = x =
 Standard Deviation of the Data = *\l  "^  = s =

        Degrees of Freedom for s = n - 1 = df =

                                   tl-cc.df =
                                       c
Standard Error for the Mean concentration = -:=•=
                                      Vn

                                  n  = |iija=
      Upper One Sided Confidence Interval = x + tj_a
         If M-ua< Cs then circle Clean, otherwise circle Dirty:
                Based on the mean concentration, the sample area is:
=
=
=
	 	
r=
r =
a=
.05
2
21
44.1







Clean Dirty
Testing Percentiles                              [2]    P0
                                                   [4 or 5] Zj.a
        Number of Samples with Concentration Greater than Cs [6] = r

                       Proportion of Contaminated Samples = — =  p
                   Standard Error for the Proportion =
      Test Statistic = p +
                                                         = U   =
                                                       .5
                                                      1.645
                                                                           .286
                                                     .0986
                                                                           .465
          If UL < PQ then circle Clean, otherwise circle Dirty:
 Based on the proportion of contaminated samples, the sample area is:
                                             Clean    Dirty
Date Completed:  EXAMPLE


Complete WORKSHEET 7 for other chemicals
                         Completed by
EXAMPLE
                                                Page
              of
                                         B-16

-------
                        APPENDIX B: EXAMPLE WORKSHEETS



                      WORKSHEET  2  Attainment Objectives

See Section 3.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
            SITE:
Former XYZ Disposal Site
                   NUMBER(g) AND DESCRIPTION [1]
  SAMPLE AREA:	1. Old Lagoon Area
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.

Sample Collection Procedures to be used (attach separate sheet if necessary):	
For example: One foot soil core, 2 inches in diameter, etc.
 Probability of mistakenly declaring the site clean = a =  .05
 Chemical
to be tested      Chemical
 Number         Name
     j
                         Cleanup
                         Standard
                       (with units)
                           Cs
  Parameter to test:
 Mean
Yes/No
Proportion
    PO
1




Chemical #3




30




Yes




25%




Secondary Objectives/ Other purposes for which the data is to be collected:
Use the Chemical Number (j) to refer on other sheets to the chemical described above.
Attach documentation describing the lab analysis procedure for each chemical.

Date Completed:  EXAMPLE                   Completed by      EXAMPLE

Use additional sheets if necessary.                                        Page	of

Continue to WORKSHEET  3
                                         B-17

-------
                       APPENDIX B: EXAMPLE WORKSHEETS


              WORKSHEET  3  Sampling Design  and Analysis Plan

See Chapter 4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
           SITE:
Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [ 1 ]
 SAMPLE AREA:	1. Old Lagoon Area
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.

Sample Design:            |    | simple Random

                               Systematic Random Sample

                               Stratified Sample
   X
Chemical Comments on the Prob of Type n error Alternate Parameter value
to be tested Sample Design and Chance of concluding the for the specified (3
Number [2] Analysis Plan site is dirty when it is clean Mean Proportion
j P Hi Pi
3




Use non-parametric estimation of
proportions



.20




15




10%




Date Completed:  EXAMPLE
Use additional sheets if necessary.
                       Completed by
EXAMPLE
                                              Page
              of
Continue to WORKSHEET 4 for random or systematic sampling and WORKSHEET 8 for stratified sampling.
                                         B-18

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
            WORKSHEET  8  Definition of Strata Within Sample Area

See Section 4.1 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
Former XYZ Disposal Site
                  NTJMBER(g) AND DESCRIPTION [1]
 SAMPLE AREA:	2. Old Lagoon Area
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
 Stratum       Describe the stratum and the reason
 Number             for interest in this area
                                  Volume =
                            Surface Area * Sample depth

                                     vh
                                                                                   Vh
1
2



Center of Lagoon
Edge of Lagoon



Total Volume = Zvn =
141,000 cu. ft.
94,000 cu. ft.




235,000 cu. ft.
.60
.40




Use the Stratum Number (h) to refer on other worksheets to the stratum described above
Attach a map showing the stratum within the sample area.

Date Completed:  EXAMPLE                   Completed by      EXAMPLE

Use additional sheets if necessary.                                       Page	of

Continue to WORKSHEET 9
                                         B-19

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
  WORKSHEET 9  Desired Sample Sizes for Testing the Mean Using Stratified
                              Sampling, by Chemical
See Section 6.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1 ]
 SAMPLE AREA:	2. Old Lagoon Area
     CHEMICAL:
                  NUMBER(j) AND DESCRIPTION [2]
                   Chemical #3
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
                                                                From z -Table, Appendix A
Probability of mistakenly declaring the site clean [2] = cc .05

For the Cleanup Standard = Cs = 30
Probability of mistakenly declaring the site dirty [3] = (} = .20
(
Proportion
of Sample
Stratum Area in
Number[8] Stratum[8]
h Wh
1
2




.6
.4




If the true concentration is [3] = |ii = 15
Calculate:
Stratum
Standard
Deviation
35
22



C/-i_i.,_
— coiun
B
o 2
J Cs - |ii lz
1 1 =
Unit
Sample
Cost
1
1



in Sum =
21
8.8



29.8
: A ~ 36.38
zl-a =
2
,,=
Desired final
sample size
nhd =
VcJT
21
8.9





= C/A =
B .819


VciT
17.2
7.21




A =
1.645

.842
Calculation
check
%
25.64
10.74



36.38

Date Completed:  EXAMPLE
Use additional sheets if necessary.
Continue to WORKSHEET 10
                    Completed by
EXAMPLE
                                          Page
              of
                                        B-20

-------
                        APPENDIX B: EXAMPLE WORKSHEETS
WORKSHEET  10  Desired Sample Sizes for Testing a Percentile Using Stratified
                                Sampling, by Chemical
See Section 7.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
            SITE:
                          Former XYZ Disposal Site
                   NUMBER(g) AND DESCRIPTION [1]
  SAMPLE AREA:	2.  Old Lagoon Area
                   NUMBERS) AND DESCRIPTION [2]
     CHEMICAL:                       3. Chemical #3
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
 Probability of mistakenly declaring the site clean [2] = a
      Proportion Exceeding Cleanup Standard [2] = PQ =
Probability of mistakenly declaring the site dirty [3] = (3 =
                     If the true proportion is [2] = Pj =
                      Calculate:
                                       P    P
                                       P0'P1
                                                          .05
                                                          .25
                                                           .2
                                                           .1
                                                         .00364
                                                                   From z -Table, Appendix A
                                                                       zl-a =

                                                                       Z1-P =
        Proportion Proportion  Stratum
        of Sample  of dirty   Standard   Unit
Stratum    Area in    Samples   Deviation Sample
Number[3] Stratum[3]             &K     Cost
                                                                 Desired final
                                                                  sample size
                                                                              Calculation
                                                                                check
           Wh
                     Ph
                                                                                 "h
1
2



.6
.4



.145
.036



.352
.184



1
1



C = Column Sum =
B = C
.211
.074



.285
Divided
by A =
/A =
B 78.3
.211
.074



16.54
6.76



A
Date Completed: EXAMPLE
Use additional sheets if necessary.
Continue to WORKSHEET 11
                                               Completed by
                                                                   EXAMPLE
                                                                       Page
                                                                                 of
                                          B-21

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
   WORKSHEET  11  Desired Sample Sizes for All Chemicals and Parameters

See Section 6.4 or 7.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
                          Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1]
 SAMPLE AREA:	2. Old Lagoon Area
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
              Enter the Desired Sample Size (nncj)by Stratum for each combination of chemical
                       and parameter to be tested, from WORKSHEETS 9 and 10.
Desired Sample Size
                                     Stratum number h
                          12345
by stratum and:
chemical
Chemical 3
Worksheet 9
Chemical 3
Worksheet 10
Chemical 	
Worksheet 	
Chemical 	
Worksheet 	
Chemical 	
Worksheet 	
Maximum n^
for all Chemicals
and Parameters
nhdmax
Fraction of
Collectible
Field Samples Rf,
A nhdmax
A~ Rh
A Rounded up to the
Next Integer = n^f ;
the field sample size
17.2
16.54



17.2
.95
18.1
19
7.2
5.75



7.2
.95
7.6
8



























Date Completed:  EXAMPLE

Use additional sheets if necessary.

Continue to WORKSHEET 12
                                             Completed by
EXAMPLE
                                                                    Page
              of
                                        B-22

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
         WORKSHEET  12  Data Calculations, by Stratum and Chemical

See Section 6.4 or 7.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
           SITE:
                   Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1]
 SAMPLE AREA:	2. Old Lagoon Area
                  NUMBER AND DESCRIPTION [3]
      STRATUM:	1. Center of Lagoon
                  NUMBER(j) AND DESCRIPTION [2]
     CHEMICAL:                       3. Chemical #3
Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
                                         Samples Size [11] = nnf =

                                       Cleanup standard[2] = Cs =

                                        Method Detection Limit: =

               Concentration used when no concentration is reported =
                       Was the    Reported
                       Sample   Concentration
Sample     Sample   Collectible?      If
Number       ID        0 = No    Collectible
   i                     l = Yes
                                                     Is Xj Greater
                                       Concentration   than Cs?
                                       Corrected for    l = Yes
                                      Detection Limit    0 = No
                                                         Yi
                                                                             19
                                                                             30
                                                                            (xi)
1
2
3
4
5
6
7
8
9
10
9301
9302
9303
9304
9305
9306
9307
9308
9309
9310
Total from previous page
Column Totals:
1
1
0
1
1
1
1
1
1
1
17
40
#NA
19
31
0
7
10
15
21




A 9

17
40
0
19
31
5
7
10
15
21
0
1
0
0
1
0
0
0
0
0



B 165


C 2
289
1600
0
361
961
25
49
100
225
441



D 4051
                          = A
                                      Completed by
Date Completed:  EXAMPLE

Use additional sheets if necessary.

Complete WORKSHEET 12 for other chemicals or to WORKSHEET  13
                                                       rh=C
                                                                 EXAMPLE
                                                                     Page
                                                                       of
                                        B-23

-------
                       APPENDIX B: EXAMPLE WORKSHEETS


         WORKSHEET  12  Data Calculations, by Stratum and Chemical

See Section 6.4 or 7.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
           SITE:
    Former XYZ Disposal Site
                  NUMBER® AND DESCRIPTION [1]
 SAMPLE AREA:	2. Old Lagoon Area
                  NUMBER AND DESCRIPTION [3]
      STRATUM:	1. Center of Lagoon
                  NUMBER(j) AND DESCRIPTION [21
     CHEMICAL:                       3. Chemical #3
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
                                                 Samples Size [11] = nhf =

                                               Cleanup standard[2] = Cs =

                                                Method Detection Limit: =

                       Concentration used when no concentration is reported =
                                                        19
                                                       30
                       Was the    Reported                  Is xj Greater
                       Sample  Concentration   Concentration   than Cs?
Sample      Sample   Collectible?      If        Corrected for    l = Yes
Number       ID        0 = No    Collectible    Detection Limit    0 = No
   i                     1 = Yes                       x;
                                                      (Xi)
11
12
13
14
15
16
17
18
19

9311
9312
9313
9314
9315
9316
9317
9318
9319

1
1
1
1
1
0
1
1
1

18
27
12
28
94
#NA
13
22
23

18
27
12
28
94
0
13
22
23

0
0
0
0
1
0
0
0
0

324
729
144
784
8836
0
169
484
529

Total from previous page


Column Totals:
A   17
165
2

B 402
C 3
405

D 16050
                       nh = A
                                         rh =
Date Completed:  EXAMPLE
Use additional sheets if necessary.
                        Completed by
EXAMPLE

    Page	
                                                         of
Complete WORKSHEET 12 for other chemicals or to WORKSHEET 13
                                         B-24

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
         WORKSHEET  12  Data Calculations, by Stratum and Chemical

See Section 6.4 or 7.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume. 1
           SITE:
                   Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1]
 SAMPLE AREA:	2. Old Lagoon Area
                  NUMBER AND DESCRIPTION [3]
      STRATUM:	2. Edge of Lagoon
                  NUMBER(j) AND DESCRIPTION [2]
     CHEMICAL:                       3. Chemical #3
Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
                                         Samples Size [11] = %f =

                                       Cleanup standard[2] = Cs =

                                        Method Detection Limit: =

               Concentration used when no concentration is reported =
                                                                             19
                                                                             30
                       Was the   Reported                   Is xj Greater
                       Sample   Concentration   Concentration   than Cs?
Sample     Sample   Collectible?      If        Corrected for    l = Yes
Number       ID        0 = No    Collectible    Detection Limit   0 = No
   i                     1 = Yes                      Xj           yj
                                                                    (Xi)"
1
2
3
4
5
6
7
8
9

9320
9321
9322
9323
9324
9325
9326
9327
9328

Total from previous page
1
1
0
1
1
1
1
1
1

9
16
#NA
0
0
5
8
10
9

9
16
0
5
5
5
8
10
9



1
0
0
0
0
0
0
0
0
0

81
256
0
25
25
25
64
100
81




Column Totals:
                   8
                       nn = A
                                                     67
|c    o   |D
657
                                                       rh =
                                      Completed by
Date Completed:  EXAMPLE
Use additional sheets if necessary.

Complete WORKSHEET 12 for other chemicals or to WORKSHEET 13
                                                                 EXAMPLE
                                                                     Page.
                                                                       of
                                        B-25

-------
                       APPENDIX B: EXAMPLE WORKSHEETS


     WORKSHEET  13  Sample Area Analysis for the Mean Using Stratified
                              Sampling, by Chemical
See Section 6.4 in " Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
                   Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1 ]
 SAMPLE AREA:	2. Old Lagoon Area
                  NUMBERS) AND DESCRIPTION [21
     CHEMICAL:                      3. Chemical #3
Stratum
Number[3]  [3]


   h      Wh
Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.


            [12]


           % = ,
   [12]       [12]

-  =B_    , ^ D-xh2A
Whxh    F =
                                                                           (nh-D
1
2



0.6
0.4



17
8



23.65
8.38



Grand Totals:
408.99
13.70





14.19
3.35




G 17.54
8.661
0.274




H 8.935
4.6883
0.0107




I 4.6990
                                              [2]     cc =
                                             [2]    Cs =

                              Mean concentration = G = "x =

                           H2
       Degrees of Freedom = -r- Rounded to an integer = df =

                                                 tl-a,df =
        Standard Error for the Mean concentration =


  Upper One Sided Confidence Interval = x +
                                                                           .05
                                                                            30
                                                                          17.54
                                                                            17
                                                                           1.74
                                                                           2.99
                                                                          22.74
         If |iua< Cs men circle Clean, otherwise circle Dirty:
                Based on the mean concentration, the sample area is:
                                                          Clean    Dirty
Date Completed:  EXAMPLE
Use additional sheets if necessary.

Continue to   WORKSHEET 14
                                       Completed by
                                      EXAMPLE
                                                             Page
                                                    of
                                         B-26

-------
                       APPENDIX B: EXAMPLE WORKSHEETS
    WORKSHEET  14  Sample Area Analysis for a Percentile Using Stratified
                               Sampling,  by Chemical
See Section 7.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
Former XYZ Disposal Site
                  NUMBER(g) AND DESCRIPTION [1]
 SAMPLE AREA:                        2. Old Lagoon Area
     CHEMICAL:
                  NUMBER(j) AND DESCRIPTION [2]
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
Stratum
Number[3]    [3]
                         [9]
            [9]
[9]
                                                              p=Whph
1
2



Grand 1
0.6
0.4



17
8



0.18
0.00



totals:
0.1453
0



0.1059
0



0.00308
0





G 0.1059
H .00308
                                                     [2]
                                                    [4 or 5]
                       Proportion of Contaminated Samples = G = p =
                        Standard Error for the Proportion =

                                    Test Statistic = p+ sp zi_a = T =

           If T < PQ then circle Clean, otherwise circle Dirty:
                Based on the mean concentration, the sample area is:
» =
,
.25
1.645
.106
.055
.197
Clean Dirty
Date Completed:  EXAMPLE
Use additional sheets if necessary.
                    Completed by
              EXAMPLE
                                           Page
                            of
                                         B-27

-------

-------
           APPENDIX C:  BLANK WORKSHEETS
         The worksheets in this appendix may be used or modified to document the
decisions, record data, and make calculations to determine if the waste site attains the
cleanup standard. These worksheets are referred to in the document.  Appendix B provides
examples of how to fill out the worksheets.
                                     C-l

-------
                        APPENDIX C: BLANK WORKSHEETS



                          WORKSHEET  1  Sample Areas

See Section 3.1 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
           SITE:
Sample
 Area
Number
   g
Describe the sample areas and the reasons for treating each area separately.
Use the Sample Area Number (g) to refer on other sheets to the sample areas described above.
Attach a map showing the sample areas within the waste site.

Date Completed:	                  Completed by	

Use additional sheets if necessary.                                       Page	of.
Continue to WORKSHEET 2
                                         C-2

-------
                        APPENDIX C: BLANK WORKSHEETS

                      WORKSHEET  2  Attainment Objectives
See Section 3.3 in "Methods for Evaluating the Attainment of Cleanup Standards." Volume 1
            SITE:
 SAMPLE AREA:
                  NUMBER(g) AND DESCRIPTION [1J
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
Sample Collection Procedures to be used (attach separate sheet if necessary):	
Probability of mistakenly declaring the site clean = a =


Chemical Cleanup Parameter to test:
to be tested Chemical Standard
Number Name (with units) Mean Proportion
j Cs Yes/No P0

























Secondary Objectives/ Other purposes for which the data is to be collected:
Use the Chemical Number (j) to refer on other sheets to the chemical described above.
Attach documentation describing the lab analysis procedure for each chemical.
Date Completed:	                  Completed by	
Use additional sheets if necessary.                                        Page.
of
Continue to WORKSHEET 3
                                          C-3

-------
                        APPENDIX C: BLANK WORKSHEETS

              WORKSHEET 3  Sampling Design  and Analysis Plan
See Chapter 4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
 SAMPLE AREA:
                  NUMBER(g) AND DESCRIPTION [1]
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
Sample Design:            |    | Simpl{, Random Sample
                               Systematic Random Sample
                               Stratified Sample
Chemical Comments on the Prob of Type n error Alternate Parameter value
to be tested Sample Design and Chance of concluding the for the specified (3
Number [2] Analysis Plan site is dirty when it is clean Mean Proportion
j P m PI

























Date Completed:	
Use additional sheets if necessary.
Completed by
                       Page
of
Continue to WORKSHEET 4 for random or systematic sampling and WORKSHEET 8 for stratified sampling.
                                          C-4

-------
                         APPENDIX C: BLANK WORKSHEETS
    WORKSHEET  4   Sample Size for Testing the Mean Using Simple Random
                                      Sampling
See Section 6.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
  	If the mean concentration is not to be tested for this chemical, continue to WORKSHEET 5
            SITE:
  SAMPLE AREA:
                   NUMBER(g) AND DESCRIPTION [1]
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
Probability of mistakenly declaring the site clean [2] = a



                              [2]        [3]
                 Appendix A

   j      P        z\.n       Cs        \LI         !
                                                                  From z -Table, Appendix A
                                                              ]      *•«-[
Chemical
Number   [3]
  [2]
  From
  z table
Appendix A
                                                           A=l
Calculate:


  Cs-m
                              Fraction of samples expected to be analyzable = R =


                                                              Max nj   _
                                                              	R   =B =

                      B rounded up = Sample Size for  Testing Means = nf =
Date Completed:	
Use additional sheets if necessary.

Continue to WORKSHEET 5
                             Completed by



































Column Maximum, Max n: =






                                                    Page	of.
                                         C-5

-------
                        APPENDIX C: BLANK WORKSHEETS
  WORKSHEET 5  Sample Size for Testing Proportions Using Simple Random
                                      Sampling
See Section 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
If the mean concentration is not to be tested for this chemical,  continue to WORKSHEET 6	
            SITE:
 SAMPLE AREA:
                  NUMBER(g) AND DESCRIPTION [1]
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
                                                                  From z -Table, Appendix A
  Probability of mistakenly declaring the site clean [2] = a

Chemical         From                  Calculate:
Number  [3]     z table    [2]       [3]
   [2]

   j      P       ZI-B     PO       PI    A = ZI-
                                                                      zi-ct=[
                               Fraction of samples expected to be collectible = B =
                                                              Max nj  _ r _
                                                                  B    ~^~
     C rounded up to the next integer = Sample Size for Testing Proportions =



































Column Maximum, Max n: =






Date Completed:	

Use additional sheets if necessary.

Continue to WORKSHEET 6
                                               Completed by
                                                                       Page	of.
                                          C-6

-------
                        APPENDIX C: BLANK WORKSHEETS
 WORKSHEET  6  Data Calculations for a Simple Random Sample, by Chemical
See Section 6.3 or 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards." Volume 1	
           SITE:
 SAMPLE AREA:
                  NUMBER(g) AND DESCRIPTION [1J
     CHEMICAL:
                  NUMBER(j) AND DESCRIPTION [2]
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
             Maximum Sample Size from Worksheets 4 and 5 = Sample Size =
                                                Cleanup standard [2] = Cs
                                                Method Detection Limit: =
   Concentration used when it is reported as less than the method detection limit=

                       Was the   Reported                   Is xj Greater
                       Sample   Concentration   Concentration   thanCs?
Sample Sample Collectible? If Corrected for l = Yes
Number ID 0 = No Collectible Detection Limit 0 = No
i l=Yes Xj yj (xj)2




















Total from previous page
Column Totals:


























































A

B
C


D
                        A = n
                                                               C = r
Date Completed:
                                              Completed by
Use additional sheets if necessary.                                        Page _ of .
Complete WORKSHEET 6 for other chemicals or continue to WORKSHEET 7
                                         C-7

-------
                        APPENDIX C: BLANK WORKSHEETS
      WORKSHEET  7   Inference for Simple Random Samples by Chemical


See Section 6.3 or 7.3 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
 SAMPLE AREA:
                  NUMBER(g) AND DESCRIPTION [1]
     CHEMICAL:
                  NUMBER(j) ANDDESCRIPnON[2]
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.




Testing  the Mean                                 [2]      
-------
                        APPENDIX C: BLANK WORKSHEETS
            WORKSHEET  8   Definition of Strata Within Sample Area

See Section 4.1 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
           SITE:
 SAMPLE AREA:
                  NUMBER(g) AND DESCRIPTION [1]
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
 Stratum       Describe the stratum and the reason
 Number             for interest in this area
      Volume =
Surface Area * Sample depth

          Vh
                                                                                   Vh

                                                                                   Svh
                                  Total Volume =
Use the Stratum Number (h) to refer on other worksheets to the stratum described above
Attach a map showing the stratum within the sample area.

Date Completed:	                   Completed by	

Use additional sheets if necessary.

Continue to WORKSHEET 9
               Page.
of
                                          C-9

-------
                         APPENDIX C: BLANK WORKSHEETS
  WORKSHEET 9  Desired Sample Sizes for Testing the Mean  Using Stratified
                                Sampling, by Chemical
See Section 6.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
            SITE:
 SAMPLE AREA:
                   NUMBER(g) AND DESCRIPTION!!]
     CHEMICAL:
                   NUMBER® AND DESCRIPTION [2]
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
  Probability of mistakenly declaring the site clean [2] = a

                       For the Cleanup Standard = Cs =

 Probability of mistakenly declaring the site dirty [3] = ji =

                    If the true concentration is [3] = Ui =
                 Calculate:

         Proportion
         of Sample  Stratum
Stratum    Area in    Standard
Number[8] Stratum[8] Deviation

   h       Wh       &h
 Unit
Sample
 Cost

  Ch
                  C = Column Sum =


                           B = C/A =

Date Completed:	

Use additional sheets if necessary.

Continue to WORKSHEET 10
                                                                    From z -Table^ Appendix A
              Desired final
              sample size
                 nhd =
              B*Wh*6-h
Completed by.
Calculation
  check
                           A =
                        Page.
   of
                                           C-10

-------
                         APPENDIX C: BLANK WORKSHEETS
WORKSHEET 10  Desired Sample Sizes for Testing a Percentile  Using Stratified
                                Sampling, by  Chemical

See Section 7.4 in "Methods for Evaluating the Attainment of Cleanup Standards." Volume 1	
            SITE:
  SAMPLE AREA:
                   NUMBER(g) AND DESCRIPTION [lj
     CHEMICAL:
                   NUMBER(j) AND DESCRIPTION [2]
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
  Probability of mistakenly declaring the site clean [2] = a

       Proportion Exceeding Cleanup Standard [2] = PQ =

 Probability of mistakenly declaring the site dirty [3] = [3 =

                      If the true proportion is [2] = PI =
                 Calculate:
                                   po-pi
= A =
         Proportion Proportion   Stratum
         of Sample  of dirty    Standard    Unit
 Stratum   Area in   Samples  Deviation Sample
Number[3] Stratum[3]             &n     Cost
                     Ph
                         = Column Sum =
                              B = C/A =
                    From z -
                                                                               Appendix A
                 Desired final
                  sample size
                    "hd =
                  B*Wh*o-h
Calculation
  check
                          A =
Date Completed:	

Use additional sheets if necessary.

Continue to WORKSHEET 11
Completed by.
                        Page
                                   nh
   of
                                          C-ll

-------
                        APPENDIX C: BLANK WORKSHEETS
   WORKSHEET  11  Desired Sample Sizes for AH Chemicals and Parameters

See Section 6.4 or 7.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1  	
           SITE:
 SAMPLE AREA:
                  NUMBER(g) AND DESCRlPnON[lJ
       Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
              Enter the Desired Sample Size (nj^by Stratum for each combination of chemical
                       and parameter to be tested, from WORKSHEETS 9 and 10.
Desired Sample Size
                                     Stratum number h
                          12345
by stratum and:
chemical
Chemical 	
Worksheet 	
Chemical 	
Worksheet 	
Chemical 	
Worksheet 	
Chemical 	
Worksheet 	
Chemical 	
Worksheet 	
Maximum n^
for all Chemicals
and Parameters
nhdmax
Fraction of
Collectible
Field Samples Rn
A nhdmax
A %
A Rounded up to the
Next Integer = nnf 5
the field sample size













































Date Completed:	

Use additional sheets if necessary.

Continue to WORKSHEET 12
                                              Completed by.
                                                                     Page
of
                                        C-12

-------
                        APPENDIX C: BLANK WORKSHEETS
          WORKSHEET  12  Data Calculations, by Stratum and Chemical
See Section 6.4 or 7.4 in " Methods for Evaluating the Attainment of Cleanup Standards," Volume 1
            SITE:
  SAMPLE AREA:
                  NUMBER(g) AND DESCRIPTION [1 ]
      STRATUM:
                  NUMBER AND DESCRIPTION [3]
     CHEMICAL:
                  NUMBER(j) AND DESCRIPTION [2]
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
                                                 Samples Size [11] = n^f =
                                               Cleanup standard[2]  = Cs =
                                                Method Detection Limit: =
                       Concentration used when no concentration is reported =
                       Was the    Reported                   Is Xj Greater
                       Sample  Concentration  Concentration   thanCs?
                       nh = A
Date Completed:
Completed by
Use additional sheets if necessary.
Complete WORKSHEET 12 for other chemicals or to WORKSHEET 13
                       Page
Sample Sample Collectible? If Corrected for l = Yes
Number ID 0 = No Collectible Detection Limit 0 = No
i l = Yes Xj yi (xO2




















Total from previous page
Column Totals:












































A




B
C













D
of
                                        C-13

-------
                        APPENDIX C: BLANK WORKSHEETS
     WORKSHEET 13   Sample Area Analysis for  the Mean Using Stratified
                               Sampling, by Chemical
See Section 6.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
            SUE:
 SAMPLE AREA:
                  NUMBER(g) AND DESCRIPTIONil J
     CHEMICAL:
                  NUMBERQ AND DESCRIPTION [2]
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
Stratum
Number[3]   [3]

   h      Wh
[12]       [12]       [12]

      - _B    ,  D-xh2A
                                Whxh    F =
                                                                             (A-l)
Grand Totals:
                                                             H
                                                       [2]     cc =    [
                                                      [2]    Cs =    f
                                      Mean concentration = G = x =

                                   H2
               Degrees of Freedom = -r-  Rounded to an integer = df =

                                                          tl-a,df =
                Standard Error for the Mean concentration =
          Upper One Sided Confidence Interval = x + s^ ti_a>(jf = |iua =
          If |iTja< Cs then circle Clean, otherwise circle Dirty:
                Based on the mean concentration, the sample area is:
Date Completed:	                  Completed by	
Use additional sheets if necessary.

Continue to   WORKSHEET 14
                                               Clean    Dirty
                                                  Page
of
                                          C-14

-------
                        APPENDIX C: BLANK WORKSHEETS
    WORKSHEET  14   Sample Area  Analysis for a Percentile Using Stratified
                               Sampling, by Chemical
See Section 7.4 in "Methods for Evaluating the Attainment of Cleanup Standards," Volume 1	
            SITE:
  SAMPLE AREA:
                   NUMBER(g) AND DESCRIPTION [Ij
     CHEMICAL:
                  NUMBERfj) AND DESCRIPTION [2]
        Numbers in square brackets [] refer to the Worksheet from which the information may be obtained.
 Stratum
Number[3] [3] [9] [9] [9]
C 2 2 Sh2
nh





Grand 1






























"otals:
G
H
                                                      [2]    P0 =
                                                    [4 or 5] Z!.a =
                       Proportion of Contaminated Samples = G = p =
            = sp =
                        Standard Error for the Proportion =
                                    Test Statistic =  p+ sp zi_a = T =
            If T < PQ then circle Clean, otherwise circle Dirty:
                Based on the mean concentration, the sample area is:
                    Clean    Dirty
Date Completed:	
Use additional sheets if necessary.
Completed by
                       Page	of.
                                         C-15

-------

-------
                     APPENDIX  D:  GLOSSARY

Alpha (a) In the context of a statistical test, a is probability of a Type I error.

Alternative Hypothesis  See hypothesis.

Analysis Plan The plan specifying how the data are to be analyzed once they are
          collected, including what estimates are to be made from the data* how the
          estimates are to be calculated, and how the results of the analysis will be
          reported.

Attainment The achievement of a prescribed standard/level of concentration.

Attainment Objectives  Specifying chemicals to be tested, specifying the cleanup
          standard to be attained, specifying the measure or parameter to be compared to
          the cleanup standard, and specifying the level of confidence required if the
          environment and human health are to be protected

Beta (P)  In the context of a statistical test, p is probability of a Type II error.

Binomial Distribution A probability distribution used to describe the number of
          occurrences of a specified event in n independent trials. In this manual, the
          binomial distribution is used to develop statistical tests concerned with testing
          the proportion of soil units in a simple random sample having excessive
          concentrations of a contaminant (see Chapter 7). For additional details about
          the binomial distribution, consult Conover (1980).

Coefficient of Variation The  ratio of the standard deviation to the mean for a set of
          data or distribution, abbreviated cv. For data that can only have positive
          values, such as concentration measurements, the coefficient of variation
          provides a crude measure  of skewness.

Confidence Interval A sample-based estimate of a population  parameter expressed as a
          range or interval of values, rather than as a single value (point estimate).
                                      D-l

-------
                           APPENDIX D: GLOSSARY
Confidence Level The degree of confidence associated with an interval estimate. For
          example, with a 95% confidence interval, we would be 95% certain that the
          interval contains the true value being estimated. The confidence level is equal to
          1 minus the Type I error (false positive rate).

Conservative Test A statistical test for which the Type I error rate (false positive rate) is
          actually less than that specified for the test.  For a conservative test there will be
          a greater tendency to accept the null hypothesis when it is not true than for a
          non-conservative test.

Distribution The frequencies (either relative or absolute) with which measurements in a
          data set fall within specified classes. A graphical display of a distribution is
          referred to as a histogram.

Estimate Any numerical quantity computed from a sample of data. For example, a
          sample mean is an estimate  of the corresponding population mean.

False Positive Rate The probability of mistakenly concluding that the sample area is
          clean when it is duty.  It is the probability of making a Type I error.

False Negative Rate The probability  of mistakenly concluding that the sample area is
          dirty when it is clean.  It is the probability of making a Type II error.

Geostatistics A  methodology for the analysis of spatially correlated data. The
          characteristic feature is the use of variograms or related techniques to quantify
          and model the spatial correlation structure.  Also includes the various techniques
          such as kriging, which utilize spatial correlation models.

Histogram  A graphical display of a frequency distribution.

Hot Spot Localized elliptical areas with concentrations in excess of the cleanup standard,
          either a volume defined by the projection of the surface  area through the soil
          zone that will be sampled or a discrete horizon within the soil zone that will be
          sampled.
                                        D-2

-------
                           APPENDIX D: GLOSSARY
Hypothesis  An assumption about a property or characteristic of a population under
          study. The goal of statistical inference is to decide which of two
          complementary hypotheses is likely to be true. In the context of this guidance
          document, the null hypothesis is that the sample area is "dirty" and the
          alternative hypothesis is that the sample area is "clean."

Inference The process of generalizing (extrapolating) results from a sample to a larger
          population.

Judgment sample A sample of data selected according to non-probabilistic methods.

Kriging  A weighted-moving-average interpolation method where the set of weights
          assigned to samples minimizes the estimation variance, which is computed as a
          function of the variogram model and locations of the samples relative to each
          other, and to the point or block being estimated.  This technique is used to
          model the contours of contamination levels at a waste site (see Chapter 10).

Less-Than-Detection-Limit A concentration value that is below the detection limit. It
          is generally recommended that these values be included in the analysis as values
          at the detection limit

Lognormal Distribution A family of positive-valued, skewed distributions commonly
          used in environmental work. See Gilbert (1987, p. 152) for a detailed
          discussion of lognormal distributions.

Mean  The arithmetic average of a set of data values. Specifically, the mean of a data set,
                                        n
          \i,  x2,..., Xn, is defined by  x = £ xj/n.
                                        i=l

Median  The "middle" value of a set of data, after the values have been arranged in
          ascending order. If the number of data points is even, the median is defined to
          be the average of the two middle values.

Nonparametric Test A test based on relatively few assumptions about the underlying
          process generating the data. In particular, few assumptions are made about the
          exact form of the underlying probability distribution.  As a consequence,
          nonparametric tests are valid for a fairly broad class of distributions.
                                       D-3

-------
                           APPENDIX D:  GLOSSARY
Normal Distribution A family of "bell-shaped" distributions described by the mean and
          variance, p. and a2.  Refer to a statistical text [(e.g., Sokal and Rohlf (1973)]
          for a formal definition. The standard normal distribution has u, = 0 and a2 = 1.

Null Hypothesis  See hypothesis.

Ordinary Kriging A variety of kriging which assumes that local means are not
          necessarily closely related to the population mean, and which therefore uses
          only the samples in the local neighborhood for the estimate.  Ordinary kriging is
          the most commonly used method for environmental situations.

Outlier A measurement that is extremely large or small relative to the rest of the data
          gathered and that is suspected of misrepresenting the true concentration at the
          sample location.

Parameter A statistical property or characteristic of a population of values. Statistical
          quantities such as means, standard deviations, percentiles, etc. are parameters if
          they refer to a population of values, rather than to a sample of values.

Parametric Test A test based on relatively strong assumptions about  the underlying
          process generating the data. For example, most parametric tests  assume that the
          underlying data are normally distributed.  As a consequence, parametric tests
          are not valid unless the underlying assumptions are met See robust test.

Percentile The specific value of a distribution that divides the set of measurements in
          such a way that P percent of the measurements fall below (or are equal to) this
          value, and 1-P percent of the measurements exceed this value. For specificity,
          a percentile is described by the value of P (expressed as a percentage). For
          example, the 95th percentile (P=0.95) is that value X such that 95 percent of the
          data have values less than X, and 5 percent have values exceeding X. By
          definition, the median is the 50th percentile.

Physical sample or soil sample A portion of material (such as a soil core, scoop, etc.)
          gathered at the waste site on which measurements are to be made. This may
          also be called a soil unit.  A soil sample  may be mixed, subsampled, or
          otherwise handled to obtain the sample of soil that is sent for laboratory
          analysis.
                                       D-4

-------
                           APPENDIX D: GLOSSARY
Point Estimate See estimate.

Population The totality of soil units at a waste site for which inferences regarding
           attainment of cleanup standards are to be made.

Power The probability that a statistical test will result in rejecting the null hypothesis
           when the null hypothesis is false.  Power = 1 - p, where p is the Type II error
           rate associated with the test. The term "power function" is more accurate
           because it reflects the fact that power is a function of a particular value of the
           parameter of interest under the alternative hypothesis.

Precision  See standard error.

Proportion The number of soil units in a set of soil units that have a specified
           characteristic, divided by the total number of soil units in the set This may also
           be expressed  as a proportion of area or proportion of volume that has a
           specified characteristic.

Random Sample A sample of soil units selected using the simple random sampling
           procedures described in Chapter 5.

Range The difference between the maximum and minimum values of measurements in a
           data set.

Robust Test A statistical test that is approximately valid under a wide range of
           conditions.

Sample Any collection of soil samples taken from a waste site.

Sample Area The specific area within a waste site for which a separate decision on
           attainment is to be reached.

Sample Design  The procedures used to select the sample of soil units.

Sample Size The number of lab samples (i.e., the size of the statistical sample). Thus, a
           sample of size 10 consists of the measurements  taken on 10 lab samples.
                                       D-5

-------
                           APPENDIX D: GLOSSARY
Sequential Test A statistical test in which the decision to accept or reject the null
          hypothesis is made in a sequential fashion. A sequential test for proportions is
          described in Chapter 8 of this guidance document.

Semi-variogram Identical to the term "variogram." There is disagreement in the
          geostatistical literature as to which term should be used.

Significance Level  The probability of a Type I error associated with a statistical test.
          In the context of the statistical tests presented in this document, it is the
          probability that the sample area is declared to be clean when it is dirty. The
          significance level is often denoted by the symbol a (Greek letter alpha).

Size of the physical sample This term refers to the dimensions of a physical sample or
          soil unit.

Skewed Distribution  Any nonsymmetric distribution.

Soil Sample See physical sample.

Standard Deviation A measure of dispersion of a set of data.  Specifically, given a set
          of measurements, xl5 x2, ..., x^ the standard deviation is defined to be the
          quantity, s =  V	—j	, where x is the sample mean.

Standard Error A measure of the variability (or precision) of a sample estimate.
          Standard errors are often used to construct confidence intervals.

Statistical Sample  A collection of chemical concentration measurements reported by the
          lab for one or more lab samples.

Statistical Test  A formal statistical procedure and decision rule for deciding whether a
          sample area attains the specified cleanup standard.

Stratified Sample A sample comprised of a number of separate  samples from different
          strata.
                                       D-6

-------
                           APPENDIX D: GLOSSARY
Stratum A subset of a sample area within which a random or systematic sample is
          selected.  The primary purpose of creating strata for sampling is to improve the
          precision of the sample design.

Symmetric Distribution A distribution of measurements for which the two sides of its
          overall shape are mirror images of each other about a center line.

Systematic Sample A "grid" sample with a random start position.

Tolerance Interval  A confidence interval around a percentile of a distribution of
          concentrations.

Type I Error The error made when the sample area is declared to be clean when it is
          contaminated.  This is also referred to as a false positive.

Type II Error The error made when the sample area is declared to be dirty when it is
          clean.  This is also referred to as a false negative.

Variance The square of the standard deviation.

Variogram A plot of the variance (one-half the mean squared difference) of paired sample
          measurements as a function of the distance (and optionally of the direction)
          between samples. Typically, all possible sample pairs are examined, distance
          and direction. Variograms provide a means of quantifying the commonly
          observed relationship that samples close together will tend to have more similar
          values than samples far apart.

Waste Site The entire area being investigated for contamination.

Z Value Percentage point of a standard normal distribution.
                                      D-7

-------

-------
                           INDEX OF  KEY  WORDS
alpha, 2-5
analysis plan, 4-6,9-8
ARAR (applicable and relevant or appropriate
  requirements), 1-1,1-7
beta, 2-5
Cs (cleanup standard), 3-5,6-1
compositing methods, 5-15, 6-26
data quality objectives, 3-11
degrees of freedom, 6-3,6-24,6-28
exact test for proportions, 7-9,7-10
false negative, 2-5,2-7
false positive, 2-5, 2-7, 3-11
geostatistical methods, 10-1
grids, 5-5
homogenization, 10-6
hot spots, 9-1
kriging, 10-2
  indicator kriging, 10-9, 10-10
  probability kriging, 10-9,10-10
mean concentration, 2-8,2-9, 3-3,
  3-10, 6-1, 6-16
median, 3-6
null hypothesis, 2-3
outliers, 2-16
parametric procedures, 6-1
proportions or percentiles, 2-9, 2-15,7-1
quality assurance/quality control, 5-18
SARA  (Superfund Amendments and
  Reauthorization Act of 1986), 1-1
sample
  judgment sample, 2-18,4-2
  random sample, 2-18,4-2,4-4, 5-3
  stratified sample, 2-18,4-4
  systematic sample, 2-18,4-2
sample area, 3-1
sample size determination, 2-14,6-7,6-8
  for random samples (mean), 6-7
  for stratified random samples (mean), 6-13,
  for systematic samples (mean), 6-20
  for simple random samples (proportions/
   percentiles), 7-5
  for stratified samples (proportions/
   percentiles) 7-13,7-17
  for a normal or lognormal population
   using tolerance intervals (proportions/
   percentiles), 7-21
sampling, 4-1
  random sampling, 4-1,4-2,7-5
  sequential sampling, 4-2,4-6, 8-1
  stratified sampling, 4-1,4-4,6-12, 7-12
  systematic sampling, 4-1,4-2, 6-21, 6-22,
    6-23
sampling location, 5-1
  random samples, 5-3
  stratified samples, 5-13
  systematic samples, 5-5
sampling plan, 4-1
serpentine pattern, 6-21,6-25,6-26,6-27
simple exceedance rule method, 7-11
soil unit, 2-18
spatial characterization, 1-4
standard deviation, 6-2,6-4
standard error
  estimation of, 6-17, 6-18, 6-21, 6-23,
    6-25,7-7,7-18
standards, 1-6
  background-based, 1-7
  cleanup, 1-6, 3-5
  risk-based, 1-8,2-11,2-12
  technology-based, 1-7
strata, 4-4
subsampling, 5-14
Superfund remediation, 1-6
upper percentile, 3-6
vadosezone, 1-4
                                             IND

-------

-------

-------
K» .
    Cover Photos: McKin Site, Gray, Maine, before and after cleanup
                   Maine Department of Environmental Protection

-------