United States
Environmental Protection
Agency
Environmental Monitoring
Systems Laboratory
PO Box 15027
Las Vegas NV 89114-5027
EPA 600'4-84-043
May 1984
Soil  Sampling
Quality Assurance
User's Guide
Cooperative Agreement
CR 810550-01

-------
                                          EPA-600/4-84-043
              PROJECT SUMMARY
      SOIL SAMPLING QUALITY ASSURANCE
               USER'S GUIDE
                    by


  Delbert S. Earth and Benjamin J. Mason
         vironmental Research Center
         versity of Nevada, Las Vegas
         Las Vegas, Nevada  89154
           Cooperative Agreement
               CR 810550-01
     Kenneth W. Brown, Project Officer
   Exposure Assessment Research Division
Environmental Monitoring Systems Laboratory
    Office of Research and Development
             Las Vegas, Nevada
                 March 1984

-------
         SOIL SAMPLING QUALITY ASSURANCE USER'S GUIDE




                      PROJECT SUMMARY
     An adequate quality  assurance/quality control (QA/QC)



program requires the  identification  and quantification of all



sources  of error  associated  with each  step of  a monitoring



program so  that the resulting data will be of known quality.   The



components of error,  or  variance,  include those associated with



sampling, sample preparation, extraction,  analysis,  and residual



error.  In the past,  major  emphasis  has been placed on  QA/QC



aspects of  sample analysis and closely  associated operations such



as  sample preparation and  extraction.   For monitoring a



relatively inhomogeneous medium  such as  soil the  sampling



component  of variance  will usually significantly exceed the



analysis  component.  Thus, in this case a  minimum adequate QA/QC



plan  must  include a  section dealing with soil sampling.  The



purpose of  this document is  to provide  guidance in QA/QC aspects



related to  soil sampling.







     Generally soil monitoring is  undertaken to carry out the



provisions  and intent of applicable environmental laws with high



                             1

-------
priority requirements associated with hazardous waste management.



The objectives  of  soil monitoring programs  are  often  to obtain



data on the  basis of which to answer one or more of the  following



questions.







     o  Are the concentrations of  specified soil pollutants  in a



        defined study  region  significantly different  from  the



        concentrations  in a control  region?



     o  Do the concentrations of  specified soil pollutants  in a



        defined study  region exceed  established threshold action



        levels?



     o  At  the measured concentrations of  specified  soil



        pollutants in a defined  study  region what is  the



        associated risk of  adverse  effects  to public health,



        welfare, or  the environment?







 For each of these applications the QA/QC required to determine



 precisions and confidence levels for the data cannot be specified



 without giving careful consideration to the consequences of



 making an error, for example,  in a decision to  require, or not to



 require,  cleanup of a contaminated region.  It follows in general



 that to  be maximally  cost-effective  and  defensible  the  QA/QC



 objectives of a soil  monitoring program cannot be separated  from



 the  objectives of  the  soil monitoring program itself.

-------
     Approximately  20  percent  of  the total monitoring program



sample  load should be allocated to QA/QC with about 5 percent of



this being dedicated  to  the analytical effort and 15 percent to



the sampling effort.   Soil  sampling programs must incorporate



statistical designs  and QA/QC plans to provide quantitative



measures of both precision and representativeness.







     Control  samples are  normally  as  important  to a soil



monitoring study as are samples  taken from the study region.  The



data  from control samples aid  in  the interpretation  of the



results from  the study  region and also help  to identify sources



and important  transport  routes  for the soil  pollutants.



Accordingly,  the same level of effort and degree of QA/QC  checks



should go into selecting and sampling  a control  region as goes



into sampling the  study region.







     Experience has  shown that requiring approximately 5 percent



of  all analytical  samples to be duplicate  samples will  provide



adequate QA/QC for  determining  variance  between samples collected



at  approximately the  same  site.   A precision  less than +20



percent is probably unrealistic for a field  soil sampling  effort.



Table  1 provides recommendations  for confidence levels and



precisions  for soil  sampling  related  to  hazardous  waste



investigations.

-------
                           TABLE 1




         ILLUSTRATIVE CONFIDENCE LEVELS  AND  PRECISIONS








                                     Confidence Level  Precision



                                        (Percent)      (Percent)








     Emergency Cleanup Activities           90             20



     Remedial Response Studies              95             20



     Planned Removal Studies              >  95           £ 20








     All  statistical  sampling plans  are based on frequency



distributions with  the most common  being normal or log  normal.



Generally the  concentrations  of  pollutants  in  soil  and



transport-related properties of these pollutants are distributed



log  normally.   In  addition to obtaining information on the areal



distribution  of  soil pollutants it is necessary to determine  the



distribution  with depth.








     Both  Type  I (false positive) and  Type  II  (false negative)



errors should  be considered  in  hypothesis  testing.  Tables  are



provided for  use in determining the required  number of samples to



achieve defined  precision and confidence levels.  The location of



sampling is  important, and a  random process should normally be



used for selecting  specific sampling sites.   Stratification of



the  sampling  region may reduce  the variance in cases where the



variance is considered to be unacceptably large.   Compositing of



                              4

-------
samples is generally not recommended since it allows  no estimate



of the variance  among the samples  being composited.








     Suggested QA/QC procedures  for soil  samples include



preparation of the  following samples, generally on  the basis of



one QA/QC  sample for each 20 samples:  field blank,  sample  bank



blank, reagent blank, calibration  check standard, spiked extract,



spiked  sample,  total recoverable, laboratory control standard,



re-extraction,  split extract, triplicate  sample,  and duplicate



sample.








     The major  technique used to detect bias  in a soil sampling



effort is the adding of known amounts of  standard solutions to



some  of the samples and comparing  the resulting data.   It is



especially difficult to demonstrate the complete absence of bias.
     The confidence interval for soil samples is bounded by  the



 confidence limits (bounds of uncertainty about the average caused



 by  the variability of the experiment).   The confidence interval



 is  used in the development of  control  charts, in  identifying



 outliers,  and in determining  if  a set  of samples exceed some



 established standard.   Generally the analysis of variance of  the



 data provides the  best method  for obtaining the information



 needed  for calculating  the confidence interval.  An approximation



 of  the confidence interval can be  obtained by use of the ranges



                              5

-------
of replicates  in  a  series.  The tolerance limits are  similar to



the confidence  limits but are used to  identify  the interval and



limits into which data from the individual samples should  fall.







     The  simplest  test of hypotheses  is  either comparison of



two  mean  values  or comparison  between  the mean  and  some



established standard,  or action,  value.  The Student's t test is



generally used  for both cases.







     Once objectives  have been  defined  for a soil monitoring



 study,  a total study protocol, including  an appropriate QA/QC



 program must be prepared.  Usually not  enough  is known about  the



 sources and transport  properties  of the  soil pollutants to



 accomplish this in a cost-effective manner without additional



 study.   The suggested approach is to conduct  an  exploratory study



 including both a literature  and  information  search followed by



 selected field measurements  based on an assumed  dispersion model.



 The data  resulting from this  exploratory study  serve as the basis



 for the more definitive  total  study protocol.   If one  is  dealing



 with a  situation requiring possible emergency action to protect



 public  health, it is necessary to compress  the planning and  study



 design  into a short time period  and proceed  to  the  definitive



 study without delay.   In either  case, the objectives of  the



 monitoring study constitute the driving force for all elements of



 the study design including the QA/QC aspects.

-------
     To develop  the  exploratory  study  protocol  with its



associated QA/QC  plan one needs  to combine into  an assumed



dispersion model the information obtained prior to  any  field



measurements.  On  the basis of  this model the standard  deviation



of the mean  for  soil  samples  is estimated.   Value judgments  are



used to define required precision and confidence levels (related



to acceptable  levels  of Type  I or Type II errors).   A control



region is  selected.  The numbers  of required samples  may then  be



calculated.  Additional samples  should be required to provide  for



the  validation  of the assumed model.   The locations of  the



sampling  sites  should be selected by an appropriate combination



of judgmental (use of  the assumed model), systematic  (to  allow



for  the fact  that the  model may  be wrong),  and random  (to



minimize bias)  sampling.  Sampling and sample handling must be



accomplished  according to  standardized  procedures based on



principles designed to achieve  both data of adequate  quality  and



maximal cost-effectiveness.  Particular attention should be given



to  factors surrounding the disposition  of  non-soil  materials



collected with  the soil samples.







      The  requirements for QA/QC for the exploratory  study need



not  be as stringent as for the more definitive study  in the sense



that acceptable  precisions and confidence  levels may be  relaxed



somewhat.   Allowance should be made,  however, for the collection



of   a  modest  additional number  of QA/QC  samples  over that



specified in the QA/QC plan to verify that the QA/QC  study design



                              7

-------
is adequately achieving  its  assigned objectives.  Also, all



normal  analytical QA/QC checks should be used.







     If the exploratory study  is conducted well, it will provide



some data  for  achieving  the overall objectives of  the total



monitoring study;  it will provide a check of  the feasibility and



efficacy of all aspects of the monitoring design including  the



QA/QC  plan;  it will serve  as a training vehicle for  all



participants; it  will pinpoint where additional measurements  need



 to be  made; and it will provide a body  of information and data



 which  can be incorporated  into the final report for  the  total




 monitoring  study.







     For the more definitive study the  selection of  numbers of



 samples and sampling  sites, sample collection  procedures, and



 sample  handling methods  and procedures  follow  and build on the



 principles  discussed  and  results  obtained  in the  exploratory




 study.







      Frequency  of  sampling is an  important  aspect of the more



 definitive study  which usually  cannot be addressed  in the



 exploratory study  because of  the relatively  short time span  over



 which  the exploratory study is  conducted.   The required frequency



 of sampling depends on  the objectives of the study, the sources



 of pollution,  the pollutants of interest, transport rates, and



 disappearance  rates  (physical,  chemical, or  biological



                               8

-------
transformations  as  well  as  dilution or dispersion).  Sampling



frequency  may be  related  to  changes  over time,  season, or



precipitation.   An  approach that has been used  successfully has



been to provide intensive sampling early in the  life of the  study



(e.g., monthly  for the  first year) and then  to decrease the



frequency as  the  levels begin to drop.  The  important principle



is  that the  sampling  should be  conducted  often  enough that



changes in the concentrations of soil pollutants important  to the



achievement of the monitoring objectives are not missed.







     The important  questions to be answered in the  analyses and



interpretation of QA/QC data  are,  "What  is the quality  of the



data?" and "Could the same  objective  have been achieved through



an  improved QA/QC  design which may  have  required   fewer



resources?"  It is desirable to  provide summarized tables of



validated QA/QC data in the  final report.   This approach  allows



users  to verify the reported  results  as well as begin to build a



body of QA/QC experimental  data in the  literature  which allow



comparisons to be made  among studies.   Special emphasis  should be



placed on how overall levels  of precision and confidence were



derived  from the  data.  If portions of the  study  results are



ambiguous and supportable  conclusions cannot  be drawn  with  regard



 to the reliability  of  the  data, that situation must be  clearly




 stated.

-------
     The adequacy of all  aspects  of the QA/QC plan should be



examined in detail with emphasis  on  defining for future studies



an appropriate minimum  adequate plan.  Some aspects of the QA/QC



plan may have been  too  restrictive, some may not  have  been



restrictive enough.   Soil  monitoring studies should have checks



and balances built into the  QA/QC plan which will identify early



in the  study whether the plan  is adequate and; if required, allow



for  corrective action to be taken  before the study continues.



This is one of the  major advantages of conducting an exploratory



study.







     There is insufficient knowledge dealing with soil monitoring



studies to state with confidence which portions of the QA/QC plan



will be generally applicable to all soil monitoring studies  and



which portions must be varied depending on site-specific factors.



As  experience  is gained  it may  be possible to provide more



adequate guidance  on this subject.   In the meantime  it is



recommended  that many  important  factors of QA/QC plans be



considered as  site-specific until proven otherwise.







     Another important  aspect of QA/QC is auditing.  The purpose



of an  audit  is  to insure that all aspects of the QA/QC system



planned for  the  project are in  place and functioning well.   This



includes all aspects of  field,  sample bank and  laboratory



operations.  Whenever a problem is  identified  corrective  action



should be initiated and  pursued  until corrected.  Sample



                              10

-------
chai n-of-cus tody procedures  and raw  data are  checked as



appropriate  and results of blind QA/QC samples  routinely inserted



into the sample  load are  reviewed.  Spot-checks  of sampling



methods  and  techniques, sampling and analysis calculations, and



data transcription are performed.  Checks are made  to ascertain



that required  documentation has  been maintained and in an orderly



fashion,  that each of the recorded items is properly  categorized,



and cross-checking  can be easily accomplished.  Checks are made



to  insure that data recording conforms to strict document control



protocols and  the program's QA/QC plan.







     It is  recommended  that  an audit of the overall QA/QC plan



for sample documentation, collection, preparation, storage, and



transfer  procedures be performed just before sampling  starts.



This is  to  review critically the entire  sampling operation  to



determine  the  need  for any corrective  action early  in  the



program.







     The  project  leader  of a  soil monitoring  project  is



responsible  for ascertaining that  all  members of his project team



have adequate training and experience to carry out satisfactorily



their  assigned  missions and functions.   This is  normally



accomplished through  a combination of required classroom



training, briefings on the specific monitoring project  about  to



be  implemented,  and  field training  exercises.  Special training
                              11

-------
programs  should be completed by all personnel prior to their
involvement  in conducting audits.
                              12

-------
                                           EPA-600/4-84-043
      SOIL SAMPLING QUALITY ASSURANCE
               USER'S GUIDE
                    by


  Delbert S. Barth and Benjamin J. Mason
       Environmental Research Center
      University of Nevada, Las Vegas
         Las Vegas, Nevada  89154
           Cooperative Agreement
               CR 810550-01
     Kenneth W. Brown, Project Officer
   Exposure Assessment Research Division
Environmental Monitoring Systems Laboratory
    Office of Research and Development
             Las Vegas, Nevada
                 March 1984

-------
                            NOTICE
The information in this document has been funded wholly or in
part by the United States Environmental Protection Agency under
Cooperative Agreement CR 810550-01 to the Environmental Research
Center.  It has been subject to the Agency's peer and
administrative review, and it has been approved for publication
as an EPA document.

-------
                         ABSTRACT


     The  inherent  inseparability of a cost-effective  Soil
Sampling Quality Assurance/Quality Control  (QA/QC) Plan from  the
objectives of a soil monitoring program is emphasized.  Required
precisions  and confidence  levels for the data  cannot be defined
until  the decisions which will be made on the basis  of the data
are clearly stated and the consequences of making Type I  (false
positive) or Type  II (false  negative) errors are weighed.
General and specific objectives for Soil Sampling QA/QC Plans  are
presented  and discussed with special emphasis on cases associated
with the management of hazardous wastes.  Selected statistical
considerations  are presented with special attention  to analyses
of variance of soil  monitoring  data, methods  of calculating
required  numbers  of soil samples to achieve desired precisions
and confidence levels, possible  applications  of Kriging,  and
assignment of control  limits  to QA/QC data.  The  value of an
exploratory or  preliminary  study  to  the  cost-effective
achievement of  both the soil  monitoring  objectives  and the
objectives  of the Soil Sampling QA/QC Plan  is strongly  emphasized.
The  value of developing a  hypothetical model to estimate the
distribution in space and  time of  soil pollutants and thus  to
assist in the design of the  monitoring network is discussed.
Methods for determination  of the  number and  locations of  soil
sampling  sites;  sample collection  methods and procedures to
include frequency  of sampling;  sample handling to include
labeling, preservation,  preparation for analysis, and transport;
together with QA/QC aspects of all of the above are presented and
discussed.  The  ultimate goal in the analysis and intepretation
of data is  to build a body of representative and comparable  data
on  the basis of which both general guidelines  (factors applicable
to  all sites) and  specific  guidelines  (factors  which  are
site-specific)  may be  developed for Soil Sampling QA/QC Plans.
Finally, the importance of systems  audits and training  to the
achievement of  soil sampling QA/QC objectives is presented and
discussed.
                             111

-------
       SOIL SAMPLING QUALITY ASSURANCE USER'S GUIDE

                         CONTENTS
Notice		ii
Abstract	iii
Figures   	vi
Tables    	vi

1.   Introduction 	      1
       Background 	    1
       Objectives 	    4
       Audience 	    4
       Approach 	    4
2.   Objectives of Quality Assurance-Quality Control
       Plans	    6
       Introduction 	    6
       General identification of the objectives  ....    8
       Objectives for background monitoring 	   14
       Specific objectives for monitoring  in support
         of CERCLA	15
3.   Statistical Considerations 	   30
       Introduction 	   30
       Distribution of soil sampling data	31
       Statistical designs  	   32
       Data analysis	40
4.   Exploratory Study   	   50
       Introduction 	   50
       Number and location of sites for  sampling   ...   51
       Sampling and sample handling 	   53
       Analysis and interpretation of data	55
5.   Selection of Numbers of Samples and Sampling  Sites   57
       Introduction 	   57
       Number of sampling sites required  	   58
       Location of sampling sites   	   62
       Quality assurance aspects  	   63
6.   Sample Collection   	   65
       Introduction   	   65
       Size of samples and method of collection  ....   65
       Boring Log	66
       Frequency of sampling  	   66
       Quality assurance aspects  	   67
7.   Sample Handling and Documentation   	   69
       Introduction   	   69
       Container preparation, labeling,  preservation
         and sample preparation 	   69
       Quality assurance aspects  	   75


                              iv

-------
8.    Analysis and Interpretation of QA/QC Data  ....  76
       Introduction   	  76
       Presentation of data Summaries	76
       Presentation of results and conclusions  ....  81
       Quality assurance aspects  	  81
9.    Systems Audits and Training  	  83
       Introduction   	  83
       Sample bank audit	84
       Daily log	86
       Bank logs	86
       Sample collection audits 	  86
       Field audits	87
       Data management audits  	  88
       Training	88

References	89
Appendices
  A.  Tools for estimating number of samples  to achieve
      specified levels of precision and confidence   .  .  93
  B.  Tables for use in calculating confidence and
      tolerance limits and judging the validity of
      measurements	99
                                v

-------
                            FIGURES
Number                                                      Page

  2-1        Data acquisition flow for hazardous materials    16
  2-2        Monitoring data flow	   21
  2-3        Technology and transfer data flow	   23
  3-1        Acceptance region for H:yo=30.0   	   35
  3-2        Type II or  3 error	   35
  A-l        Number of degrees of freedom required  to
             estimate the Standard Deviation within P% of
             its True Value with confidence level Y •  . • .   94
                            TABLES

   2-1        Effect of  increases  in analyses  and  sample  size    8
   3-1        Analysis of variance of a nested soil
               sampling design   	    34
   3-2        QA/QC Procedures for soil samples	    41
   3-3        PCB study  to determine contamination of
               an area  (hypothetical data)	    48
   5-1        Number of  samples  required  to  achieve  a
               sampling precision  (P) at a  confidence
               level of (1-a)	    60
   5-2        Number of  samples  required  to  achieve  a
               sampling precision  (P) at a  confidence
               level of (1-a) and a power of  (1-6)	    61
   7-1        Sampling containers, preservation requirements,
               and holding  times  for soil samples	    71
   7-2        Accountable document control requirements   .  .    74
   A-l        Estimated  number of  samples to achieve
               specified levels  of precision  and  confidence
               when coefficient  of variation  (CV) is known     95
   B-l        Percentiles of the  t distribution	100
   B-2        Confidence interval  for averages 	   101
   B-3        Single classification factor (c^) to estimate
               standard deviation  from range, and equivalent
               degrees  of freedom  (/)a	102
  B-4        Tolerance  interval  for individuals 	   103
  B-5        Critical values for discarding invalid
               measurements 	   104

-------
                         CHAPTER 1
                        INTRODUCTION
BACKGROUND

     Data resulting  from any monitoring or  sampling program
cannot be  evaluated  and  interpreted with  confidence unless
adequate  quality assurance methods  and procedures have  been
incorporated into  the program design.   An  adequate  quality
assurance program requires  that all sources  of error associated
with each  step of the monitoring or sampling effort be identified
and quantified.

     Sources  of error  are then analyzed with  an appropriate
statistical  design to yield estimates of the various  components
of  variance.   The component  of variance analysis is based  upon
the premise  that the total variance for a particular  population
of  samples is  composed of the sum of the variances from each  of
the identified sources of error  plus an error  term which is  the
sample to sample variance (a2).  The population variance (a2)  is
usually unknown: therefore, it must be estimated from a set  of
samples collected from the population.  The total sample variance
(s2) is estimated from the summation of the sum of squares  (SS)
from  each  of  the  identified components of variance plus the error
SS.  For example:

              SSt = SSs+SSp+SSex+SSa+SSer

       where  SSt = Total SS
              SSS = Sampling SS
              SSp = Sample Preparation SS
              SSex= Extraction SS
              SSa = Analysis SS
              SSer= Error SS

The result of  this analysis provides a measure of the precision
of  the analysis  plus confidence  limits.

     To date the most  highly developed aspect  of quality
assurance  undertaken i-n  support of monitoring programs has
been  for the  analytical procedures.  Such an approach is not
adequate in  cases where the medium being  sampled  is  not

                              1

-------
homogeneous, which is particularly true for  soil and may also be
true in some  instances for air, water, sediments or  foods.   For
example,  two soil samples  taken a few feet apart may differ in
important  characteristics or chemical pollutant concentrations  by
an  order  of magnitude or more.  Therefore,  quality assurance on
the  analytical  results  is  a necessary but not  sufficient
condition for assessing total sample variability within the soil
population being sampled.  The analytical errors may  account  for
a negligibly small portion of the total variance.  In view of the
above it is clear  that for  soil  monitoring programs a more
comprehensive quality  assurance  program  is mandatory.  This
document will address the quality assurance  aspects  associated
with soil  sampling.

     Prior  to  proceeding with the quality  assurance aspects it
is necessary to discuss soil monitoring  in general,  monitoring
objectives, and possible actions which may be taken on the basis
of the resulting monitoring data.  Clearly it  is not  possible  to
separate the  required  quality assurance  procedures for soil
monitoring from the objectives or purposes for which  the soil  is
being monitored.

     Soil monitoring may be:

     o   carried  out to meet  the provisions and  intent  of
         environmental laws such as the Resource Conservation  and
         Recovery  Act  (RCRA),  the Comprehensive Environmental
         Response, Compensation, and  Liability Act  (CERCLA),  the
         Federal  Insecticide,  Fungicide,  and Rodenticide Act
          (FIFRA), the Toxic Substances Control Act  (TSCA), etc.

     o   source,  transport or  receptor oriented

     o   conducted to  determine  the  presence of  specified
         contaminants in comparison to an appropriate  background
         level

     o   conducted to determine the  levels of contaminant and
         their  spatial and temporal distribution

     o   conducted to provide  input  into  an exposure and risk
         assessment study

     o   part of  a compliance  monitoring scheme to measure the
         efficacy of control  actions

     o  part  of  a research,  technology  transfer,  or
        environmental model validation study.

     Presumably, in  each case where soil  monitoring is deemed
necessary, administrative or legal actions are likely to be taken

-------
on the basis of an evaluation and interpretation of the resulting
data.   The  consequences of taking or not talcing action must  be
clearly understood before it is possible to establish  an
allowable  confidence band for quality assurance of  the data.
After weighing  and evaluating the consequences, a value judgment
must be made by  a responsible official concerning the  acceptable
probability of making a Type I (false positive) or a Type  II
(false negative) error (see Chapter 3 for discussion of  Type I
and Type II errors).  It is not possible to design a meaningful
quality assurance program until this  step has been taken.

     The acceptable  probability for each type of error must be
established in  relation to the consequences of making  such  errors
and depends upon the specific objectives of the soil monitoring
program.   The  Type I error is the error most  often  used  in the
literature.   In environmental monitoring,  however,  the Type II
error may be more  important than the Type  I  error.   The  cleanup
of  a  highly  toxic  spill would  be an example  where a false
negative could create major problems  for the environmental
manager.   The Type II error would  lead  the manager  to conclude
that a cleanup of some areas is not necessary when  in fact the
action levels are being exceeded and cleanup is  necessary.  The
Type I and Type II  error for the QA/QC effort should  be equal to
the error  levels chosen for the sampling  effort  itself.  This
acceptable probability  in different cases may, for  example, range
from  20  percent to 1 percent or less.  In some circumstances the
value judgment may simply be a  statement  of an allowable error
not to be exceeded in the final data.

     There may  be  a temptation in many cases  to avoid making the
necessary value judgments concerning acceptable probabilities of
making different kinds of  error.  The  course  of  action often
substituted for the difficult  value  judgment is  to adopt as a
guiding  principle the concept  that one  should  always strive to
achieve the highest precision and level  of confidence (or lowest
error)  possible with  exisiting  available resources.   The
 resulting  data are then used as  the basis for making decisions
with  the assumption that  this  guiding  principle  gives  the best
 possible result.  Obviously such  an  approach will  rarely, if
 ever, be cost-effective.   Two types of  errors are possible.   The
 data may be much better than required  which indicates resources
 have been wasted,  or  the  data may not be of adequate quality
 thereby resulting in costly  decisions of  doubtful validity.  This
 point may be summarized  by stating that resource availability is
 an important factor  for  consideration in the establishment of
 quality assurance programs, but resource availability should  not
 be accepted as the sole determinant of  required quality assurance
 methods and procedures.   Maximal cost-effectiveness should be  the
 overall goal.  This  generally means  that  a minimum  adequate
 quality assurance plan must be defined and then implemented.

-------
OBJECTIVES

     This  document  is  intended  to serve  as  a user's  guide
presenting  and explaining  selected principles and applications of
methods and procedures for  establishing  adequate  quality
assurance on  soil  sampling  aspects of environmental monitoring
programs.  The  soil  sampling aspects treated  include sample  site
select ion, sample collection,  sample handling, and analysis and
interpretation of resulting  data.  No detailed treatment of
analytical quality assurance procedures  is  given since that
important aspect of the  overall problem has been adequately
treated  elsewhere   (USEPA-600/4 - 82 - 030 a -df  1982;
USEPA-600/4-84-012,1984).   It should be noted, however, that
sampling quality  assurance procedures are not fully separable
from analytical quality  assurance procedures.   This is
particularly  true  for sample collection and handling procedures.
If  an intact,  timely,  and  representative sample of proper  size
and constituency is not  adequately delivered to the analytical
laboratory,  the analytical  quality assurance procedures cannot be
expected to  yield meaningful results.  Thus the soil  sampling
quality assurance procedures presented here  should be viewed as
an  important  integral part of  the overall quality assurance  plan.


AUDIENCE

      This  document has been  developed to serve as a user's guide
 for anyone designing, implementing  or overseeing soil monitoring
programs.   It is  especially  applicable for personnel responsible
 for regulatory programs where soil monitoring is an important
 integral element.  Special attention is  given to soil sampling
examples related to CERCLA since such applications are deemed to
constitute high priority for  soil  sampling programs.  Many  of the
principles and procedures discussed, however, are applicable to
other situations as well.
 APPROACH

      Following  the  presentation in Chapter 2 of  general  and
 specific objectives of quality assurance plans a brief  survey of
 selected applicable statistical methodology is presented.

      Discussion  of  the value  of an exploratory study  to  the
 subsequent design of a soil sampling quality assurance program
 leads logically into more detailed discussions of  sample  site
 selection,  sample collection,  and  sample handling.   These
 detailed  discussions  will  include minimal coverage of  soil
 monitoring protocols, per se, since they were recently treated in
 a  comprehensive document (Mason,  1983).   The  focus  of  the

-------
discussions here will  be  quality  assurance.   The  goal of  each
discussion will be the development of design features  for sample
site selection, sample collection, and sample handling to meet
quality  assurance objectives  of  defined precision and levels of
confidence for each subject area.

     Similarly, the goal  of  the discussion concerning analysis
and interpretation of data  will  be  focused on  quality assurance
aspects.  Special  attention will be focused on a technique known
as the components of  variance analysis.  This analysis results
from the use of a  statistical  sampling plan  designed to measure
as many of  the  sources of variation as can be identified and
sampled in a  cost effective  manner.   The  analysis further
identifies the  amount of total  sample error  (or variance)  that
results from each  component in the sampling  - analysis chain.
The last two subjects treated are program audits and personnel
training.

     To the maximum extent feasible throughout this  report we
will present concepts and principles  first  and then present
selected examples  of how these concepts  and principles may be
applied in  realistic situations.

-------
                          CHAPTER 2
     OBJECTIVES OP QUALITY ASSURANCE-QUALITY CONTROL PLANS
 INTRODUCTION


     The main objective  of any soil  sampling quality
 assurance-quality  control  (QA/QC)  plan  is  to determine  the
 quality of  the  reported  data and insure  that it is adequate to
 the degree required for the  intended end use  of  the data.   How
 this objective is met depends upon the  purpose of the particular
 sampling program.

     Soil  is by its very nature extremely variable.   Superimposed
 on this variability are other  sources of  variation or error  that
 can  be  introduced into the final result  by the sampling and
 analytical efforts.  The concept of total QA/QC has been used to
 develop a  system for assuring the quality of the results by
 attempting either to provide control of the  various steps  in  the
 analytical  process leading from sample collection  to data
 interpretation  or to provide adequate  replication  for
 statistically determining and  quantifying  the sources of
 variation or error in the chain.

     QA/QC  requires  that  each step in  this chain be shown to be
 valid.  USEPA (1976) has indicated that the  quality  of the  data
 considered  to be acceptable must be defined as quantitatively as
 possible.  The variability of  the soil and the requirement  for  a
 quantitative standard  for  acceptability  requires  that  a
 statistical  sampling plan  be developed that assures  the
 precision, bias,  completeness, comparability,  and
 representativeness of the sampling effort and  of  the resulting
 data.

     Statistical sampling is the  mechanism by which the QA/QC
 program can  determine the sampling precision  and  can provide  a
measure of the  reliability of the entire sampling  effort.
 Buffington  (1978) quotes Congressman  George  E.  Brown  Jr.  as
 saying "no number  is significant, and  subsequently worthy of
being recorded, without an estimate of its   uncertainty."   This
statement should be considered when designing the QA/QC plan for
a soil  sampling effort.

-------
     It is not uncommon to spend considerable money on a sampling
and analysis program only to find that  the  samples  were not
collected in a manner that allows valid conclusions to be drawn
from the  resulting  data.   Ku (1978) states "...the design of a
proper statistical sampling scheme depends almost entirely on  the
purpose  for which  the results are going to  be used.   Hence,
without  an explicit  and  defined purpose for an undertaking,  the
design of the sampling scheme cannot be formatted for efficient
data collection  and  for  the correct interpretation of results."
Skogerboe and Koirtyohann (1976) note "...cost considerations
most frequently  serve as  the excuse for failing to carry out even
the most  rudimentary quality assurance checks."  This failure to
accept  the  fact  that quality assurance is cost  effective
continues to plague  the  scientific community in spite of the
considerable body of literature to  the contrary.

     QA/QC programs  for  analytical laboratories are widely used
and accepted.  These  programs are strongly oriented toward  the
analytical process, not toward  the  sampling  effort.   This
orientation was indicated by Plumb  (1981) who suggested  that 15
to 20  percent of the total analytical work load be dedicated to
the quality control program.  Of this, five percent was  allocated
to the sample collection effort and 15  percent to factors that
were  considered to  be a  portion of  the laboratory effort.
However,  studies that have been done to  determine the components
of variance  (Snedecor and Cochran, 1982;  and Bauer,  1971)  in
final  monitoring data indicate that reversing the quality control
sample allocation load would not only be more cost effective  but
also  provide  a  better estimate of the quality of  the sampling or
monitoring effort.

     The  component  of  variance  analysis identifies  various
components of  the environmental sampling scheme that  influence
the  total variance of the samples collected.   The  design of the
statistical  study depends upon the purpose of the  sampling  and
the  method of collection.  A  first approximation  of  the total
variance can  be defined by the following equation  (Bauer, 1971).

                     Vt=(Vs/k) + (Va/kn),

where  k  is  the  number of samples, n the number  of analyses per
sample, kn the total  number of  analyses,  Vt the total  variance,
Va the analytical variance and Vs the sample variance.

     One of the  general  purposes of a sampling  effort  should be
to obtain samples with the smallest feasible Vt and thus the best
available precision.  Intuitively one can identify the  portion of
a  sampling  effort where  the greatest  gain would be made  in
reducing the  variation in  the  data.   Analytical programs
frequently attempt  to attain a precision of  less than +  1 percent.

-------
On the  other hand, sample to sample variation for  soils is often
on the order of 30 to 40 percent.   Increasing n  by one reduces
the  contribution from  the small  analytical  variance but has
little
       or  no effect on the total variance,
one  reduces both the larger  sampling variance
variance;  thus, increasing k produces more gain
of the sampling effort.
 Increasing  k by
and the  analytical
in the precision
     The  actual data  presented  in Table
approximately  equal  (Bauer,  1971)  show
increasing  the number  of analyses while
samples constant and vice versa.  These data
was previously described.
                                           1  where Vs and Va are
                                           both  the  impact  of
                                           holding the number of
                                           show  the  gain  that
 Table 2-1.  EFFECT OF INCREASES IN  ANALYSES AND SAMPLE  SIZE
           n
                     kn
  •t*
1
1
1
2
4
1
2
4
1
1
1
2
4
2
4
0.0993
0.0793
0.0693
0.0396
0.0248
0.32
0.28
0.26
0.20
0.16
 * total standard deviation
     This  information should  be considered in determining the
 objectives according to the steps outlined below.
GENERAL IDENTIFICATION OF THE OBJECTIVES
General
     The objectives of  any quality  assurance  effort  focus
primarily upon verification of the  required reliability of  the
sampling data to  support actions which will be taken  on the basis
of interpretation  of the data.  An  important consideration is  the
percent of  Type I or  Type II error which will be deemed
acceptable.  A decision on this matter leads to the  definition of
required confidence levels.   Reliability might be  defined as the
probability  that a particular  measure of the soil system reflects
the  true  average value for all of  the soil  in a defined
geographical area  in which the sample was taken.  This requires
that  the accuracy of  the method, the representativeness of the
sampling, and  the  comparability of  the data be determined.

                              8

-------
Accuracy and Comparability


     The  accuracy and comparability of the data are determined by
the analytical  laboratory working with the agency's quality
assurance laboratories.   Reference samples, spikes and duplicate
analyses are used as part of the  routine, daily  QC program.   The
laboratories  also  are  involved in  an interlaboratory
collaborative  program  designed  to aid in  evaluating  the
quality of the  data produced by a particular laboratory using a
particular method.   Environmental  scientists must work
cooperatively with  the analytical laboratory to insure that the
required field samples are taken for use  by  the  laboratory  for
conducting its  own  QC effort.   The investigator is  not likely to
be directly involved in the determination of  the accuracy  of  the
method used by  the  laboratory unless he has a specific reason to
further verify the laboratories'  performance.


Precision  and Representativeness


     The  precision  and the  representativeness  of  the final
results are of primary concern  to the  investigator.   Precision
measures  the repeatability of the results  obtained from analyzing
the collected soil samples.  The statistical  plan used  to  design
the  sampling effort should take  into  consideration all  of the
sources of variation likely to be encountered.

     Representativeness  of sampling has  two  components.  First
the sample taken must  reflect what  is present in the soil.  For
example if one  must measure the  macro  structure of the  soil an
undisturbed block sample must be taken and  carefully prepared for
transport to  the lab.  A sample  taken  for analysis of metals on
the other  hand  can be  disturbed.  Another example of this  aspect
of  representativeness would be a  landfill where scrap wood had
been used to  absorb  liquids placed in  the landfill.   The wood is
not  soil  and  therefore should  not be  included in the analyses;
yet,  it is a major component of  the pollutant source term and
therefore should  be measured.   Thus,  the objectives of the
sampling  will  determine  how the  sample  should be taken and
handled during  shipment,  etc.   This  first  aspect of
representativeness  is  difficult  to quantify because it is often
quite subjective.

     The  second aspect of representativeness  can  be  quantified
because it is closely  tied  to  sampling precision.  This aspect
addresses the reliability  of the mean  and the  standard deviation
as measures of  the  amount  of  a  chemical present in a particular
area.   Increased sampling  intensity,  (spatially  and/or

                              9

-------
temporally) independent  sampling, sampling audits and the use  of
confidence maps  developed  by Kriging  are all  examples of
techniques that can help  insure that the sample  is representative
of the conditions in the area  under investigation.

     The sampling plan should  be designed in such a way that  it
provides  quantitative measures  of both  precision  and
representativeness.   The QA/QC required levels of reliability
vary with the purpose of the study.   Six situations where soils
are  likely to  be sampled as  a part  of  the USEPA's mandated
responsibilities are illustrated below.   Each situation  has its
own QA/QC requirements.  The situations are:

     o     preliminary site investigations
     o     emergency cleanup operations
     o     remedial response operations
     o     sampling for litigation purposes
     o     monitoring
     o     research or technology transfer


Preliminary Site Investigation;

     The  purpose of  a preliminary  site  investigation is  to
provide information  about a specific site that can be  used  in
making initial  management decisions, and,  should further work  be
necessary, for designing a detailed and comprehensive monitoring
program.   The data  collected during the preliminary study are
often used to  determine if  a site  will  require  further
investigation.   Three basic conclusions from preliminary data
might be:

     o     No further study is needed.
     o     A detailed study is needed.
     o     A conclusion cannot be drawn without further study.

     For such conclusions  to be drawn  there  is a definite  need
to measure the reliability of  preliminary data.   However,  methods
that are often  used could be classified as "look and  see"  or
"quick and  dirty."   Many times little attention  is  given  to
statistics in  sampling  design, and the efforts are sometimes
biased by  the interests  of the particular investigator.  Any
approach  for collecting  preliminary data without adequate QA/QC
should be strongly discouraged as counterproductive.

     Judgmental sampling  is  often  used in the  preliminary  site
investigation.  This  approach is based upon the investigator's
judgment as to where the samples should be taken.  Judgmental
sampling may be the best  approach to use if there is a  limited
number of  samples allowed  by the  study  plan.   The major
disadvantages of using  judgmental sampling centers on  the fact

                             10

-------
that  the samples  are biased and  that very often  little useful
information on spatial variation is  obtained.   This approach
prevents the investigator from making certain statements about
the reliability of data derived from  the samples collected.

     On the  other hand,  judgmental  sampling can be  combined with
other sampling plans in order to insure that particular aspects
of the  site can  be addressed.   For example, examination of
historical  data and a site visit  may  indicate  areas where
pollutants were  believed to be placed and how a plume  from this
location may  have moved in a particular  direction.   A sampling
plan  can then  be  designed to focus  the sampling at the original
disposal area and along the  suspected  plume  axis.  The exact
sampling locations can then be randomly or sequentially located;
the exact choice depending upon the  purpose of the study.


Emergency Cleanup Operations


     The  emergency  cleanup  operation (immediate removal) has a
single focus  —  remove the pollutant  as  quickly as  possible  to
achieve either a  background  level or a  level that  is not an
unacceptable  threat to human health,  welfare or the environment.
The  principal  role of QA/QC relates to a reliable  determination
that cleanup  operations have been adequate.

     The  contamination may be limited  in scope; e.g., a group of
leaking barrels  lying on the surface  or  a spilled tank truck.   A
leaking landfill, on  the other hand, could present an entirely
different demand on the investigator.  Resources are  limited  for
site  investigations no matter what the  level in the  chain  leading
from discovery to abatement.  The emergency cleanup operation
often leads to one of  the higher steps in the chain.  Criteria
for determining  if  the emergency  cleanup operations should be
elevated to either a  planned  removal or a remedial response
operation are based to some extent on cost and the length of  time
required to implement the action (USEPA, 1982).

     The  fact  that  the emergency cleanup  operation can  lead to
one of the more  definitive types of monitoring operations  places
a  demand on the investigator to collect as much information on
the area as is possible within the time  and resource  constraints.
This  demand suggests  that an  adequate  statistical  design
incorporating appropriate QA/QC  measures should  be used to
provide  the quality of data needed for the  decisons that  must be
made by the On-Scene Coordinator.  A decision made on the basis
of data  collected  following a statistical  design is more likely
to be defensible as there is an identified measure of reliability
that can be placed  on the decision process.

                              11

-------
Remedial  Response Operations


     As has  been previously identified, the remedial response
operation is one of  the possible steps  in  an environmental
cleanup  process.  Studies conducted under a remedial response are
normally  quite detailed and have the latitude  to provide  for a
thorough  soil investigation.   The  designs used  in  these
situations are likely to be directed  to provide not only a
quantitative evaluation of  the pollutant data but also a spatial
evaluation as well.  The remedial  response operation  which is
designed as a permanent remedy, at times allows the investigator
to develop a data base on time trends.

     Many of the  actions  used  to control  the impacts of a
pollutant depend upon the nature and properties  of the soil at
the  site.  The  studies  conducted for remedial response usually
attempt to develop soil physical and chemical  characterization
data along with the pollutant data.

     Similar to remedial response,  planned removal operations
may also involve extensive monitoring and sampling  programs.
The  detail  and  extent  of  a  planned  removal monitoring and
sampling program may be quite extensive.   However, the planned
removal  operation is limited by time (six months), thus limiting
the investigator  from collecting data for time trend analysis.


Sampling for Litigation Purposes


     All  of the previously mentioned sampling situations  have a
potential for litigation.  There are, however, a group of studies
that  are designed for  specific litigation goals.  Potential
questions may arise in court  that must be addressed  by studies
designed to provide  specific data.   Negotiations may require
studies that are  designed to aid in drawing  conclusions and
making decisions  prior to taking a negotiating stance.   These
studies   require that  a good  experimental design be  used.
Statistical designs incorporating appropriate QA/QC measures  and
including "chain-of-custody"  procedures are one of the  best ways
to provide the kinds of information needed in  these situations.

     The  litigation study is often similar to a research study  in
that there is a specific hypothesis that is being  tested or
evaluated by the investigator.  Details will vary as the purpose
changes.   The confidence levels desired and thus the cost of the
studies  are usually higher  than those encountered in other types
of field  sampling efforts.

                             12

-------
Monitoring Studies


     The purpose of a monitoring study may be routine or it may
be the outcome  of  a  negotiation process.   The  study is often
designed to measure  pollution levels  and changes  in  spatial and
temporal distributions.  Evaluation  of these  changes may  require
the use of techniques such as intervention analysis or Kriging to
aid in  identifying patterns which may not  be  directly  obvious
from the monitoring  data itself.  These  studies  can occur over
long periods of  time and have a routine  aspect to them.

     The design of a  monitoring study and  the associated QA/QC
procedures frequently is dictated by a  negotiated  settlement or
by some administrative review process.  Designs that are costly
often  have little chance of being   implemented.   The more
efficient the  design, the better  the  likelihood that it will be
implemented.  The reliability of the data combined with a  measure
of cost effectiveness should be the primary consideration for
selecting the  monitoring design.  This  implies that the QA/QC
program design should also be minimally  adequate.   Too often,
cost is the  excuse  for failing to apply  even  limited  statistical
and QA/QC criteria in the monitoring  design selection  process
 (Skogerboe and  Koirtyohann, 1974).

     Properly designed,  quality assured, monitoring  plans allow
the  investigator to  make comparisons  within  a particular
monitoring data base and also  to make comparisons  with other
monitoring  studies.   Thus, each properly designed monitoring
study  can add  to our overall knowledge of the environment rather
than just provide data  of  limited reliability for a limited use.


Research or  Technology Transfer Studies


     These  studies  are  by  nature designed to answer  specific
questions.   The purpose  is  to  increase knowledge or to make a
decision about some characteristics  of the soil system.  The
experimental  design should  be statistically and  quality
 controlled.   Failure  to  provide a measure  of  the reliability of
 the data will ultimately  lead  to  the  results of the  study being
questioned or invalidated  by peer review.
                              13

-------
OBJECTIVES  FOR BACKGROUND MONITORING


General


     Generally the design  of  soil  monitoring programs requires
that the levels  of  defined hazardous or potentially  hazardous
substances and  their spatial  and temporal trends be measured for
some  specific  purpose.   Often it  is  critical  not  only  to
quantitate levels and trends but also to link the existing levels
to sources.  This is necessary  to enable adequate control  actions
to  be taken whenever  a situation that is  hazardous to  human
health,  welfare,  or the environment is  identified.   often the
situation is  complicated by the  fact that multiple  sources
contribute to the measured levels.

     The  situation  is  further complicated  by the presence of
pollutants of recent origin mixed with pollutants of past  origin.
This  mixing becomes especially important when the investigator
attempts to trace the migration from source to receptor and also
 in  predicting what future levels are likely to be after various
proposed control measures are implemented.

      Identification of  spatial and temporal trends  along with
 linkage of  observed measurements  to  sources requires  that
adequate background,  reference or  control  samples be taken.

      In  the absence of  such background samples,  interpretation of
 the resulting  data may become extremely  difficult, if not
 impossible.  The burden of proof that background samples are not
 necessary, for a particular soil monitoring study, rests with the
principal investigator.   In  the absence of  such proof a prudent
 investigator will  insure that  the collection  of  adequate
background samples is  included  in the  monitoring study design.
 Furthermore,  some  EPA regulations concerning regulatory
monitoring (U. S.  Code  of  Federal Regulations, 1983) specifically
require background sampling.

      Since measured levels in presumably  higher  concentration
areas  will be compared to background levels, QA/QC procedures are
just  as critical for the  background measurements  as  they are  for
the study  area measurements.   Thus,  for background  sampling, a
QA/QC procedural umbrella  must cover the selection  of  appropriate
geographical areas, the selection  of  sampling  sites  within the
geographical areas, sampling, sample storage and/or  preparation,
sample analysis, and interpretation of the resulting data.

     Under  most circumstances  background data will  not be
available  for a given monitoring  location.   These data must be

                              14

-------
acquired  either  before or during  the exploratory or  preliminary
investigation phase.  The intensity  of  the background sampling
that  is undertaken depends upon  the pollutants being measured,
the soil characteristics and variability, the levels of pollutant
likely to be found in the study area and the purpose of the study.
Further discussion of the role of  background studies is presented
in Chapter 4.


SPECIFIC OBJECTIVES FOR MONITORING IN SUPPORT OF CERCLA


     The specific QA/QC precision and confidence level objectives
for any sampling study are controlled in part by the goals  of  the
particular study.  Three example situations where soil sampling
will likely  be undertaken are:

     o    hazardous  materials  investigations for areas  such as
          abandoned landfills or  chemical  spills
     o    monitoring studies
     o    technology transfer

     The data  flow that can occur in each of these situations is
outlined in  Figures 1,  2  and 3.   The data  generated in  each
category can  provide  input  into the  development of  plans and
specifications  for  the other situations.  Data that have  been
subjected  to  a good  QA/QC plan can be  relied  upon as a resource
for the development of new data.

     The main  area  where the magnitude  of  the soil sampling can
be controlled  is  in the  precision required  by  the  sampling
designs.    The accuracy of  the sampling is unknown because the
true average is not available.   Repeated sampling  for  a  high
precision  nevertheless must  rely  on  the analytical accuracy
obtained in the  laboratory to  insure  that the  methods  used
measure what  is  present in the soil sample.  Thus  the accuracy
for analysis applies  only to  the samples  while the  accuracy for
the  study depends on the degree to which  the samples are
representative  of  the  area.

      The  steps  outlined below are  designed  to provide  a
monitoring effort  with the  needed sample precision  and
representativeness (USEPA,  FR 44:233, 1979 and Bauer,  1971).


      1.   Identify the objectives of  the  study.  These should
 reflect the specific items  of  information  that are required to
make  the decisions  that will  follow  achievement  of the  study
objectives.
                               15

-------
I
End
        End
                         I
                       Preliminary Site
                        Investigation
                                  Ijuigoncy
                                  Clean-tp
 Figure  2-1.  Data acquisition flow for hazardous materials.

                                      16

-------
          Remedial
          Response
         Litigation
          Sampling
         Monitoring
           Effort
Figure 2-1.   (Continued)
              17

-------
      Corrective
        Action
   Design Corrective
       Measures
Figure 2-1.   (Continued)
         18

-------
     Remedial
     Response
     Clean-up
     Complete
        is
    Litigation
     Planned
     Planned
     Renoval
     Planned
     Renoval
Figure 2-1.  (Continued)

        19

-------
        Is
     Research
      Needed
      Research
       Studies
Figure 2-1.   (Continued)
       20

-------
            Design
       Monitoring Study
            Acquire
        Monitoring Data
            Exceed
           Standards
           Initiate
       Corrective Action
Figure 2-2.   Monitoring data flow,

                21

-------
    Litigation
    Anticipated
    Litigation
      Studies
Figure 2-2.  (Continued)
        22

-------
                Technology
                 Transfer
                 Studies
                   •ere
                Objectives
                    Met
Figure 2-3.  Technology  transfer data flow,
                    23

-------
     2.   Determine the components of variance  that are built  into
the statistical design.  Proper stratification of the study area
will  lead to several sources  of variation.  The sources of
variation that can be controlled  by the sampling are determined
by  the  particular  sampling design used and by the pattern of
sample collection superimposed  over the area.  An analysis of
variance  of  the  data  provides  information which can be used in
calculating components of variance.  Table 3-1 gives an example
of  a  component of variance study on a set of hypothetical soils
data.

     3.   Choose  the desired confidence level.   A confidence level
of  95 percent or  better is desirable; however, this is often not
possible because  of either fundamental constraints  or the
economics of  the  situation.  The  investigator may have to  select
a  level, determine  the  number of  samples  and  replications
required and then compare  this with the resources  that are
available and  recalculate the confidence  level that  can be
reasonably attained with the resources available.  Of course, if
the revised  level  is  not adequate to allow  achievement  of the
study objectives, then more resources must be found, the study
objectives must  be  revised, or  the study must be abandoned.  The
major point to be made here is that the confidence level  should
be chosen before the  study is  conducted and  not after the data
are collected.

      4.   Obtain sampling data  from other  studies that  have
similar characteristics to the one  being designed.  Particularly
desired  data are those  where the  results were derived  from
replicated samples.  These data will be used to calculate average
parameters  such  as  the coefficient of variation, variance and
confidence intervals for early stages of the sampling process.

      5.  Calculate the mean and range of each set of replicates.

      6.   Group the  sets  of replicates according to concentration
ranges  and by  the types  of samples  that are believed to be
similar.  An example of the groupings might  be samples  in the
range from 0 to less than 10, and 10 to less than 25 mg/1, etc.,
or by soil type such as those that  were in sand, silt or clay.

      7.   Calculate  the  critical difference Rc (number not to  be
exceeded to maintain adequate QA/QC) by noting that  for any group
of n duplicate  analyses that are considered similar  to  each
other,   their  ranges [R^]  and  means  [X^] can be  used to
                              24

-------
estimate the critical difference [Rc] between similar  future
duplicate analysis  for any specific concentration  level  [C].
Specifically,
                  Rc = 3.27   (C)
                            n
                                     n
z
     where X^ = Xf + Xj+j and Rj_  = Xj_ -
                  2
     8.   Develop a table  of  Rc  values for various  concentrations
that span the  range  of concentrations of interest.   (A  similar
approach makes  use  of confidence limits based on  the standard
deviation rather than the range. )  These data are used to accept
or  reject a  set  of  replicated samples.  The replicates  are
usually  duplicate samples,  and therefore the difference  between
the  two values should lie within the critical range.   If  the
sample  is rejected,  the analyses should be rerun  if possible.
The  indication  is that one or both of the samples or the  analyses
were variant.   Discarding of the results should  only   be done
after careful review of the data.   There are  situations in soil
sampling where the coefficient of variation can  reach hundreds of
percent  due to  the variability in the soil system;  therefore,  the
suspected outlier may in fact be a part of a wide  distribution.
This tends to be the case in situations where very  high levels of
chemicals have been spilled over small areas or where chemicals
have flowed through  desiccation cracks, animal burrows or  old
root channels.  Observations made by the field party  can aid in
making  a decision  to discard a sample if appropriate comments
have been noted  in  the  log books at  the time of  sample
collection.

     9.   The constructed  preliminary Rc table  is used until data
are acquired during  the sampling.  As the analyses proceed  the
results  are combined with those from previous studies.  At the
point where approximately fifteen pairs (USEPA, FR  44:233,  1979)
of  results are acquired from the particular study  area a  new
table should be  calculated based upon the average  range of  the
data that has  been accepted to date.

     10.   The  data  collected during  the preliminary  or
exploratory site investigation and during an emergency  response
activity become  the  data base upon which  later studies  are
evaluated and/or designed.

     The specific goals of each type of study will control the
required precision and confidence levels and the differences that

                              25

-------
are detectable  between  the  samples  and background.  General
guidelines are given below for each of the situations covered.


Preliminary  Site  Investigation


     The  preliminary or exploratory investigation is  the
foundation upon which many of the other studies  in  hazardous
waste work  is  based.   Following preliminary  data evaluation a
decision is  made  to either pursue a more definitive  study or else
to  terminate the investigation.  The reliability of data obtained
from these  investigations should  be as high as  feasible.   The
initial plan should attempt  to set the confidence  level above 95
percent.  If resources limit the number of samples  that can  be
taken, then the investigator should calculate the  precision that
can be attained based upon the  number of samples  that can  be
collected.   If  this new level of precision is deemed adequate,
the study should  proceed.

     Using  five percent duplicate samples would provide adequate
QA/QC sampling under most  situations  (Plumb,  1981);  however,
there should  be a minimum  of two  sets  of duplicates in each
strata sampled.   Improved precision can be gained by  increasing
the number  of replicates at  each QA/QC site sampled.  Analyses of
splits of each of these samples provide data for within sample
variation.   The latter information can be delayed until later
studies unless the within sample variation is expected to provide
information  that  can be used  in making decisions  about the site.

     When resources  are limiting, a two staged sampling design
may prove to be  advantageous.   The  first stage  would have a
limited scope and provide only  limited data.  Once  a decision was
made that there was  a possible source  of pollution present,
additional  sampling using  a higher confidence  level would be
initiated.   The  first stage could drop to a confidence level  of
around 80  to 90 percent.   The second stage should  rely on
confidence levels of 95 percent  or higher.

     As previously stated coefficients  of variation of different
soil parameters  can range from a low of a few percent  to a high
of  several  hundred percent.   For  example,  a study  of soil
variability done  by Beckett and  Webster (1971) indicated that it
was  not uncommon for coefficients ranging from 23 to  84 percent
to occur within  the same soil  series.  The use of a  precision of
less  than  +20  percent  for  parameters related to transport is
probably unrealistic for any field soil sampling effort.  On  the
other hand,  parameters related to static soil properties such as
percent clay, bulk density,  etc.,  can have coefficents of
variation less than + 20 percent.

                              26

-------
Emergency Cleanup


     The emergency  site sampling  is designed to identify those
areas where  soil is contaminated and  requires  containment  or
removal. A numerical sampling criterion is usually developed for
determining  which areas are clean.   Frequently  the techniques
used  for making  these determinations are rough field  tests that
lack the refinement of the laboratory analyses.  Grids are  often
used  as a  basis  for surveying the  site.  These situations offer
an opportunity  for more definitive information to be obtained  if
adequate QA/QC  procedures are built  into the sampling efforts.

     The use of grid sampling increases the likelihood  of finding
the presence or absence of a chemical deposited in the soil.   By
adding randomly placed duplicate or  triplicate sampling locations
at 5 to 10 percent of the locations,  the  precision of  the data
can  be determined.   Sampling replication coupled with appropriate
quality assurance checks in the laboratory, insures that the data
meet some predetermined level of precision and accuracy.

     Analyses  of splits of  the samples  will  provide data to
measure the  within sample variation.  Replicate extractions from
a  subset of the  samples provides a means of measuring the
variation introduced  by the extraction procedures.   Replicate
analyses of a  subset of the extractions provides a measure  of the
analytical component  of the total variance.   These measures  of
the  components  of  variance  provide information  that  can be
extremely valuable in designing any subsequent studies.

     The need  for  increased precision  occurs  as the  levels
approach an  assigned  threshold of significance or a predetermined
reference level such as a cleanup action level.  A sample  that is
highly contaminated needs little QA/QC  to determine if it exceeds
background; but, a sample level close  to background must  have  a
high precision  in  order  to detect  a  difference.  An attempt
should  be  made to  provide  a field measurement  that  meets  a
reliability that has  at least a 90 percent confidence level  and  a
precision of at least 20 percent.


Remedial Response  Studies


     These   studies  by their nature may end  up in litigation;
therefore, a confidence level of 95 percent  or better  should be
used to provide the  quality of data that is  needed.  The areas to
be surveyed  should be stratified  and sampled according to  a
design  that can  be  used  to determine spatial variability as well
as concentration ranges.   The  experimental  design should provide
                               27

-------
for an  identification and  evaluation  of  the components of
variance  for the study  situation.  This  requires that a suitable
statistical design  be formulated and  that appropriate QA/QC
procedures be implemented.

     Studies conducted for  remedial  response purposes often
attempt  to determine pollutant  behavior  and  pollutant
distribution in  order to provide a permanent remedy  preventing
the migration  or release of  hazardous substances into  the
environment.  The experimental designs  used in these cases should
be reviewed by a statistician and  should be  focused on those
areas  that are  required to meet the  specific objectives of the
study.   The objectives  should be distinctly defined.   If several
studies  are being  addressed with one  set of samples,  priorities
should  be set to insure that the main purpose of the  study is  not
lost because of interference or statistical confounding.

     If the sampling during exploratory or emergency response
investigations has been done properly, there will be a sound
basis  for determining  the  sample  size and sampling  site
distributions.  Where  pollution migration is  suspected,  the
design will have to  incorporate information on the vertical
distribution as well as the horizontal distributions.    It is  not
enough to sample the  surface soil when the pollutant is expected
to be moving down with  the permeating water.  Nested  designs  are
required to provide the vertical components of the variation.


Planned Removal Studies


     These studies are usually continuations of those initiated
during  emergency cleanup studies.   They are conducted in  more
detail and should  be designed  to provide specific information
needed to resolve  control  option issues.   The  number  of
replications may have to be increased if there is  a  need for a
higher  level of precision.   Greater care may  be  required in
determining the  location and distribution of sampling sites in
order to provide the detail that is  needed to test a specific
hypothesis.

     An attempt  at cost recovery is a likely successor to these
studies.  The QA/QC must be adequate in  order to insure that  the
analytical and sampling methods and the statistical  designs will
withstand detailed examination that results when litigation is
involved.   The  confidence level should  be 95 percent or better
and the precision should  be reduced  to 10  to 20 percent if
possible.   Preliminary data  can be  used to determine if  this
precision reduction can be accomplished  within reason.  A  more
sophisticated statistical design may have to be implemented  that


                             28

-------
will  provide  increased precision without increasing  costs
(Hammond,  et  al., 1958).

Monitoring

     Soil monitoring is  often associated with  measuring the
changes that  occur with space and time.   Monitoring studies  may
be  undertaken  as  part of  a  litigation  effort or they may be
undertaken in order to determine  the trends in a pollutant's
concentration.   Precision  and confidence  levels will have to be
adjusted to match levels  required to achieve  the objectives of
the studies.   Some form of  staged or split plot design may be
needed in order to be  able  to  determine  the components of
variance.   Techniques  such as intervention  analysis (Hipel, et.
al., 1978) and  trend line analysis  will  aid  in  determining if
there have been  changes in the  concentration patterns over time.
                              29

-------
                         CHAPTER 3



                 STATISTICAL CONSIDERATIONS
INTRODUCTION


     This chapter reviews the role of statistics in the QA/QC
process.   Statistics deals  with the mathematics of  collection,
analysis and interpretation of data.  Without statistics there
would be  little basis for determining if a particular  sample was
reliable.  There are numerous texts and references dealing with
statistics.  A number of these references have direct  bearing  on
the  collection of soil samples. The techniques presented in these
references will  not be  discussed  in detail.   The user  is
encouraged to utilize the  referenced materials  if additional
and/or more detailed information is required.   This chapter will
emphasize  those techniques  that  have  direct  bearing  on
determining the precision of the soil sampling and the  resulting
analysis.

     Box  (1974)  stated,  "Environmental data are usually highly
variable  and it is by facing this fact, rather than running away
from it, that we solve  some of our problems."  He gives credit
to Sir Ronald Fisher for the concept that "...we can  exploit the
patterns of natural variation in data to  design  enquiries and
experiments so that errors are  minimized."  Techniques for
exploiting natural variability in soil sampling quality assurance
are presented below.  The basic techniques that have  a bearing  on
soil sampling  involve the  use of statistical designs that are
capable  of providing  the information needed  to perform  a
component of variance analysis.

     The  component of variance analysis enables the  environmental
scientist to determine the  amount of variation that is associated
with the soil  itself, and  the variations associated with the
collection process, handling processes  such as  sample shipping,
storage, and preparation, and the analyses.  Box  (1974) sums up
the role  of the component of variance  analyses by  stating,  "An
appropriate study of components of variance — how much variation
is associated with chemical analysis, how much with  the sampling
method,  how much with change of location..., together with the
knowledge of how much it will  cost to  take a sample and perform a
chemical analysis — enables  us to devise a testing scheme which
                              30

-------
can be  dramatically more accurate and economical than one naively
chosen." Most soil sampling efforts fail  to provide  for the
needed  statistical control and thus fall into  a category of  being
one of the studies that Box would classify as  a testing scheme
that is  "naively chosen."


DISTRIBUTION OF SOIL SAMPLING DATA


     All statistical sampling plans are based upon the frequency
distribution of the data collected by the environmental scientist
and/or  they are based upon the pattern that is anticipated  after
reviewing data collected under similar conditions.   One sample
collected  from  a  location tells us little  about the reliability
that we  can place upon the number that results  from the  analysis
of  that sample.   A number  of  samples taken from that location
provides information on the variation of the  values that can be
expected if all possible samples of soil were collected from the
same location.  The variation in the data is  as much  a  property
of  the  soil as is the amount of material upon which the analysis
is done.

     The environmental scientist can obtain information on the
distribution of the  values likely to  be found in  an  area by
conducting an exploratory  or  pilot  study.  The  exploratory
studies  conducted during the initial phases of  an investigation
can provide an indication of  the site  specific  frequency
distribution pattern.  The environmental scientist is  interested
in  finding the location of pollutants; therefore, the pilot  study
should provide information on both contaminated areas  and areas
of  background soils.   EPA's  Regional Laboratories and  EPA's
National Enforcement Investigation Center in  Denver, Colorado can
provide information concerning the  frequency distribution of
background samples and also information  on the distribution of
analyses performed on  contaminated  soils. Another source of
information concerning  distribution of  soil  pollutant
concentrations  is the  report, Environmental Monitoring at Love
Canal (USEPA, 1982).

     The two most common frequency distributions encountered in
the  soils  literature  are the  normal and  the log-normal
distributions.   Where discrete events  such as  decay of
radioactive particles occur, the Poisson distribution is the  more
common  fequency distribution  (Kempt home and Allmaras, 1965).
Rao, et  al.  (1979) have reviewed the frequency distribution of
spatial variability for soil physical properties.  They indicate
that "soil  properties such as  bulk density, organic matter
content,  clay content, and soil-water content at a given tension
are generally characterized by normal distributions....  However,
flow  related soil properties such as air permeability, saturated
hydraulic conductivity, soil-water flux, pore-water velocity, and
                             31

-------
solute dispersion  coefficients  have been reported to be
log-normally distributed."  These two  distributions are  "the most
frequently observed statistical distributions for describing  the
spatial variability of soil physical properties."

     The  distribution pattern of chemical constituents of soil
including pollutants is truncated at  zero.  This characteristic
often gives rise to a frequency curve that is skewed  to the  low
end  of the concentration scale.   When this  situation is
encountered the  data  can usually  be transformed  to a  normal
distribution by taking the log of the  data.


STATISTICAL DESIGNS


     The  design of the sampling study must be determined before
the sampling is undertaken.  Frequently, improper designs contain
factors that are  correlated with each other in  such  a fashion
that the sampling itself determines the  outcome of  the
investigation.   This  is in  part  avoided  by  proper  use of
randomization  during the collection process.   Designs such as
those  outlined by  Mason  (1983) or Peterson and Calvin  (1965)  can
greatly aid in  determining the components  of variation  likely to
be encountered  in the  soil sampling effort.

     The soil  scientist  often wants to provide information on the
spatial  distribution  of the  concentration  of  a particular
constituent of the soil.   Techniques such as those outlined by
Davis  (1973)  for  Kriging, trendline analysis,  discriminant
function analysis,  cluster  analysis  and Fourier  analysis
provide  the needed  information  for evaluating the spatial
patterns  in an area.   Kriging provides one benefit that is useful
from the  quality  assurance point  of view.   Kriging has  the
capability of providing  a  reliability map of the area under
investigation.   This map  is  generated from the fact that
pollution samples located  close to  each other are often  related
because  of their  position  within a  pollutant plume  or  within
their original geological  deposition  sequence.  A series of
articles  by Burgess  and Webster (1980a), Burgess and Webster
 (1980b),  Webster  and Burgess (1980), and Flatman (1984) give a
good presentation of  the use of Kriging in  soil sampling  and
discuss  the use of the error map.  Mason  (1982) also reviewed and
described the  use  of Kriging and its application to soil sampling.
Campbell  (1978, 1979)  discusses the use  of discriminant function
analysis  and  autocorrelation in evaluating  the spatial
distribution of soil properties.

     Soil  sampling  faces a spatial distribution  problem of
another sort in the fact that sampling with depth is required of
the data  collection effort.  Purely random samples  cannot be
taken  once the investigator penetrates the surface of the  soil as
                              32

-------
each sample  is  now composed of subsamples.  Some form of nested
design  is  required  in order for the investigator  to handle  this
problem.  Table 3-1 shows the analysis of variance that would be
undertaken for  this type of soil data  (Snedecor and Cochran ,
1982, p.  248  with modifications).


Type I  and Type  II  Errors


     Statistical sampling plans  are designed not only to measure
the  components  of  variance but also to  aid  the  environmental
manager in making  informed decisions about  contaminants  found
during an investigation of a site.  The most  common decision
would  answer the question, "Is the area contaminated  or  not?"
The environmental scientist should  design an experiment that  will
provide data for testing the hypothesis  that the study area mean
value is  equal to the background mean value.   Mathematically  this
hypothesis is expressed as


                Ho  : (yS * ^B  I°S = °B >
where  y  =  mean and o^ = variance, the  subscripts S = study area
and B = background.  If this hypothesis is  true then there is no
difference between the two areas.   If  the  hypothesis is false
then the alternate  hypothesis given below is used.


                   Ha : (ys * PB }
       The alternate  hypothesis means that the  two  mean  values
are different.   A similar test  can  be  used to determine  if the
pollutant in the study area  soils  equals  some action level  such
as a clean up level  or an environmental  standard.

     The  tests  of  the hypothesis are made  on  the basis of  either
the  normal distribution  or a  normalized  distribution.  The
hypothesis is  accepted or  rejected on the  basis of a comparison
of a  sample mean with the  values  delineating the  acceptance
region  (See Figure 3-1).   Definition  of the acceptance  region
includes  the acceptance of a probability  that the sample  mean
lies  outside of the  acceptance region (in  the shaded portion  of
the  distribution shown in  Figure  3-1).   This probability  is
defined  by the significance  level, the confidence level  or the
Type I error (all three terms are synonymous).


                              33

-------
 TABLE 3-1.  ANALYSIS OF VARIANCE OF A NESTED SOIL SAMPLING DESIGN.
Location
i-l,...,a
1


2


3


4


Depth
j-1, ..
1
2
3
1
2
3
1
2
3
1
2
3
Ana 1-1
. ,b X
3.28
3.52
2.88
2.46
1.87
2.19
2.77
3.74
2.55
3.78
4.07
3.31
Ana 1-2
ijk
3.09
3.48
2.80
2.44
1.92
2.19
2.66
3.44
2.55
3.87
4.12
3.31
Sum- Anal
Xij
6.37
7.00
5.68
4.90
3.79
4.38
5.43
7.18
5.10
7.65
8.19
6.62
Sum-Loca Total
X£.. X..


19.05


13.07


17.71


22.46 72.29
I.   C - (X...)2/(abn)  - 72.292/24 - 217.7435




II.  Analysis:  £X£jk2-  C - (3.282+..+3.3l2)  - C - 10.2704




III. Depths:  ZXij2/n -  C - (6.372+...+6.622)/2 - C - 10.1905




IV.  Locations: IXi..2/bn - C - (19.052+...+22.4&2)/6 - C - 7.5603




V.   Depths in Locations - III-IV-10.1905-7.5603-2.6302




VI.  Analysis of Sample - II - III =10.2704 - 10.1905 - 0.0799






                             ANOVA TABLE
Source of Degrees of Sum of
Variation Freedom Squares
Location
Depths /Location
Analysis/Depths/
Location
Total
3
8
12
23
7.5603
2.6302
0.0799
10.2704
Mean
Square
2.5201
.3288
.067
Components of
Variation
VA +
VA +
VA
nVD + bnVL
nVD
n « 2,b » 3,a « 4,s2 « 0.0067 estimates V^ or variance due to analysis




SD2 . (o. 3288-0.0067)/2 - 0.1610 estimates VD or variance due to depths




,L2 . (2.5201-.3288)/6 - 0.3652 estimates VL or variance due to  location
                                     34

-------
                              ACCEPTANCE REGION
                   202              /. • 30.0
Fig. 3-1   Acceptance region for H. po = 30.0.
                                                       39.8
                    ti = 10.0
  HE 3-2 - T)T>e H or P  error.
20.2
30.0
39.8
                                           35

-------
     The  Type I error whose probability is denoted by a, is the
case where we reject a hypothesis when  in  fact  it is  true.   A
second error called the Type  II error, whose  probability  is
denoted by 0 , is the  case where  we accept a hypothesis when  in
fact it is  false.  The values of ot and 8 interact with each other.
When ot goes up 8 goes  down.  The  two  types of error are defined
in terms of  their probabilities and  can  be  controlled to desired
levels by appropriate  sampling.

     The Type  I  error  is most frequently encountered  in
statistical tests  used  in the  literature.   The significance  or
confidence levels most often  used are the  90, 95 and 99 percent
levels.  The Type II error  (false  negative)  is also used  to
identify  the  power of  the  test which is equal to (1-B).

     Figure  3-1 presents the acceptance region  for  testing the
hypothesis H:(ys  = 30.0).  The  shaded portion represents the
probability  of  a  Type  I  error  (Juran et al., 1979). Figure 3-2
shows a  situation  where a sample with a  mean of  10 is  compared
with  the distribution  presented in  Figure  3-1.  Figure 3-2 is a
graphic  presentation of  the  hypothesis that  the mean of our
sample   is equal  to  30.   The  shaded portion of  this  figure
represents  the portion of  the distribution that would give a Type
II error  (Juran  et al.,  1979).


Number and  Location of Samples

     One of  the most useful tools for increasing the reliability
of  the data  collection process  is  the  replication of  samples.
Appendix A  provides a set  of  tables  and a graph for use  in
determining the  number of  samples needed  to  detect a difference
between  samples  or  to estimate the standard deviation with a
precision (p) at a particular confidence  level.   A computer code
developed  by Heimbuch  (1982)  is also available for calculating
the  sample  size.  The location of these samples should  be
determined  by  use  of some  form of random selection  process.
However, samples  are collected  from a  systematic grid pattern
when  Kriging will be employed in the analysis of the  resulting
data.

       Further gain is made in  increasing  the precision  of the
analytical  results from the  sampling effort if the sample design
can  incorporate  stratification of sampling locations.   This
technique  makes use  of  the observation  that certain locations
tend  to  be  similar  in  their properties and they  in  turn are
different  from  samples  taken in other  locations.  A typical
stratification  used in soil science is the soil type. Soil types
are chosen because they  represent an assemblage of soil units
that are  closely  related  in their  physical and  chemical
properties.
                              36

-------
Role of  Quality Assurance in Experimental Design


       The  Quality Assurance Officer  should  be  intimately
involved in  the review of the experimental or sampling design
proposed by the  investigator.  He  should insure that  the
information  obtained provides measures  of the  components  of
variance that are  identified in the field.  An additional quality
check that should be undertaken as part of the QA  program is  the
review  of "the design by qualified soil scientists and  other peers
that are in  a position to provide the necessary oversight of  the
sampling effort.


     Broms  (1980) makes the following statement;  "There should
be a balance between the soil investigation method,  the quality
of  the soil samples, and the care  and skill spent on  the
preparation and the testing of the samples.  There is no point in
spending time  and money on  careful  sample preparation and on
testing if the quality of the samples is poor."   The QA program
must address the total flow of information from the  design to the
reporting of the results.  The sampling design is  the foundation
of  the  whole study, therefore, it should  be given maximum support
if the purposes of  the sampling effort are  to be met.


Components of Variance


       The  component of variance analysis  provides an  estimate of
the  sources of variation that contribute to the total variation
seen in  the sampling.  It therefore provides an estimate of the
variance of the analytical process,  of  the sampling and of  any
other factors  incorporated  in the experimental design.  Bauer
 (1971) and Snedecor and Cochran  (1982)  discuss  the use of  the
components of variance analysis.  An excellent example of the use
of this technique  is provided in a report  by  the  Electric  Power
Research  Institute  (Eynon and Switzer,  1983) on the sources of
variation encountered in using the RCRA  extraction procedure on
utility  ash.   The example  presented  in  Table 3-1  gives  the
components of variance  for hypothetical  sampling data.   This
technique  can only be used  if the design  of the experiments is
done in such a manner that  the variation  due  to  the  parameters
can  be  evaluated.    This  is accomplished  by appropriate
replication at each level  in the  design.
                               37

-------
Compositing of Samples


       One of the  main techniques used to  reduce sampling  and
analytical costs is the use  of the composite sample.   Combining
the samples from several sampling locations reduces not only  the
costs  but also the variation of the mean  obtained by  sampling.
The technique  is  used extensively by agricultural workers when
estimating the amount of fertilizer to be  placed on a farmer's
fields.   Since  the primary  purpose of QA/QC  is to measure  the
precision of  the samples obtained, this technique  should  be
avoided  if  at all possible.  Peterson and Calvin (1965) make  the
following  statement about this technique:

       "It should  be pointed  out  that the  composite
       samples provide only  an estimate of the mean  of
       the population  from which the samples  forming
       the composite are drawn.  No estimate  of  the
       variance  of the mean, and hence, the  precision
       with which the mean  is estimated can be  obtained
       from  a  composite of  samples.  It  is  not
       sufficient to analyze two or more  subsamples
       from  the same composite to obtain an estimate of
       the variation within the population.   Such a
       procedure would  permit  the estimation  of
       variation among subsamples within the composite,
       but not the  variation among samples in the field.
       Similarly, if composites are formed from samples
       within different parts of  a  population, the
       variability among  the parts,  but  not the
       variability within  the parts, can be estimated.
       If an estimate of the variability among  sampling
       units within the population  is required, two or
       more  samples taken at  random within the
       population must be analyzed separately."

     Youden  and Steiner   (1975)  caution against the use of  the
 composite sample  for much the  same  reasons as  those outlined
 above.  There  is no measure  of the precision,  and  therefore  the
 reliability,  of  the data obtained.


 Split Samples, Spiked Samples and Blanks


     Split  samples, spiked samples  and  blanks are used to provide
 a measure of the  internal  consistency of the samples and to
 provide an estimate of  the components of  variance and  the bias in
 the analytical process.
                              38

-------
     Samples can be split  to:

     o     Provide samples to both  parties in a litigation or
          potential litigation situation.

     o     Provide a measure of  the within sample variability
          (this is needed in order to determine the influence  of
          other factors that may  be confounded with  sample
          splitting.)

     o     Provide materials  for  spiking  in order  to test
          recovery.

     o     Provide a measure of  the sample bank and extraction
          error.

     The  location of the sample  splitting  determines the
component  of variation  that is measured by  the  split.   A split
made in the sample bank  measures error introduced at that level.
A split made in the field  measures field handling along with the
within sample variation.   A split or series of subsamples made in
the laboratory for extraction purposes measures  the  extraction
error.

     Spike samples are  prepared by adding a  known amount of
reference  chemical to  one of  a  pair of  split samples.  The
results of  the  analysis  of a split compared with the non-spike
member  of  the split measures the recovery of  the  analytical
process and also provides  a measure of the analytical bias.

     Spike samples are  difficult to prepare with soil material
itself.  Frequently the  spike solution is added to the extract of
the  soil.  This  avoids  the problem of mixing, etc. but does not
provide a  measure of the interaction of the chemicals in the soil
with  the spike,  nor  does it provide  an evaluation of the
extraction efficiency.

      Blanks provide  a  measure  of  various cross-contamination
sources, background  levels in  the reagents,  decontamination
efficiency and  any other potential error that  can be introduced
from sources other than the sample.  For  example,  a  trip blank
measures  any contamination that may be introduced into the sample
during  shipment of containers from the  laboratory  to the field
and  back to the  laboratory.  A field blank measures input from
contaminated dust or air  into  the sample.  A decontamination
blank  measures  any chemical that  may have been  in the sample
container or on  the tools  after  decontamination is completed.

     The  number of QA/QC samples have been selected  by a rule of
thumb that one out of every twenty samples is to be  assigned to
each of the categories of  samples.  This ratio  has been used
successfully in  several major  USEPA studies (USEPA,  1982, 1984).
                              39

-------
Table  3-2 presents  the  breakdown of QA/QC samples used in  these
previously conducted monitoring studies.


DATA ANALYSIS


     The  topics that follow  are designed to  provide insight into
the use  of statistical  techniques for evaluating  the  data
obtained during  an  investigation.   They  are not by any  means
exhaustive, but  are designed to provide the basis for  designing
the  quality assurance  portions of a sampling  effort and  to
provide the basis for  obtaining the most  benefit from  the  data
acquired.


Bias


     The  variation  seen in analytical data can  be composed  of
variation within the  sample  itself, variation introduced  in
sample collection or preparation and variation in the analysis  of
the samples.  The variation can further  be divided  into sample
variation and  bias.  Bias  identifies the  component of  the
sampling error that causes the mean value of  the sample data  to
be either higher or lower  than the true mean  value of the samples.
An example of a  bias would be  the error in  analytical results
introduced by  an instrument being out of calibration during a
portion  of the analysis.   Laboratories  usually introduce
reference samples  into their  sample  load in  order to detect  these
changes.  Bias in soil sampling  is  difficult to  detect.   The
presence of bias can be proven by use of one of the techiques
described below. On  the other hand it is difficult  to prove that
bias is  not present  because the  absence  of bias may be  the result
of the inability to measure it rather than its actual absence.


Standard Additions—   It  is  necessary   to conduct special
experiments in order  to detect bias in the sampling  effort.    The
major technique used is that of adding  known amounts of standard
solutions to the samples:  it is recommended that this  be done  in
the field or in a  field laboratory.   The main problem encountered
is that mixing soils  to obtain  homogeneity  is difficult  in a
laboratory  much less in the field. Several known quantities of
the  standard are  added  to  samples taken  in the field.   The
results  should follow the equation for a straight line:

                           y « a + bix

Bias is  indicated  if  the  data  do not follow the straight line
equation, or if  a  < 0, and if precision  errors are  small.  If  the

                              40

-------
         TABLE 3-2.  QA/QC PROCEDURES FOR  SOIL SAfFLES
     Procedure
1.  Field Blanks
        Comments

One for each sampling team
per day.  A sample container
filled with distilled,
de-ionized water, exposed
during sampling then analyzed
to detect accidental or
incidental contamination.
2.  Sample Bank Blanks
The field blank, about 40% of
them, passed through the
sample preparation apparatus,
after cleaning, to check  for
residual contamination.
3.  Reagent Blank
One for each 20 samples  to
check reagent contamination
level.
4.  Calibration Check  Standard
One  for each 20 samples  to
check instrument calibration.
5.  Spiked Extract
6.  Spiked  Sample
 7.   Total  Recoverable
One  for each 20 samples  to
check  for extract matrix
effects on  recovery  of known
added  analyte.

One  for each 20 samples. A
separate aliquot of  the  soil
sample spiked with NBS Lead
Nitrate to  check for soil and
extract matrix effects on
recovery.

One  for each 40 samples, a
second aliquot of  the sample
is  digested by a more
vigorous method  to  check the
efficacy of the protocol
method.
                                         41

-------
                           TABLE 3-2
    Procedure

8.  Laboratory Control Standard
                                                 nts
9.  Re-extraction
 10.  Split  Extract
 11.  Triplicate  Sample
 12. Duplicate Sample
One for each 20 samples.  A
sa«ple of NBS River Sediment
carried through the
analytical procedure to
determine overall method
bias.

One for each 20 samples.  A
re-extraction of the residue
from the first extraction to
determine extraction
efficiency.

One for each 20 samples to
check injection and
instrument reproducibility.

One for each 20 samples.  The
prepared sample is split  into
three portions  to provide
blind duplicates for the
analytical  laboratory  and a
third replicate for  the
referee  laboratory  to
determine  interlab  precision.

One  for  each 20 samples  to
determine  total random error.
                                          42

-------
units of x and y are the same, the  value of b]_ should  be unity;
and significant deviations from unity  indicates  a  proportional
bias (Allmaras, 1965).

     Occasionally a  laboratory will prepare "spikes" using only
the standard  solutions.   This should  be avoided if possible
because of  interactions that may  occur between components of the
soil matrix and the chemicals under investigation.   Radioactive
isotopes can provide a useful indication of the recovery of the
analytical process.


Internal Consistency—  If several  samples of soil  of different
size are analyzed  for a constituent,  the  results  should fit  a
linear equation of the form:

                           y = a + b2Z

where  "Z" is the quantity of sample analyzed.  The amount of
chemical detected  should be directly related to  the quantity  of
the  sample analyzed.  The equation should be linear;  if not, bias
is indicated.  The "b" term depends upon the proportion of  soil
used.   The value  of "a" should not be less than zero.  A negative
value for "a" and a constant value  for  "b" indicate  a negative
bias.   Consistency measurements  can be combined with the method
of standard additions in  order to provide a  more  definitive
evaluation of bias.
Analytical Procedures--  Analytical methods that make use of
different basic theories can provide  information on the presence
of bias.  This  is especially  true if one  of the  methods is
considered to  be the "standard" for  the industry.   An example
from  wet chemistry would be to make use of a colorimetric and a
gravimetric method.  Bias is present  if  the results of the  two
methods do not agree.

     The  use  of referee laboratories can also aid in determining
the reliability of  the data and in detecting bias in  the analysis.
Techniques such  as charge balance,  summation of parts (more
important in physical analyses), comparative values and  simple
examination  of  the data can  all  provide  information on the
analytical reliability of the sample  results.


Confidence Intervals


     One  of  the major  calculations performed on the data is the
calculation of the  confidence interval (CD.  This value is  used
in much of the  work undertaken in QA/QC.   It is used in the
development of the  control charts, in identifying outliers  and in
                              43

-------
determining if a  set of samples exceed  some pre-identif ied
standard.   Two approaches can  be used for  calculating  the
intervals.   Where statistical  designs  have been used in the
sampling and analysis of  the samples, the Analysis of Variance
(ANOVA)  provides  the best method for obtaining  the needed
information for calculating  the confidence interval.    An
approximation of  the interval  can be made by use of  the range
between or among two or more samples in a series.   Bauer  (1971)
presents  both techiques.  When the  number  of  replicates is
limited,  i.e., two or  three, the range may offer a better
estimate of  the confidence  level.

     The confidence interval is bounded by the confidence limits
(CD.  The confidence limits are "the bounds of uncertainty about
the  average caused by the variability of the experiment" (Bauer,
1971).  The  limit is defined by the following equation.


                          CL = x + ts/v/n~


where x  * mean, s = standard deviation, n = number of samples and
t * Student's t value at  the desired level of confidence  and  n-1
degrees  of freedom (See Appendix B, Table B-l, for values of t).
The range  can also  be used to  calculate the  CL by using  the
following  equation:


                          CL » x + AIR


where I R = the sum of the k ranges and A = a factor derived from
the relationship between  the range, the standard deviation and t.
The  values of A are presented in Appendix B, Table B-2, but can
be calculated from  the following relationship:


                          A =
where  k = the  number of  groups of data and n the  number of
samples in each group.   The values  of GI  and the equivalent
degrees of freedom for t  are obtained  from Appendix  B,,  Table
B-3, which is based upon the  relationship s = R/CI, the  defining
equation for the  relationship between standard deviation and
range.

     The tolerance limits  (TL) are similar to the CL but are used
to identify the interval and  limits into  which the individuals  of
                             44

-------
the  population should fall.   The defining equation for this value
is:
                           TL = x + ts


     or                    TL - x + IZR
where  lisa  factor relating the  range and  the  standard
deviation.   The value of I is obtained  from Appendix  B, Table
B-4, or can be calculated from the  relationship:
Outliers


     Dixon  (1953, 1965)  has provided techniques  for identifying
data points  that  lie outside acceptable  limits and  thus can be
considered  to  be points  from a  different  population.   The
differences  may be due to bias,  cross  contamination  or other
factors that have altered the  chemical content of  the sample or
the  results obtained  by  the analysis of the  sample.   One
technique  that  is identified by Bauer (1971) makes use of the
range for determining if a particular value is a true outlier or
merely a value  on the edge of  the particular distribution,   the
statistic


                          ti = |  x - x| /R


is  used to  make the decision (x =  mean).  Appendix  Bf Table B-5,
lists the values  of ti that are the  critical values  for various
sample sizes.


Propagation  of Errors


     Consider  the case where  several measures  (each with some
associated error) are made on the same  or matched  samples  and
then an estimate is calculated from the measured values.  In this
case, when the measured  values are statistically independent  and
the errors  normally distributed,  the  following equations may be
used to estimate  the variances of the calculated values, where  x
                              45

-------
and  y  represent the measured  variables, s represents the standard
deviation of  the  measured  variables, and  f (x.)  and f (x,y)
represent the calculated  estimates:


For Single Valued Function  of  an Observation;

   f (x)	Estimated variance of f (x)

   In x            s2/x2
   sin x           (cos2x)s2

   xP              P2x2(p-l)s2


For More Than One Attribute on  the  Same  Sample;

   f (x,y)	Variance of  f (x/y)	

   x + y           sx2+sy2+2sxy

    xy          y2sx2+x2sy2+2xysxy

    x/y        [sx2+(x/y)2sy2-2(x/y)sxy]/y2

   In(x/y)      (l/x2)sx2+(l/y2)sy2-(2/xy)sxy

   ln(xy)        is true.
 The  following  equation  is used to calculate the t  value.  This
 value is compared with the value  of  t taken  from the table  in
 Appendix B.
                               46

-------
           X  - X
            1	2
      S =  s (— + .1) 3j
             n i  n2

             n s 2    n  s  2  ,
where s =  t  V   n 2  *  1
                1 +   2-2

and  xi and  X2 are  the mean values of the  two samples and ni  and
n2 are the number  of  samples collected  from  the  respective
populations.   s is the pooled sample standard deviation and  sj.2
and S2^ are the individual  sample  variances.   In the  example
given previously,  the two samples were from the background  and
the study area.  If  ts  < t at an assigned confidence level for  ni
+  n2~2 degrees of  freedom, then the difference between the means
is not significant at that confidence level.   If ts > t, then  the
difference is significant at that confidence level.

     A similar  equation, given below, can be used to test  the
comparison between the  sample mean and some  action level  such  as
a  clean up level.   An  example of a clean up level would be the 50
ppm level used for PCB's in soil.

                     t  = X - AL
                       c
where AL = the action level s is the standard  deviation (s) and
n is the number of  samples.  A one-tailed  test is used in  this
case  because  interest  is only in those cases  where x exceeds the
action level.   If the calculated  tc  value  exceeds the t value
taken from the table  in Appendix B for (n-1) degrees of freedom
at an assigned  confidence level, then, the  hypothesis  is false
and the sample  exceeds the action level.

Example:
     A  preliminary  study  is  done in an area suspected of being
contaminated with polychlorinated  biphenyls   (PCB's).   Sixteen
soil  samples  were collected from both the study area and from a
background area.  Table  3-3 lists the data and  the analyses  done
on the data.
                              47

-------
TABLE 3-3.  PCB STUDY TO DETERMINE CONTAMINATION OF AN AREA
            (HYPOTHETICAL DATA)

           Background Area (ppb)          Study Area (ppm)
           35.8       38.5                47.0     50.0
           45.5       36.0                62.0     49.6
           35.5       40.5                47.0     53.5
           32.0       35.5                59.5     68.0
           50.0       45.5                40.0     60.0
           39.0       37.0                57.5     45.0
           37.0       36.0                48.5     42.5
           47.0       53.0                53.0     58.7


XB = .0402 ppm    SB2 = .000037       nB = 16    CVB* = +15.1%
xs = 52.61 ppm    SS2 = 60.2598       "s = 16    CVg  = +14.8%

*CV - Coefficient of variation  in %


     This  example does  not  need the use  of statistics to
determine that there is a difference between  the two  population
means.  The t value  is calculated as follows:

      r  16(.000037+60.2598) *h
s *   l       16+16-2      J
 s .  /32.1386 « 5.6691


    m   52.61 - .0402  ,  26.2281
  C     5.6691/-27IT

 the t  value  for 99 percent  confidence  level  and a  one  tailed  test
 is 2.131  for  15 degrees  of freedom  (dof).   A  value  of 26.2281 is
 so  large as  to be  improbable  in samples  of  this size  so  the  null
 hypothesis is rejected,  i.e., the samples  with mean of 52.6  come
 from a  different population  than the  background samples.

     A  more  appropriate use of  this test  would  be to decide  if
 the study.area exceeds  the 50 ppm clean up action level.   The tc
 value is  calculated as  follows:

       52.61 - 50.00  _  1.344
  *c   7.7626X
 and from Appendix B,  tgo%,  15 = 1.341;

                                48

-------
this  indicates  that the study area concentration may be different
from the value  of 50 as  this value  of tc  could occur in 10
percent of the  samples.  Since the study  area mean may exceed  the
50 ppm action level,  we could make  the decison to  clean the
entire site.   This is not economical  because it is apparent that
some locations exceed the value and others do  not.   A  statistic
can  be calculated that is analogous  to the tolerance limit (TL).
We can assume that the standard  deviation  of the  true mean
(SOppm) equals the sample standard  deviation.  This yields  the
following equation:

                       TL1  =  y  + Ks
where K =  the normal deviate for the selected confidence level
which may be obtained  from the t table at infinite dof.  Assuming
a  95% confidence level  the TL1 becomes 50+(1.96)(7.7627) or 34.7
to 65.2.   In  this case  the  TL  =  y +  ts  =  52.61   +
(7.7627) » 52.61  +  (1 .753) (7.7627)  = 39.0 to 66.2  so
there is little difference between TL and TL' .  Only  one value
exceeds the  50 ppm  level  if one uses  either  the modified
tolerance limit (TL1)  or  the tolerance limit (TL).

If one decides  that the upper limit cannot  exceed 50 ppm under
any circumstances, it  would be necessary to  calculate  the range
of values  that could be considered equal to or  less than 50.  On
the basis of the lower limit therefore all of the samples would
lie within the  range and would have to be removed.   This latter
case  is the more conservative  but also the more costly  decision.
                               49

-------
                          CHAPTER 4
                      EXPLORATORY STUDY
INTRODUCTION
     Once objectives have been  defined which involve  the need for
soil sampling the next step is to develop a  total  study protocol
including an appropriate QA/QC  program.  In order  to  develop this
protocol answers to the following questions  must be available or
estimates must be made.

     o   What are  the likely sources  of the  pollutants of
         concern?
     o   How  have these sources varied  in the  past compared to
         their present emissions?
     o   What are  the important transport  routes which
         contribute  to soil contamination?
     o   What is the geographical extent of the contamination?
     o   What  average concentrations of the pollutants exist at
         different locations and how do these vary as a function
         of location and time?
     o   Do localized areas of high concentrations  exist and if
         so, where are they and what are their concentrations?
     o    Is it  possible to stratify the sampling region in such
         a  way as  to reduce  the  spatial variations within
         strata?
     o   What are  the soil characteristics, hydrogeological
         factors, meteorological or climatic  factors, land use
         patterns,  and agricultural practices affecting the
         transport and distribution  of the pollutants of concern
         in soil?
     o   What  is  an appropriate background,  or control region
         to use for  the study.
     o   What  are the acceptable levels  of precision and both
         Type I and Type II errors for this study?

     If detailed and  specific  answers to  all of these questions
were available in advance, there would be no need  to  conduct the
study.   The recommended approach is to conduct  an  exploratory
study that  includes  both a literature and information search
along with  selected  field measurements made on the basis of  some
assumed dispersion model.  If one is dealing  with an emergency
                             50

-------
E   uation such as a hazardous chemical  spill, there usually will
not be time to proceed in the deliberate  fashion recommended here.
Thus  for the emergency situation it  is  often  necessary to
compress  all of the planning and study design  into a very  short
time  period and  then proceed  to the  final definitive study
without delay in order to afford  maximum protection to  human
health,  welfare, and the environment.   The following will assume
that there is a reasonable amount of time available to conduct an
exploratory study prior to the conduction of  the more  definitive
study.

     To be maximally  cost  effective  a major  element of the
exploratory study is a literature and information search  and an
information-seeking  series  of interviews.  Much information
pertinent to the questions asked above may already be  available
in the published literature,  in the  files of  governmental
agencies  or industrial corporations, in ongoing or  completed
research at local universities, or  in the knowledge  of local
citizens.  A carefully planned  and organized effort  should be
mounted  to accumulate  what  relevant  information is  available.
Only after this information has  been collected, collated  and
evaluated should any field measurements be  taken.  It is  a  good
policy to adhere to a reasonable  fixed period of time  for  the
collection and  analysis of available information,otherwise  this
process could drag on interminably.  Only at  the end of  the  fixed
time  period, and based on whatever information is available at
that time, should the design and implementation of  the  field
measurements portion of the exploratory study  be undertaken.

     Even though the principal  subject  of  this document  is QA/QC
for soil  sampling  it is not possible to separate the QA/QC from
the  total soil  monitoring study design.   The objectives  of the
monitoring study are the driving  force for all elements  of  the
study design  including the  QA/QC aspects.  The results  of the
exploratory  study will be a  set of information and field data
that  will serve as the basis for  the design of a definitive
monitoring study that includes  the total  integrated QA/QC  program
for  all media.   One  element  of  this total  QA/QC program will be
the soil sampling QA/QC plan.


NUMBER AND LOCATIONS OF SITES  FOR SAMPLING


     What is  desired  for  the  final definitive study  is the
appropriate  number  of sampling  sites at the  appropriate locations
to obtain data  on the  basis of  which mean concentrations and
standard deviations for the  regions  of interest may  be determined.
A method has  been described  in  Chapter 3 for calculating the
number of required sites in  a given region if one  knows the
required precision,  the standard  deviation of  the mean, and the
required levels of  confidence  (related to acceptable  levels of
                              51

-------
Type I and  Type  II  errors).   The precision and  confidence levels
must be specified, usually on  the basis of a value judgment  by  a
responsible official  of the  agency conducting  the soil sampling
study.   The standard  deviation  of  the  mean  of  the  total
population  of  soil  samples in the  study region  must be estimated
on the basis of the  standard  deviation of a suitable  sample of
the total population obtained  during the exploratory study.

     In general, a suitable soil sample from a number of possible
soil samples may be selected on the  basis of random, judgmental,
or systematic  sampling.  A major input into selecting the optimum
sampling design is the information accumulated prior to  the field
sampling phase.   Usually the  optimum approach  will  be  a
combination of  judgmental  and systematic random  sampling.
Assuming  that  appropriate information has been  obtained in the
preliminary or information-gathering phase of the exploratory
study,  a  model may  be hypothesized describing a likely spatial
distributon of  soil contamination as well as identifying a  likely
control area.   Selection of a desired number and location of
sampling sites  on the basis of a model is a judgmental  approach.
Subsequently the model must  be verified by collecting a limited
number of samples in areas outside of the suspected contamination
zone.

     For  example, suppose  it  is  suspected that an  abandoned
hazardous waste  site  is leaking wastes into the groundwater.
Further suppose  that  the groundwater is being used to irrigate
crops in the vicinity.  Preliminary information  has  identified
some  of the pollutants that  have been placed  in the  waste site
and has determined the hydraulic gradient extending from the  site
location.  The recommended sampling approach  is to establish  a
radial grid system with the center at the waste site and the  zero
azimuth line along  the direction of the hydraulic gradient.   The
largest number of  samples would then be taken along the  zero
azimuth  and  along  the  +  45° azimuth  from zero.   This is
judgmental sampling.   However to make sure that  important  data
are  not missed some additional samples should  be taken close to
the waste site  along each 45°  azimuth (5 additional directions).
This  adds systematic  sampling to take care of cases  where, for
example, some immiscible waste constituents may  be moving  in  a
direction different from the hydraulic gradient, the hydraulic
gradient  has not been  properly defined, or  there  are other
sources contributing  to the  soil contamination.  The location of
the  samples taken  along each axis and in the  pollutant plume
should be selected at  random.

     For  the  selection  and sampling  of a control area  a
combination of judgmental and random sampling is recommended.
Based on  the  available information and  the assumed  transport
model, select a background or  control region which is  similar  to
the  study area in  every important  aspect except for the expected
absence of contamination by selected waste  constituents.   Select
                              52

-------
a minimum  of 12-15 sampling locations at randoir  from  the
background area  to  obtain data for calculating  the  mean  and
standard deviation  of the background concentrations of selected
waste constituents.   It is recommended that approximately  the
same degree of effort go into the selection and  sampling of  a
control  area as goes  into the selection and sampling of  the study
area.

     The QA/QC program  for the exploratory study  need not be as
stringent as that for the more definitive  study.   Keep in mind,
however, that  reasonable levels  of precision and confidence must
be attained in order for  the resulting data  to serve as  an
adequate  foundation for further studies.   As a minimum,  it is
recommended  that  duplicate samples be collected  from at  least 5
percent  of  all  sampling locations and that another 5  percent of
all samples  be  split  into triplicate samples.  Furthermore  it is
recommended  that a  modest  number  of  additional  independent
QA/QC soil  samples  be taken  on a random  basis  at approximate
mid-points  between  selected sampling points in  regions where the
hypothetical model predicts the highest  concentrations  will be
found.   Data  from these additional QA/QC samples  will give some
measure of how well  the QA/QC plan is achieving its  objectives.

     Duplicate sample results will help  to establish precision
among different samples collected from the  same site.   Triplicate
splits  of  samples provide a measure of precision within  a single
sample, which tests  the homogeneity  of the sample.   The
additional  QA/QC samples will provide data to use in  evaluating
possible changes  in  means and  standard deviations when additional
sampling points  are added.   If  the two  groups of  samples (study
design and additional QA/QC) are statistically equal,  the samples
can  be  combined.  If not,  then  there  is  an indication of either
some form of bias in the  sampling design or one or more  errors in
the  assumptions inherent  to  the  sampling design.  This matter
must be carefully evaluated  in order to determine  if additional
sampling is needed.  In addition  to  the above QA/QC checks on
sampling, all normal  analytical  QA/QC  procedures  such as field
and  trip blanks, etc.,  should be  operative for the exploratory
study.


SAMPLING AND SAMPLE  HANDLING


      It  is  assumed  that an approved protocol will be followed for
 handling, labeling,  transporting and chain-of-custody  procedures
 for  sample containers and samples.

 •Generally a number on  the order of  6 to 15 should  be adequate
 with the exact number being determined by a value  judgment based
 on study  objectives,  site-specific  factors,  and  available
 resources.
                               53

-------
If appropriate  approved operating procedures for  these subjects
are not available, then they must be prepared prior  to beginning
the study.  Sample volumes will be specified by  the analytical
laboratory depending on the analytical methods to be used and  the
desired sensitivity.   Accordingly, principal attention will be
focused here on sampling methods, preparation of the samples  for
analysis,  and QA/QC aspects of both.

     Some  major concerns in sampling design include identifying
the required depth of sampling,  whether or not sequential samples
at  different  depths will be required, whether samples should be
composited,  frequency  of  sampling,  sample  preparation  for
analyses  and  QA/QC aspects of  all of the above.  In deciding how
to deal with these concerns one  must constantly keep in mind  the
objectives for  which the  soil monitoring study is being conducted.
The exploratory study  provides a  limited opportunity  to
investigate some of  the above subject areas experimentally to
determine what  effect the sampling parameters  may have on  the
QA/QC aspects of the  total   study.   The expenditure of modest
additional resources in the exploratory  study may  well lead to
more cost-effective designs for  the final definitive study.

     The  sampling device used  to acquire the exploratory samples
should be consistent with the objectives of the final study.   The
simplest  sampling tool  should  be used.  Where the contaminant is
believed to be  on the surface, a soil punch or trowel may be used.
If the contaminant is  soluble or is expected to be located more
than a meter below the  surface a truck mounted core sampler  such
as a split spoon sampler  should  be used.

     Surface  sampling  should be  augmented with a modest
number  (see footnote, page 53)  of sequential  samples taken  down
to 1.5 meters in order  to determine  if the pollutants have moved
downward.  These additional samples should be  located in the  area
of major  contamination.  Data from  these  samples will provide
information for deciding  if more than the surface soil needs to
be sampled in the final definitive study.

     With regard to  compositing  of  soil samples,  the major
concerns are that the samples be representative and that  high
concentrations not  be significantly reduced by being averaged
with lower level samples.  It   is recommended  that to improve
representativeness at  least  four different  samples taken  in the
vicinity of each selected sampling  site be  composited  into  a
single sample.  In addition a modest  number  (see footnote,  page
53) of single non-composited QA/QC samples  should be collected
from sampling  sites in  high concentration areas for comparison of
resulting data  with  that  from composited samples.

     The  exploratory  study is not  designed to  obtain information
on temporal patterns in  soil concentrations  since the study is
expected  to  be completed in a relatively short period  of  time.
                               54

-------
Thus,  temporal  trends  will normally be addressed in the final
study.   If  it  is possible to select a time for the  exploratory
study,  it  is  best  conducted at a  time  when the soil
concentrations would be expected  to be at a maximum.   It may be
necessary to use the  hypothesized dispersion model  in order  to
make this decision.  For example,  the sampling normally should
not  be done  immediately  following  a heavy rain, during winds
exceeding 20 knots, or during a time when the ground is frozen.

     Sample  preparation  for analyses  introduces some
possibilities  for  errors.  The sample  preparation  may  involve
drying,  grinding, mixing,  and sieving.   Prior to any sample
preparation procedures vegetation,  sod, or other non-soil
material must  be removed from the soil fraction.

     Grinding and mixing equipment as well as  any  sieves used
must be carefully  cleaned between each sample in order to avoid
cross-contamination.   The final rinse water used for cleaning
equipment should be sampled in order to  provide a sample blank
for  use  in evaluating the decontamination efficiency.  Collection
of one  sample  blank after processing a  group of 20  samples has
been used successfully in a number of EPA studies (USEPA 1982,
1984).   These  samples should be submitted to the laboratory along
with the  other  QA/QC  samples.  After evaluation of the sample
blank data  obtained from the exploratory study, a decision may be
made to increase or decrease the  frequency of blank collections.

     One of the most  serious possibilities for error during the
sampling process is discarding vegetation, sod or other non-soil
material  collected with the soil sample.  It is recommended that
all discarded  material be  retained,  including  any  materials
retained  on  the  sieve.   Ten percent of these samples should  be
sent to the analytical laboratory for analysis with the remainder
being  archived.   The  results  of  these analyses will give  a
quantitative estimate of  the possible  errors  introduced by
removal  of non-soil materials.   Care must be taken  in evaluating
and interpreting these data as data quality will be a function of
analytical  capability.


ANALYSIS  AND INTERPRETION OF DATA


     Analysis  and interpretation  of  the  total  integrated
information and data resulting from the  exploratory  study will
provide  the  basis for designing the  final definitive monitoring
study including all elements of the QA/QC plan.  For example,
decisions must  be made on whether  or not the selected control
area is adequate and appropriate;  whether the hypothesized model
is  valid;  whether the study area should be stratified and if so,
how; what number of  samples  should be collected at  what

                              55

-------
locations; whether or not the QA/QC plan for sampling is  adequate
and if not, how it should be changed; etc.

     If the  exploratory study is  conducted well, it will provide
some data  for achieving the  overall  objectives of the total
monitoring study? it will provide  data  concerning the feasibility
and efficacy  of most aspects of the monitoring  design including
the  QA/QC plan;  it will serve  as  a training vehicle for all
participants; it will pinpoint where additional measurements need
to  be made; and it will provide a body of information  and data
which may be  incorporated into the final report for the total
monitoring study.
                               56

-------
                          CHAPTER 5



      SELECTION OF NUMBERS OF  SAMPLES AND SAMPLING SITES
INTRODUCTION


     The QA/QC plan must be designed in such a way as to allow
for estimation  of errors in the determination of,  as  a  minimum,
mean  concentrations and standard deviations of the means.   In
some cases  the  primary interest may be in the determination of a
reasonable mean of extreme values (the stratum having the highest
mean concentration)  which must  be compared to an  acceptable
threshold level.   In  the  latter case protective actions will
generally be  required if the acceptable threshold level is deemed
to  be exceeded.   For this case the QA/QC plan must provide data
on the basis  of which one may state with  what confidence level
the  threshold  action concentration is, or is not, exceeded. Both
Type I and Type II  errors must  be taken into consideration.
These errors  can  only  be controlled by choosing an appropriate
number of samples.

     On  the  basis of  data from the  exploratory study the
following information will be available.

     o    Mean concentrations  and standard deviations of  the
          means for stratified regions  (assuming it  was deemed
          necessary to stratify the study region)
     o    Mean concentrations and standard deviation of the mean
          for the control region
     o    Results  of tests,  at specified confidence levels,  to
          determine whether or not the mean concentrations  in all
          strata are significantly different from the control
          region mean concentration
     o    Results  of tests,  at specified confidence levels,  to
          determine whether or  not peak or maximum measured
          concentrations exceed any established threshold action
          levels
     o    Some measure,  through analysis of variance  tests,  of
          the distribution of observed  variances among various
          elements of the  sampling plan, sample handling,  and
          sample  analysis
                              57

-------
     The problem now  is  to determine which elements of the
exploratory study provide sufficient  information  to meet  the
program objectives, and  where additional measurements will be
necessary.  Generally, since the exploratory study was designed
to provide only a limited  sample of the desired study population
(and perhaps a biased sample) it will be necessary to obtain  some
additional measurements  to improve  the levels of precision and
confidence, to  confirm the results,  and  to expand  the
measurements to cover regions not previously sampled.


NUMBER OF SAMPLING SITES REQUIRED


     The minimum number  of samples ,n,  required to achieve a
specified precision and confidence level may be estimated by  use
of  the tables given in Appendix A.  For values outside the range
of the tables the following  equations may be used (Mason, 1983):
              2         I \ 2                        t2        (CV) 2
        n _  t(a/2,n-l) *0)    or its  equivalent  n =  fo/2fn-l)	

                 P2                                   P2


where a is the  standard  deviation of the  mean  for the total
population,  p is  the precision and must be in the same units as
Qt  the confidence level  is 1-a, and ^01/2,n-1) is tne value of
Student's t distribution for the two-tailed test  at  n-1 degrees
of  freedom.   In the  second equation CV is the coefficient of
variation and if  CV is in percent, p must be in  percent.   These
equations must be solved by iteration  or graphically since t  is a
function of n-1.  Normally  the two-tailed t test  is  used unless
the  objective  of the  study is to  determine whether or not a
defined value is  exceeded.   In the  latter case  a  one-tailed t
test is used and  t    -.is  substituted  for
                 a, n-1

                    fca/2,n-1

The  tables in Appendix A are labeled  to make them applicable for
both one and two-tailed tests.

     If one approximates  the population mean y by a sample mean,
x, obtained from  the exploratory  study, it  is  possible to
calculate the  required number of  samples to achieve  a specified
precision at a specified confidence  level.  The  next problem is
to  estimate what sources  of error may be involved  in proceeding
in this fashion.

     First, this entire  approach assumes that the measurements
are independent of one another and are  distributed normally.   If
one  or the other,  or  both, of these assumptions  is not valid,

                              58

-------
undetermined  errors may be introduced.   Furthermore the standard
deviation,  s, of  the sample obtained in  the  exploratory study may
not  be a good approximation of the standard deviation, o,  of  the
entire population.  This problem can be  evaluated by  comparing
the  standard  deviation obtained in the final definitive study  to
the one obtained  from the exploratory study  and testing to see  if
they are different  at some prescribed level  of confidence.

     If the  normal distribution assumption is not valid, then  an
assumption  of a log normal  distribution may  suffice.   If the
measurements  are  dependent on one another, it may be possible  to
replace classic statistical techniques with  Kriging.  Examination
of  the data collected  from the exploratory study should enable
one to make decisions on these matters.

     If one  of the objectives of the final  definitive study is  to
compare measured  peak values to an acceptable  threshold level,
another possible source of error must be dealt with.  This source
of error is addressed by modifying the equation so  that the
minimum sample size, n, is given by the following formula (USEPA,
Dallas Lead,  1984):
       „ .   (a*Btn-l)°a   Qr its equivalent n .  VB,n-VCV'd
                                                    P
Where a =  probability of a  Type I error,  6 = probability of  a
Type II error,  03 = standard deviation of the  differences between
all  points  in the  strata having the highest mean concentration
and the acceptable threshold value,  pa  = minimum value of the
difference  considered to be above  the  threshold value, and
tfct+Brn-l)  is the value  of Student's t distribution  needed for
the  one-tailed test at n-1 degrees of freedom.  Note that if the
study region is stratified, this sample size would be  applicable
only to the stratum having the  highest mean  concentration.

     Table 5-1  gives some examples of determinations of required
minimum numbers of samples for  hypothetical  data of different
standard  deviations to achieve different levels of precision and
confidence  where the presented values  come  directly from the
appropriate tables in Appendix A.   Table 5-2  gives similar
examples for hypothetical data  to determine  the required minimum
numbers of  samples needed to decide if an  acceptable threshold
value has been exceeded  at  different levels of precision and
confidence.
                              59

-------
Table 5-1.  Number of Samples Required to Achieve an
Analytical Precision (p) at a Confidence Level of (1-a)
Standard Deviation
(%)
1


5


10


50


100


Confidence
Level
(%)
99
95
90
99
95
90
99
95
90
99
95
90
99
95
90
1
10
6
5
166
96
69
664
385
271
-
-
—
-
-
-
Number of Samples
Precision
(%)
5 10 50
322
221
211
10 5 2
631
521
31 10 3
18 6 2
13 5 2
664 166 10
385 99 6
271 70 5
664 31
385 18
271 3
100
1
1
1
2
1
1
2
2
1
5
3
2
10
6
3
                          60

-------
Table 5-2.
Precision
(1-6).*
  Number of Samples Required to Achieve an Analytical
(p)  at a Confidence Level of (1-a) and a Power of
Standard
Deviation
Confidence
Level
Power
1
Number of Samples
Precision (%)
5 10 50
100
5
10
50
99
95
90
99
95
90
99
95
90
99
95
90
99
95
90
99
95
90
99
95
90
99
95
90
99
95
90
90
80
70
90
80
70
90
80
70
326
215
165
251
155
114
203
120
83
1302
857
657
1004
619
452
813
619
452
-
-
-
16
10
8
13
8
6
11
6
5
55
36
28
43
27
20
36
21
15
1302
857
658
1004
619
452
813
471
327
16
4
3
5
3
3
5
3
2
16
10
8
13
8
6
11
6
5
326
215
165
251
155
114
203
120
83
1
1
1
1
1
1
1
1
1
2
1
1
2
1
1
2
1
1
16
10
8
13
8
6
11
6
5
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
6
4
3
b
3
3
5
3
2
* Since  6  is  the  probability of making a Type II error  (false
  negative),  the  power,  or  1-g, is  the probability  of not making
  a Type II error.
                               61

-------
     An example  of a sample calculation is  presented below:

Suppose °d = 10  percent, confidence level(1-a) = 95 percent,

Power (1-3) = 90 percent, and p2
Then n -     fto)2            = * o+B,n-l-

Solution must be by iteration.  Try ni~l =  °°.

Then ni = t2a+6/oo =  (t^+t^J2 = (1.645+1.282) 2 = 9.

Recalculate n2 with ni~l » 8.

Then n2 = (t^g+t^g)2 = (1.B60+1.397) 2 . n.

Recalculate n3 with n2-l = 10;  Then n3 =  = (1-833 + 1-383)2 = 10 = n3 = n.

Thus the desired n = 10 as shown in Table 5-2.

     It  is usually necessary to calculate  the probable total cost
of the program,  including all QA/QC samples.  By comparing  total
cost to the available resources one can determine the feasibility
of continuing with  the study.   If inadequate  resources are
available,  either more  resources must be obtained, the  study
objectives must  be changed, or the  study  should  be  abandoned.
Experience  indicates  that the  second item, changing the  study
objectives, is the most likely option.


LOCATION OF SAMPLING SITES


     The  location of  the  sampling sites will depend on whether
the  sampling is  random,  judgmental, systematic,  or  some
combination.   All information  and data resulting from  the
exploratory study should be used to  assist  in  the location of
sampling  sites.   Assuming that  a reasonable validation  of  the
model used for designing the exploratory study has been achieved,
the  study region  may  be stratified as  deemed appropriate.  An
approach using a combination of judgmental  and random sampling is
recommended  where the stratified areas to be sampled are located
judgmentally,  but the  specific  sampling sites within the
stratified areas are selected on a random basis.

     In the  event it  is impossible  to  obtain  a  sample at  a
randomly selected sampling location,  a  sample should be  obtained
                              62

-------
from  the closest available alternate site.  For example, the
selected site may  be beneath an  asphalt or concrete parking lot,
a road,  or  a building.   In these cases it is  not recommended  to
use drastic  measures such as jackhammers or picks to collect soil
samples.   Any errors introduced by moving to the closest
alternate site  are not apt  to influence unduly the overall
results.

     The movement  of pollutants  over  the surface of a site and
through the  soil mass may be strongly influenced by the shape  of
the  terrain,  by soil type, by geological formations,   by
vegetation and by land  use.    If any of these  factors  are
important,  the  sampling design should be stratified in order  to
include each important factor.  For example, a liquid pollutant
deposited  on  a  hill  top  will move  down  slope.   Maximum
concentrations  are likely to occur in low areas as opposed  to
ridge  tops.  Stratification of such an area into three strata -
ridge top, hill  side, valley floor - is recommended.  This design
allows  the  analysis  of  variance to remove the variation due  to
these three  strata  from the total  error term;  thus, the estimated
sample variance will be reduced.

     Within each  stratum  the location of  sample points should  be
randomly located.   The  number of sampling points assigned to a
stratum  should  be assigned on  the basis of the percentage of the
total area located  in that stratum.

     Two other techniques  can  help to identify  the  location  of
sample  sites.   The use  of Kriging to analyze the exploratory
study  data  can  generate an error map for  the site.  If the error
map is generated on the basis  of a desired confidence level, the
araas  that do  not meet this  level of reliability  can   be
identified and sampled.

     Geophysical  measurement techniques can  be used to locate
plumes  of pollutants  under some geological  conditions.   The
pollutant pattern identified under these conditions can  be used
to stratify  the  sampling by "plume" and "non-plume" areas.  There
may  be  the need to  identify  a third stratum covering   an
intergrade along the edges of  the  plume.


QUALITY ASSURANCE  ASPECTS


     The best QA/QC  plan that  can be designed on the basis  of
what is known about a  study area and a control area may not  be
adequate  to  achieve  the  desired  levels  of precision and
confidence.  Accordingly, it is recommended  that  additional
samples  be  included  in  the definitive study design that may help
to determine whether or not desired levels  for QA/QC have in fact
                             63

-------
been achieved.  Conceptually there are at least two different
approaches available  to  achieve this goal.  If an independent
(different) method  is  available for measuring  the parameters of
importance, that method can  be used and the results compared to
those achieved by the  first method.  Generally an independent
method is not available  for monitoring of contaminated soils.
Another  approach  is to deliberately overdesign the study by
taking and analyzing  many  more samples  than the recommended
numbers for a minimum adequate  design.  This is  usually deemed to
be undesirable because of  the additional cost which it entails.

     The recommended  approach  is a modification of the  over
design concept.   It is suggested  that  a modest number  (see
footnote, page  53)  of additional samples  be taken at randomly
located sites  in the stratum where  the highest concentrations are
expected.   These high concentration  areas are normally identified
on the basis of the exploratory study data.  The data should be
analyzed  for  that  stratum both with and without the additional
samples  and the  two results compared.  The comparisons should
give some  indication as  to whether or not the  desired levels  of
precision and confidence were actually achieved by the original
design.
                              64

-------
                          CHAPTER 6


                      SAMPLE COLLECTION
INTRODUCTION
     An important  segment of  the  definitive studies QA/QC plan
deals with sample collection.  Sample collection has  been covered
by Mason  (1983),  Cline  (1944)  and  Ford  et al.  (1983).  This
aspect of  the definitive studies must  be designed  to meet  the
specific objectives.  Improperly collected samples can void the
entire study.  The final protocol must provide  guidance on  such
matters as sampling methods, equipment, locations  for sampling,
and compositing; dealing with non-soil portions  of samples;
dealing with the  existence of animal burrows, root channels and
other such anomalies in the soil being sampled; and the selection
of the depth or depths that should be sampled.

     The  environmental scientist should be able to estimate the
components of the  variance,  or  error, associated with  each
element of the sample collection methods and procedures used from
the data generated by the study.   Evidence from  the  exploratory
study pertinent  to this estimation process should  be taken into
consideration.   It is recommended  that  a  minimum adequate
approach be sought consistent with the objectives  of the study,
the resources available and the designated  required levels  of
precision  and  confidence.  Also  an effort should be made to
establish some criteria or procedures for estimating, after  the
fact, whether or not the sample collection  elements of the QA/QC
plan and objectives were satisfactorily achieved.


SIZE OF SAMPLES AND METHODS OF COLLECTION


     Generally,  the  minimum sample volume is specified by the
analytical laboratory on the basis of  the selected method and
required sensitivity of analysis.  Of course, enough sample must
be collected so that the final soil sample remaining after any
non-soil materials are  discarded will be adequate to meet the
requirements of the  analysis. Sufficient  information should  be
available  from  the exploratory study to  provide  guidelines on

                             65

-------
approximately  what percentage  of the collected sample will be
discarded during the analytical sample preparation process.

     The  method  of  collecting  samples must  take into
consideration the required depth of sampling as well as  required
amounts.  The  data obtained from the exploratory study on soil
concentrations at various depths should be adequate  to allow  a
specification of required depths of sampling  for the final study.
The sample collection device must be adequate  to obtain samples
to  the required depth.   Mason  (1983)  and Ford et al. (1983)
provide information on available sampling devices and procedures.
The sampling device must be carefully cleaned between each use to
avoid cross contamination of samples.   Suggested cleaning methods
are given in USEPA (Love Canal 1982,  1984).  The type of device
selected should  be one that  provides the most cost  effective
sample.

     Frequently when  collecting soil samples anomalies such as
animal burrows,  root channels, sand lenses, desiccation cracks
and/or other factors that can alter the sample, or the collecting
procedure, and affect pollutant migration will be encountered.
These anomalies should be documented and noted on coring logs, in
sample log books, and on sampling site description forms.


BORING LOG


     When subsurface samples are being collected a boring or core
log should be prepared.  The log should indicate soil  structural
changes,  stratigraphy  changes,  the presence of rock, sand  and
gravel lenses, root channels, animal burrows, debris  and other
factors  that may be useful in interpreting the likely avenues of
contaminant migration.

     The  forms  for these logs can be designed by the investigator
or can be copied from most geological field sampling texts.   The
form should present  both a graphic and a verbal description of
the soil lying below the surface.

     Quality assurance can be maintained by periodic audit of  the
forms and by  proper training of the personnel  preparing the forms.
Properly  prepared boring logs  are  one  of the most valuable
interpretive  tools developed during an investigation.


FREQUENCY OF  SAMPLING


     The  required frequency of sampling depends on the objectives
of the  study,  the  sources of pollution,  the pollutants  of
                             66

-------
interest, transport rates,  and disappearance rates (physical,
chemical  or biological transformations as  well as dilution or
the  determination of dispersion).   Sampling frequency may be
related to changes over time, season, or precipitation.  Normally
little information will  be  obtained on sampling frequency  from
the exploratory study but  in those cases where temporal  changes
are  expected the  final  study should  incorporate  sampling
frequency into  the design.   It is  not uncommon  for many
definitive studies to be conducted over a period of one year or
more.

     The  changes expected in the concentration of pollutants in
soil are  normally associated with precipitation.  Precipitation
may  influence  the movement  of chemical pollutants downward and
aids in decomposition.  Sampling frequency associated with either
major  rainfall  events or with an accumulated amount of rainfall
can often provide  valuable information  on changes  that are
occurring.

     Monitoring studies are often designed to measure the effects
of some remedial measure on the site.   Trends are important in
these  cases.   The frequency of sampling should be designed to
measure the changes.  One  approach used successfully has  been to
provide   intensive initial sampling early in the study, then to
decrease  sampling frequency as  the levels  begin to  drop.  One
recommended  procedure  would be to sample monthly for the first
year, quarterly for the second  year, semi annually for  the next
two to three years then annually thereafter.

     Evaluation  of the  trend of the  data should  allow the
environmental  scientist to determine when the sampling  frequency
can  be reduced  or halted completely.  Monthly data may provide
the  needed data for performing statistical  tests and  for
determining the yearly variation within the data base.

     Samples  collected  for  evaluating  trends can usually be
obtained  on some subset of  the initial  year's sampling.  The
major  focus  is  mainly  on the highly contaminated and on the
immediately adjacent areas.   The environmental scientist is
primarily interested in detecting changes in these adjacent areas
in order  to provide early warning of a breakdown of the  remedial
measures.


QUALITY ASSURANCE  ASPECTS


     QA/QC procedures  of the sample collection effort must
identify and determine the magnitude of  errors associated with
characterizing  soil contamination introduced through the sample
collection effort.   Audits  (see Chapter 9) are perhaps  the most

                             67

-------
effective tool to  insure that  the sampling is done correctly.
Factors  likely to  influence  the magnitude  of  the sample
collection error are sample  size, collection methods,  and
frequency  of sampling.   The most  important of  these are  the
methods of collection and  the frequency of sampling.

     The   tools  used for collecting soil samples are limited and
are not likely to be  sources of error.  The errors  most likely
occur in  the use made of the tool.  Proper replication will
insure that the precision of  the procedure  meets  the QA/QC
objectives.

     Techniques  such as  trend line analysis or interdiction
analysis will provide a means of evaluating the effectiveness  of
the  data  obtained  from  sampling frequency studies.  The methods
outlined  in Chapter 4  for evaluating possible  errors  from
compositing can prove to be a valuable tool for evaluating the
use of  the composite  sample.

     A comparison  between  the first samples taken and the most
recent  should show a  decrease in pollutant concentrations unless
there  is  a new source of pollutants, there is migration into the
sampled soil or there is an error in the  data.  This  test becomes
a better  indicator of errors the longer the study runs.
                             68

-------
                          CHAPTER 7
               SAMPLE HANDLING AND  DOCUMENTATION
INTRODUCTION
     The  goal is  to define  the segment of the QA/QC plan dealing
with all aspects  of sample handling including the  transfer of  the
sample from  the  collecting  device  to a suitable container,
transportation of the sample, and the preparation  of  the sample
for  analysis.   The importance of all these aspects of sample
handling and possible errors  introduced  thereby will  naturally
vary  with  the  sampling methods,  monitoring  objectives,
characteristics of the soil being sampled and the physical  and
chemical properties of the pollutants of concern.


CONTAINER  PREPARATION, LABELING, PRESERVATION, AND  SAMPLE
PREPARATION


     The  sampling protocol and the QA/QC plan  must address the
following factors.

     o   Type  of container material, its size, shape and the type
         of lid.

     o   Cleaning procedures  for the containers

     o   Decontamination procedures for  sampling instruments.

     o   Decontamination procedures for  sample bank equipment.

     o   Labeling scheme and  log book entries

     o   Chain of custody procedures

     o   Sample preparation procedures in the field

     o   Sample preparation procedures at the sample  bank
                               69

-------
     Due to a lack of  specifically tested and recommended methods
dealing with the storage, handling,  construction  and types  of
containers, cleaning and decontamination  of containers, and
suggested materials for container  lids for soil  samples it  is
suggested that the  specifications and methods identified in
OSEPA,  Federal Register Vol. 44 No. 233 (1979) be utilized.

     Table 7-1  provides  general information on  recommended
containers, preservation requirements, and holding times  for
measuring selected  contaminants.   Even though these procedures
and methods were specifically designed and  tested for water
samples, they are applicable for soil sampling studies.

     For sampling studies that require a large number of samples
and/or  extensive preanalytical sample preparation  a sample  bank
may  be established.  The sample bank is the element that operates
between the field sampling effort and the analytical laboratory.
However,  for smaller studies the sample banks responsibilities
are often  incorporated  into the responsibilities  of the field
sampling team or the analytical laboratory.

     If a sample  bank is established,  sample bank personnel can
assume  responsibility for the following procedures:

     o     Custodian for  all records pertaining to the sampling,
          sample preparation as required, and  shipment of  soil
          samples to analytical laboratories.

     o     Responsibility  for record filing  and storing, for
          storing  and preparation  of soil  samples,  and  for
          dispensing  containers,  sampling  equipment and all
          custody documents such as chain-of-custody forms  and
          sample collection and analytical tags, as required.

     o     Responsibility for  updating and maintaining  the
          projects' master  log book, auditing the records  as
          required, generating  sample bank  QC  sample blanks,
          accepting QA/QC samples  for inclusion into  the
          analytical scheme, and for scheduling the collection of
          field sample blanks.

     o     Responsibility  for completing, as required, analysis
          data reporting  forms  and for assuring that  all
          chai n-of-custody requirements pertaining to  all
          field sampling, shipping and sample  bank operations,
          are adhered to.
                             70

-------
Table 7-1 Sampling Containers, Preservation Requirements, and Holding  Tines  for Soil Samples

CONTAMINANT                   CONTA1N£R         PRESERVATION           HOLDING  TIME
Acidity
Alkalinity
Aaaonia
Sulfate
Sulfide
Sulfite
Nitrate
Nitrate-Mitrite
Nitrite
Oil and Grease
Organic Carbon

  Metals
Chromium VI
Mercury
Metals except above

  Organic Compounds
Extractables (including
phthalates, nittosamines
org%nochlorine pesticides,
PCB'e nitroaromaticn,
isophorone, Polynuclear
aromatic hydrocarbons,
haloethers, chlorinated
hydrocarbons and TCOO)
Extractables (phenols)

Purgables (halocarbona
  and aroma tics)
Purgables (acrolein and
  acrylooitrate)
Ortbophosphate
Pesticides

Phenols
Phosphorus (elemental)
Phosphorus, total
Chlorinated organic
  coapounds
P.G
P.G
P,G
P.G
P.G
P.G
P.G
P.G
P.G
G
P.G
P.G
P.G
P.G
Cool, 4°C
Cool, 4°C
Cool, 4°C
Cool, 4°C
Cool, 4°C
Cool, 4°C
Cool, 4°C
Cool, 4°C
Cool, 4°
Cool, 4°C
Cool, 4°C
Cool, 4°C
G, teflon-lined   Cool, 4°C
  cap
G, teflon-lined
  cap
G, teflon-lined
  aeptum
G, teflon-lined
  septua
P,G
G, teflon-lined
  cap
P,G
G
P.G
G, teflon-lined
  cap
Cool, 4°C

Cool, 4°C

Cool, 4°C

Cool, 4°C
Cool, 4°C

Cool, 4°C
Cool, 4°C
Cool, 4°C
Cool, 4°C
14 days
14 days
28 days
28 days
28 days
48 hours
48 hours
28 days
48 hours
28 days
28 days
48 hours
28 days
6 Months
                      7 days (until extraction)
                      30 days (after extraction)
7 days (until extraction)
30 days (after extraction)
14 days

3 days

48 hours
7 days (until extraction)
30 dsys (after extraction)
28 days
48 hours
28 days
7 days (until extraction)
30 days (after extraction)
Polyethylene(P) or  Glass(G)
Sample preservation should be performed immediately upon sample collection.  For composite samples
each  aliquot  £Jiould be  preserved at the time of collection.  When impossible to preserve each
aliquot,  then samples may be preserved by maintaining at 4°C until compositing ond sample splitting
is completed.
Samplea  should be analyzed as soon as possible after collection.  The times listed are the maximum
times that  samples  may  be held before analysis and still considered valid.  Samples may be held for
longer periods only if  the aialytical laboratory has data on file to show that the specific types of
samples  under study are stable for the longer time.
For additional information see Ford et al (1983).
                                                   71

-------
     The  following  sample  bank procedures have  been used
successfully on a number of soil monitoring studies.

     A.   Issuing Supplies:

         (1) The  sample  bank  issues  as  required sample
            containers, sample collection tags,  chain-of-custody
            forms and  site description forms  to the sampling
            teams.  Sample collection tags and chain-of-custody
            forms are normally accountable documents; the sample
            bank will  log  the  forms by numerical lot identifying
            the  team  and/or the individual responsible for the
            temporary  custody  of these documents.

         (2) The  sample bank may  be required to store sampling
            equipment  in a suitable environment.   If sampling
            equipment is  stored at the sample bank, issuing this
            equipment  to the sampling teams as required will  be
            necessary.

     B.   Accepting and  Logging  Samples:

         (1)  Transfer of  sample custody from  the  sampler to
            sample bank personnel will  normally occur at  the
            sample bank.

         (2)  Before accepting  custody of any samples,  sample bank
            personnel  must check  all tags  and  forms for
            legibility and completeness.

            (a)   All  individual samples must have a completely
                 filled out sample collection tag  attached.

            (b)  Every  sample must  be  identified on  the
                 chain-of-custody form.

            (c)  Each site sampled must have a completely filled
                 out site description form.

            (d)  Any  discrepancy will be corrected before sample
                 bank personnel  will  assume custody.   If  a
                 discrepancy exists that cannot be  resolved to
                 the satisfaction of the sample bank  personnel,
                 resampling, filling  out  additional tags and
                 forms, and/or revisiting the site  to obtain
                 necessary documentation may be required.
                              72

-------
            (e)  All unused accountable documents as shown in
                 Table  7-2 must be returned  to the sample  bank
                 on a  daily basis.   However,  depending upon
                 circumstances  such as a  sampling team's
                 schedule  and route, accountable documents may
                 be retained by the sampling  team leader.   The
                 sample  bank supervisor, however, must be aware
                 of the situation.

         (3)  After  the sampler  relinquishes custody and the
            sample bank personnel  assumes  custody of  the
            samples,  each  sample must be  logged into the master
            log book.

     Preparation of soil samples for analysis may require sample
bank personnel  to dry, sieve,  mix  and  aliquot  samples
appropriately.   The preparation procedures selected are
determined by the contaminant to be measured  and the  analytical
requirements.   Various techniques  and methods for mixing and
compositing soils have been described by Oregon State  University
(1971),  USEPA (1984), and Peterson and Calvin (1965).

     It is inappropriate  to initiate a  sampling study without
first consulting with analytical personnel.   Collecting samples
that cannot be suitably analyzed will not  provide data necessary
for satisfying  the sampling objectives.

     The  possibility  of errors being  introduced in sample
preparation procedures  involving the  discarding  of  non-soil
material or of  non-sieved material as well as possible losses
during  any grinding  or drying operation  has  been briefly
discussed  in  Chapter 4.   The definitive study decisions
concerning the  non-soil fraction must be made on the basis of the
data obtained  from the exploratory study.  For example, available
data may indicate  that significant  contamination is in the
discarded portion.   If so,  it  is recommended that  the discarded
portion from ten percent of the samples collected from the area
having  the highest concentrations be analyzed.  An  estimate can
then be made of the total amount  of contamination  being  discarded
by  multiplying  the measured concentration in  the discarded
material by the total amount  of the discarded material.  Assuming
that this amount is uniformly distributed through the soil sample
remaining after non-soil materials and non-sieved  materials have
been discarded,  one can then calculate an estimated  value for the
potential soil  sample  total concentration if  none  of  the
contamination had been discarded.   Comparison of  this potential
concentration to  the actual measured concentration will  enable an
estimate of the possible error to be made.
                              73

-------
                           TABLE 7-2.  ACCOUNTABLE DOCUMENT CONTROL REQUIREMENTS
    Documentation
 Iifued by
  Numbering
       Interim
    Responsibility
        Fin* I
   Responsibility
Sanple Collection Teg*

Custody Records

Field Logbook*

Site Description fora*

Analytical Sample Tags

Laboratory Notebook*

Analytical Data Sheet*
Sample Bank

Sample Bank

Sanple Bank

Sample Bank

Sanple Bank

Laboratory

Staple Bank
Praaerialiied

Preaerialiied
Preaerialiced
Sampling Teaaj

Sanpling Tea*

Sampling Tea*

Sanpling Te«»

Sanple Bank

Analytical Laboratory

Analytical Laboratory
Sanple Bank

Sample Bank

Sample Bank

Sanple Bank

Analytical Laboratory

Sanple Bank

Sanple Bank

-------
     If the  error estimated by this process exceeds acceptable
limits specified in the QA/QC  plan, it might  be necessary to
modify sample  preparation  procedures for the definitive  study.
One might  consider a sample preparation procedure in which the
entire collected  sample  (soil and non-soil materials) is
extracted  in  the analytical laboratory.  The analytical  results
could then  be  reported  as  amounts  of  contaminant per gram of
mixed material.  At present there is no acceptable method for
proceeding  in  cases such as these.  One problem is the lack of
standard reference materials for  determining and measuring errors
in  extraction  efficiency.  One solution may be  to try different
methods of extraction and  compare  the results.   The  final
interpretation of the data must  then take  into consideration
these estimated  errors.
QUALITY ASSURANCE  ASPECTS


     The  problem is  to  quantitate overall  errors.  The
recommended procedure for verifying  that the QA/QC plan is being
carried out  properly  for this chapter's factors is a periodic
audit,  combined  with  a modest amount of extra samples and
analyses related to factors discussed above.
                              75

-------
                         CHAPTER 8


          ANALYSIS AND  INTERPRETATION OF QA/QC DATA
INTRODUCTION


     One goal  in the analysis and interpretation of data is  to
show how all aspects of QA/QC for a soil monitoring study combine
to give an overall level  of precision and confidence for the data
resulting from the  study.  Another goal  may be to  determine
whether all QA/QC procedures which were used were necessary and
adequate and  should  definitely be incorporated  into future
studies of the same type. This  entire evaluation must be closely
linked to the objectives of the study.  In summary the  important
questions  to  be answered are, "What is the  quality of the data
(maximum accuracy attainable)?" and  also," Could the  same
objective  have  been achieved through an improved QA/QC design
which may have  required fewer resources?"


PRESENTATION OF DATA SUMMARIES


     It  is desirable to provide summarized  tables of validated
QA/QC data in  the final report.   For example,  QA/QC  data
validation procedures  used in  a number of soil  sampling studies
reported by Brown and Black (1983) included validation  of sample
data sets by  checking and assessing the accompanying QA/QC data.
The criteria for  QA/QC samples  and procedures  used  to validate
all data included:
                              76

-------
Samples  and Procedures
                Example  Criteria
1.  Reagent Blanks

2.  Calibration Check
      Standards
3.  Laboratory  Control
      Standards
        Concentrations  had  to be less
        than 0.25  g/ml"1.
        Recovery must be  between 95% and
        105% of the known value for
        either the first  analysis or the
        first re-check  analysis.

        Recovery must be  between 90% and
        110% of the known value for
        either the first  analysis or the
        first re-check  analysis.
     Data  produced  by any  sampling  and analyzing system are
affected by two  types of  errors;  random  and  systematic.   The
accuracy of  any one  result then, is a function of the bias (due
to systematic error) and precision (due  to
collection  and analysis methodology.
components,  associated with extraction and
and  is assessed by the  mean  recovery
Standards and Laboratory Control Standards
                           random error)  of the
                            Bias has at least two
                           instrument efficiency,
                           of Calibration Check
                           (LCS).  The  LCS check
overall bias
determines the
for  the system;
instrumental bias.
the Calibration Check Standard
     Total  random  error  can be assessed  by analyzing duplicate
samples, but it includes errors due to sample collection,  sample
homogeneity,  sample  extraction, sample composition  (matrix
effects) and instrumental reproducibility.   These errors  can be
evaluated  by  the use  of the  other QC procedures stated above  and
are assessed by calculating the  standard  deviations  of the
various analyses.
     The  accuracy  of analysis, i.e
evaluated separately below  for the two
the following equations:
                       . ,  bias  and precision,  are
                       types of samples, using
     Recovery = Amount Found/Known Amount
                                            (1)
     Bias (B)
 Recovery - 1
                                                           (2)
                               77

-------
Difference  (D)  = | x^ -  x2|  where x^ and X2 are the analytical
results  of  paired analyses and the  average is:
                        n
                                                           (3)
and the precision is:
                s_=Precision = 0.8862  D
                                                           (4)
where 0.8862  converts  the range of  two  results to the standard
deviation (Natrella, 1963).
 then
     If  component  errors are used to assess total random error,
+ D2 +
                       )/n and
                                           (5)
precision -  [0.8862 (D].2 + D22
                        )  +
                                    s 2
                                                          I1/2-
     Equation  (3)  is suitable for  use on results where the
 concentration varies  over a  very  narrow  range.   If  the
 concentrations found vary by an order of magnitude or more,  then
 the difference should  be normalized by dividing by the average of
 the  two  values and the precision is expressed as the coefficient
 of variation (CV) which  is s/x
       Z
   n   n
  D_ L, I I X  ~ X I  »
_ - _ ( |  1    2h )
                                                           (6)
                   r   ix  - x
                  2 z   I i   2
                   n   .
              CV  = 0.8862  D
                                                           (7)
                          n
     One  of the  studies  discussed  by Brown  and Black (1983)
 involved lead contaminated  soils.  The use and  evaluation of  the
 QC analyses for this soil  monitoring study was presented as
 follows:
                              78

-------
The  limit of detection, approximately 0.25 yg ml~l, was tested on
about  10  blank  analyses  using a more sensitive  absorbance
wavelength for  lead on an AAS.  The result was less than 0.1 yg
ml"1, or 2 yg g~l for sample analysis.  This suggests  that most
of the blank analyses were less  than 2 yg g~l, but  this cannot be
stated with any confidence.  The results of the QC analyses were
as follows:
QC Sample
Calibration Check Standard
Laboratory Control Standard
Field Blank (yg ml"1)
Sample Bank Blank (yg ml~l)
Reagent Blank (yg ml~l)
Re-extraction Analysis
Total Recoverable
Split Extract (CV)
Spiked Extract
Spiked Sample
Duplicate Aliquot (CV)
Duplicate Sample (CV)
Triplicate Analysis (CV)
No.
150
147
76
77
148
17
144
147
147
147
134
129
220
Mean
101.5%
101.2%
<0.25
<0.25
<0.25
1.7%
99.8%
0.0089
99.4%
100.4%
0.053
0.189
0.144
S
2.6%
4.1%



1.4%
8.0%
0.0079
5.0%
5.1%
0.047
0.168
0.128
 (1)   Bias:   The percent recoveries indicated above for the
 Calibration Check Standards and LCS's  suggest  a  small positive
 bias  for the  method  of soil analysis, due  principally  to
 instrument reproducibility.  The result, using Equation (2), is:


           Bias  = Recovery - 1 = 1.012 - 1  = 0.012.


 (2)   Precision:  The  recovery of the analyte by the analytical
 method compared  to the "total" recoverable  method was  essentially
 equal and  re-extraction of  the residue left from the initial
 extraction indicated  an  additional 1.7 ±1.4 percent  recovery,
 also essentially equivalent.   Furthermore, the results of the
 three types  of  blank  analyses  indicate no measurable
 contamination  from reagents,  sample collection,  or  sample
 preparation.  The remaining random errors  are  evaluated below.
 Because  of  the  wide  range  of  concentration of  lead  in the
 samples, the coefficient of variation is used,  Equation (7).
                              79

-------
     Precision  (total random error)  from Duplicate  Sample
Analysis:


          CV - 0.168 or 16.8% of sample concentration.


The component random errors, summed as per Equation (5),  are:


      sx m (0.00792 + 0.052 + 0.0512 + 0.0472)1/2 = 0.085.


     These  random errors  suggest  that  reproducibility
errors(0.0079) are small and that extract matrix, sample matrix,
and sample  homogeneity errors are equivalent.  The sum of these
errors is about  half the total random error  so the sampling error
is  essentially equal to all other errors combined.

     Inter laboratory precision as calculated from the results  of
triplicate analyses, using Equation (7)  is:


    Precision * CV « 0.128 or 12.8% of sample concentration,


 (3)  Uncertainty:  The data for bias  and precision  can  be
combined to yield the uncertainty for any  reported concentration
by  use of  the following equation:


                 U - (1 + B + 2 C)                           (8)


where B  is the bias, C  is  the standard  deviation or coefficient
of  variation as  appropriate,  and 2 converts these  to  the  95
percent confidence limits.   For soil analyses, using Equation (8)
and the bias and CV derived above,  the 95% confidence bounds on  a
reported value,  x,  are:


     Soil result will  lie  between 0.676x and 1.348x yg g"1.


      It  is required  that the  QA/QC plan ensure  and document that
all data collected, whether  used for research or for  monitoring
purposes, is  scientifically valid,  defensible and  of known
precision  and accuracy.   The  described presentation  of QC data,
though  designed  for  analysis of lead in soil, can be used as a


                               80

-------
guide  for  other  sampling and data analysis protocols and/or QA/QC
plans.

     Presentation of QA/QC data  allows  readers  to  verify
conclusions drawn as to the reliability of the  data.   Such an
approach also contributes to the building of a body of QA/QC and
monitoring  experimental  data in  the literature which allow
comparisons  to  be made  between and among  studies.   Procedures
used to validate  the individual data points should be  presented
and  where  some points are discarded arguments should be presented
to support  these  decisions.


PRESENTATION OF RESULTS AND CONCLUSIONS


     Special emphasis should be placed on how overall levels of
precision and confidence were derived from the data.   Great care
must be exercised to  insure  that,  in determining results and
conclusions, assumptions are not made which were not part  of the
study  design and  which cannot be tested by data derived from the
study.   If  portions of the study results  are ambiguous and
supportable conclusions  cannot be drawn with regard to the total
reliability of the  data, that situation must be  clearly stated.
In  that event   it  is  desirable to  include recommendations for
conducting  an improved study in  such a  way  as  to clarify the
observed ambiguities.


QUALITY ASSURANCE ASPECTS


     The adequacy of all  aspects  of the  QA/QC plan should be
examined in detail  with emphasis on defining  for future studies
an  appropriate minimum  adequate plan.  Some aspects of the plan
actually used may have  been too restrictive,  some may not have
been restrictive enough.  Appropriate analyses and interpretation
of  the data should identify the actual situation.

     Future  soil  monitoring studies should  have  checks and
balances built into  the QA/QC plan which will identify early in
the study whether  the  plan is  adequate and if  necessary, allow
for  corrective action  to be taken before the study  continues.
This is one of the major advantages  of conducting an  exploratory
study along the lines  outlined  in this  report.   If there are
problems  with  the  QA/QC plan,  they will often  be identified in
the  exploratory study and  be  corrected before  major  resources are
expended.
                               81

-------
     There  is insufficient knowledge dealing with soil monitoring
studies to  state with confidence which portions of the QA/QC  plan
will  be generally  applicable to all soil monitoring studies and
which portions must be varied depending on site-specific factors.
As experience  is  gained, it  may be possible to provide more
adequate  guidance  on this subject.  In the  meantime  it  is
recommended that  the best approach is to assume that important
factors of  QA/QC plans are site-specific  and to conduct  an
appropriate exploratory study at each new  study site to verify
that various aspects  of the QA/QC plan are adequate to  meet
program objectives prior  to proceeding with  the final definitive
study.
                              82

-------
                           CHAPTER  9


                  SYSTEM  AUDITS  AND TRAINING
INTRODUCTION
     The material  for  this chapter has been obtained primarily
from USEPA  Kellogg Idaho Study  (1984).   The  first phase  of an
auditing program  for soil monitoring projects should  be the
preparation of  standard operating procedures (SOP) that  identify
the methods and techniques necessary to perform all aspects of
the required audit.   The SOP must be adequate  to  perform  onsite
sampling and  sample bank  (where applicable) audits.   The  second
phase should then be the actual  conduct  of the  required  field
audit.   Audits  are conducted by appropriate elements  of agencies
or organizations  having cognizance over the monitoring  project.
The  frequency of  auditing should be determined by  the  project
officer.  Juran et al. (1979) state that, "the activities subject
to  audit should  include any that affect quality regardless  of the
internal  organizational location."

     A system audit  is an overall evaluation of a project to:

     o   Verify  that sampling methodology is being performed in
         accordance  with program requirements

     o   Check  on the use of appropriate QA/QC measures

     o   Check methods of .sample handling,  i.e.,  packaging,
         labeling, preserving,  transporting,  and archiving in
         accordance  with progam  requirements

     o   Identify any existing quality problems

     o   Check  program documentation,  i.e., records (site
         description, chain-of-custody collection and  analytical
         tags,  field and  sample bank  log  books and field  work
         sheets)

     o   Initiate corrective action if a problem  is identified
                               83

-------
    o   Asssss  personnel experience  and qualifications  if
        required

    o   Follow-up  on  any  corrective  action  previously
        implemented

    o   Provide onsite  debrief ings for sampling team and sample
        bank personnel.

    o   Provide a written evaluation of the sampling and sample
        bank program

The purpose of  the  system  audit  is  to ensure  that the QA/QC
system  planned for the project  is in place and functioning well.

    The auditor first  must review Work Plans, Protocols, Test
Plans,  QA/QC Project Plan, and  all Program Reports.  A discussion
of the current  status  of  the project, and the identity of any
problems encountered,  with the  project officer  is  suggested
before  conducting  the  onsite  sampling audit.   Sample
chain-of-custody  procedures and  raw data  are  checked  as
appropriate and results of blind QC samples routinely inserted in
the sample load  by  sample  bank personnel  are  reviewed.
Spot-checks of  sampling methods  and techniques, sampling and
analysis calculations, and data transcription are performed.


SAMPLE BANK AUDIT


     The primary objective  is to  determine the status of all
Sample Bank documentation and  archived  samples.   Emphasis  is
placed on:

     o  Verifying  that the documentation  is  in order  and
        sufficient to establish the disposition of  any sample
        collected
                          *
     o  Determining any discrepancies that currently exist and
        initiating corrective  action as  appropriate

     o  Verifying  that the  recording of QA/QC measures  (blanks,
        duplicate spikes, blinds)  is  in accordance with  the
        QA/QC Plan

    o  Establishing procedures  for  final  disposition  and
        mechanics of transfer  of all Sample Bank holdings  upon
        termination of the operation.
                              84

-------
The  first  step  of  the  audit  is  to inventory the Sample Bank
records and archived samples.   The records  that must be inspected
are:

     o   Chain-of-custody forms
         -    Field forms
         -    Analysis forms

     o   Sample tags
         -    Field tags
         -    Analysis tags

     o   Analysis forms
         -    Individual samples
         -    Batch sheets

     o   Shipment forms

     o   Logbooks
         -    Soils
         -    Daily log

The operational procedures inspected should include:

     o   Preparation Procedures  (sample bank or  analytical
         laboratory)
         -    Drying  (if used)
         -    Sieving
              Mixing
         -    Packaging
         -    Shipping

     o   Housekeeping
              Safety
         -    Decontamination
         -    Evaluation of Swipe Samples

     o   Security
         -    Forms (documents)
         -    Samples

     o   Storage
         -    Sampling equipment
         -    Archived samples

          Check  that required  documentation has been maintained
in an  orderly fashion,  that  each of  the recorded items  is
properly categorized, and cross-checking can be easily performed.
In addition,  ensure  that data  recording conforms to  strict
document control  protocols and the program's QA/QC Plan.

                               85

-------
     The  archived  samples  inspected  can be  categorized as
follows:

     o   Soil
     o   Blanks
     o   Splits
     o   Standard Reference Materials  (SRM )
     o   Non-Soil Materials Collected with the Soil Sample

     Conduct  an audit  of the  archived  samples.   Verify  that
appropriate  samples exist for  each  entry in the  logbook.   Field
sample tags  should be  replaced by the appropriate analytical
tags,  and chain-of-custody forms  are  prepared in order to
transfer the  samples.   Detailed sample  bank procedures are
presented by USEPA Dallas Lead Study (1984).


DAILY LOG


     Check  for clear, concise entries detailing events of  the day
 (such as numbers of  samples processed), problems encountered,  and
actions  taken to solve them.   This log can provide excellent
documentation  of the operation of the  Sample Bank.


SAMPLE BANK LOGS


     Review these  logs  for complete sample information  entered.
Changes made should  be by  crossing  out so  the original entry is
still visible,  and initialing.  In addition checks  for the
 identification and documentation of split  and duplicate samples,
and  field and  Sample Bank  blanks must  be performed.


SAMPLE COLLECTION  AUDITS


      It  is  recommended  that  an  audit of  the overall QA/QC  plan
for  sample documentation,  collection,  preparation, storage,  and
transfer procedures  be performed  just before sampling starts.
The  intent of  this audit  is  to critically review  the  entire
sampling operation to determine the  need  for any corrective
action early in the  program.  Additional total program or  partial
audits can be conducted  at  various times throughout the sampling
program.

     It  is  recommended -that the Project  Officer maintain a QA/QC
Coordinator onsite  during sample collection to monitor the
sampling team's  activities,  provide  technical and corrective

                              86

-------
action  suggestions  to  the sampling teams, and  supplement
performance audits on sampling as needed.


FIELD AUDITS


     The primary objective is to determine the  status of sampling
operations.  Emphasis is placed on:

     o   Verifying that operational aspects and procedures are in
         accordance with the protocols and  QA/QC plan.

     o   Verifying  the collection of all samples  including
         duplicates and field blanks.

     o   Verifying  that documentation is  in order  and sufficient
         to establish  the  collection location of  any sample
         collected.

     o   Determining  discrepancies that exist  and initiating
         corrective action as appropriate.

     o   Collecting independent samples.

     The on-site field audit  is  to inspect  sample records and
equipment.   Records inspected include:

     a.  Chain-of-Custody Forms

     b.  Sample Tags

     c.  Site Description Forms

     d.  Log Books

     The operational procedures inspected should include:

     o   Sampling Procedures
              Equipment
         -    Techniques
              Decontamination
              Collection of duplicate and field  blank samples
         -    Security
         -    Sample storage and transportation
         -    Containers
         -    Contaminated waste storage and disposal
              Site Description Form entries
                               87

-------
DATA MANAGEMENT AUDITS


     An audit of  the  data management system by tracing the  flow
of specific samples through the system should be  performed.   In
particular,  the  ability of  the system to correctly identify  a
sample from any one of  its identification numbers should be
checked.

     Entries in the  sample bank's logbook will be the  basis  for
these performance checks.  From time to  time,  erroneous  input
information may be  used  to audit the system.


TRAINING


     The  project leader of  a soil  monitoring project is
responsible for ascertaining that all members of his project team
have adequate training  and experience to carry out satisfactorily
their assigned missions  and functions.   Until  a  field  sampling
team has worked  together long enough for the project  leader to
have verified this  from  first hand knowledge it is good practice,
in  addition to any classroom training or experience, to conduct
comprehensive briefing sessions for all involved  parties during
which all  aspects of  the sampling protocol, including  the QA/QC
plan, are presented and  discussed in some detail.  This  approach
will help the project personnel to develop into a team where  each
team member knows his own job well and knows how it fits  into   the
overall team effort.   Sufficient field training exercises  should
follow the  briefing  sessions  until  each  team  member  can
demonstrate successfully that he can perform his job  routinely
well and without delay.  Of course, on subsequent projects of   the
same general type with the same team, the training exercises  may
be reduced in' number or  dispensed with as deemed  appropriate by
the project leader.

     In summary,  the  sampling effort must include classroom  and
field training programs  that'have provided detailed instruction
and  practical experience to  personnel in  sample collection
techniques and procedures, labeling, preservation, documentation,
transport, and sample bank operational procedures.  Also, special
training programs concerning 'procedures and program documentation
should be  completed by all personnel prior to their involvement
in the conduction of any audits.
                              88

-------
                          REFERENCES


1.    Allmaras,  R.  R.   Bias.   In:  Methods of Soil Analysis,  Part
     1, Physical  and Mineralogica1 Properties,  including
     Statistics of Measurement  and Sampling.  C.  A.  Black, et
     al,  ed.   American  Society  of Agronomy, Madison,  Wisconsin,
     1965.   pp.  24-42.

2.    Bauer,  Edward L.   A Statistical  Manual  for Chemists.
     Academic  Press, New York,  New York, 1971.  193 pp.

3.    Beckett,  P.  H. T.,  and R.  Webster.  Soil Variability:   A
     Review.   Soils and Fertilizers,  34:1-15.  1971.

4.    Box, George,  E. P.   Statistics  and the Environment.  Journal
     of the Washington  Academy  of Sciences, 64(2):52-59.   1974.

5.    Broms, Bengt  B.  Soil  Sampling  in Europe:  State-of-the-art.
     Journal of  the Geotechnical  Engineering  Div.,  106:65-98.
     1980.

6.    Brown, K.  W.  and  S.  C.  Black.   Quality Assurance and Quality
     Control Data Validation Procedures Used for  the Love Canal
     and Dallas Lead Soil Monitoring Programs.  Environmental
     Monitoring  and Assessment, 3:113-112, 1983.

7.    Buffington,  J. P.   Developing Recommendations  to Improve
     Quality Assurance  for  Federal  Monitoring  Programs.   In:
     Proceedings  of the National Conference on Quality Assurance
     of Environmental Measurements, Denver, Colorado,  1978.

8.    Burgess,  T.  M. and  R. Webster.  Optimal Interpolation and
     Isarithmic  Mapping .of  Soil  Properties.   I.   The
     Semi-variogram and  Punctual  Kriging.  Journal  of  Soil
     Science,  31:315-331.   1980a.

9.    Burgess,  T.  M. and  R. Webster.  Optimal Interpolation and
     Isarithmic  Mapping of  Soil Properties.  II.  Block  Kriging.
     Journal of  Soil Science,  31:333-341.  1980b.

10.  Campbell,  James  B.   Spatial Variability of Soil.   Annals  of
     the Association of American Geographers,  69(4 ) :544-556 .
     1978.

11.  Campbell,  James B.   Locating Boundaries Between Mapping
     Units.  Math.  Geology,  10 (3 ):289-299.  1979.


                               89

-------
12.   Cline,  Marlin.  Principles  of  Soil Sampling.  Soil  Science.
     Vol  58:275-288.  1944.

13.   Davis, John  C.   Statistics and  Data  Analysis in Geology.
     John C. Wiley and Sons,  Inc., New  York, New York, 1973.

14.   Dixon, W.  J. Processing Data for  Outliers.  Biometrics,
     9:74-89.  1953.

15.   Dixon, W.  J.   Extraneous  Values.   In: Methods of Soil
     Analysis.   Part  1,  Physical and  Mineralogical Properties,
     including  Statistics of  Measurement and Sampling.   C.
     A. Black, et  al, ed.  American  Society  of Agronomy, Madison,
     Wisconsin,  1965.  pp. 43-49.

16.  Eynon, Barry and Paul  Switzer.   A Statistical Comparison  of
     Two Studies  on Trace Element  Composition of Coal Ash
     Leachates.   Final Report.    Electric Power  Research
     Institute,  Palo  Alto, California.   July 1983.  EA-3181.

17.  Flatman, G.  T.   Assessing Lead Contamination near Smelters:
     A Case Study. Paper presented at  Workshop on Environmental
     Sampling.   February 1-3, 1984.   Las Vegas, NV.

18.  Ford,  Patrick  J., Paul J. Turina,  and Douglas E. Seely.
     Characterization of Hazardous Waste Sites—A  Methods Manual.
     Vol.  II.   Available Sampling  Methods.   EPA 600/4-83-040.
     U.S.  Environmental Protection Agency,  Environmental
     Monitoring  Systems  Laboratory,  Las Vegas, Nv.  1983.

19.  Hammond, Luther C. , William L.  Pritchett,  and Victor Chew.
     Soil Sampling in Relation  to  Soil Heterogeneity.   Soil
     Science Society  of  America Proceedings, 22:548-552, 1958.

20.  Heimbuch, Douglas G.   Computer Software  Package for  the
     Determination  of Optimal Preliminary Sample  Size  in
     Two-Stage  Sampling  for  Mean Concentration.   Cornell
     University.  Utica, NY.. 1982.

21.  Hipel, Keith William,  Dennis P. Lettenmaier, and  A.  Ian
     McLeod.  Assessment of Environmental  Impacts  Part One:
     Intervention  Analysis.   Environmental  Management,
     2(6):529-535, 1978.

22.  Juran, J.  M. , F.  M. Gryna, Jr.  and R.  S. Bingham Jr., eds.
     Quality Control  Handbook, Third Edition,  McGraw  Hill.  1979.

23.  Kempthorne, Oscar and R. R.  Allmaras.  Errors of Observation.
     In:  Methods of Soil  Analysis.  Part  1, Physical  and
     Mineralogical Properties including Statistics of Measurement
     and Sampling.  C. A. Black,  et al, ed.  American  Society of
     Agronomy,  Madison,  Wisconsin, 1965.  pp.  1-23.

                              90

-------
28,
24.   Ku,  Harry H.   Statistical Sampling and Environmental  Trace
     Organic Analysis.   In:   Trace Organic  Analysis:   A New
     Frontier  in  Analytical  Chemistry.  Proceedings  of  the 9th
     Materials  Research Symposium, National  Bureau of Standards,
     Gaithersburg, Maryland, 1978.

25.   Mason, Benjamin J.   Preparation of Soil Sampling Protocols.
     (EPA 600/4-83-020).  1983.

26.   Natrella, M. G.  Experimental  Statistics, NBS Handbook 91,
     U.S. Government Printing Office, Washington, D.C.  1963.

27.   Oregon State University.  Methods of Soil Analysis  Used  in
     the Soil  Testing Laboratory  at Oregon State  University.
     Special Report  321.  Corvallis,  Oregon.  1971.

     Peterson, R. G. and L. D. Calvin.  Sampling.  In:  Methods
     of  Soil Analysis.   Part  1,  Physical and Mineralogical
     Properties,  including  Statistics of Measurement  and Sampling.
     C.  A.  Black, et al, ed.   American  Society of  Agronomy,
     Madison,  Wisconsin,  1965.   pp.  54-71.

29.  Plumb, Russell H. Jr. Procedures for Handling and Chemical
     Analysis of  Sediment and Water  Samples.  EPA-48/05-5720-10 ,
     U.S.  Environmental  Protection  Agency/Corps  of  Engineers
     Technical  Committee on  Criteria for  Dredged  and Fill
     Material,  Grosse lie,  Michigan,  1981.

30   Rao,  P. V., P. S. C.  Rao,  J.  M. Davidson,  and L. C. Hammond.
     Use of Goodness-of-Fit Tests for Characterizing the  Spatial
     Variability of Soil  Properties.   Soil Science  Society  of
     America Journal, 3:274-278. 1979.

31   Skogerboe,  R.   K. and R.  Koirtyohann.  Accuracy Assurance  in
     the Analysis of Environmental  Samples.   Accuracy in Trace
     Analysis:  Sampling,  Sample  Handling, and Analysis Volumes
     I-II.   Seventh Materials Research Symposium.   Edited  by
     Philip D. La Fleur, NBS,  Gaithersburg,  MD, 1976.

 32   Snedecor,  George W.  and  William G.  Cochran.   Statistical
     Methods,  Seventh Edition.   The Iowa State  University  Press,
     Ames,  Iowa,  1982.  507 pp.

 33   U S.  Environmental  Protection Agency.    Guidelines
     Establishing Test Procedures for the Analysis of Pollutants;
     Proposed  Regulations.   Federal Register,  44:233,  pp.
     69464-69575, Washington, D.C.   1979.

 34   U S.  Environmenta-1 Protection Agency.   Documentation  of
     EMSL-LV  Contribution  to the Kellogg  Idaho Study.  EPA
     600/X-84-052,  Environmental Monitoring  Systems Laboratory,
     Las Vegas,  NV, 1984.
                               91

-------
35.   U.S. Environmental Protection Agency.   Quality Assurance
     Handbook  for Air Pollution  Measurement  Systems.
     EPA-600/9-76-005 ,  Environmental  Monitoring and Support
     Laboratory, Research Triangle Park, N.C.  1976.

36.   U.S.  Environmental  Protection  Agency.   Environmental
     Monitoring  at  Love  Canal.    Volumes  I  -  III.
     EPA-600/4-82-030 a-d, Washington, D.C.  1982.

37.   U.S. Environmental Protection Agency.   Documentation of
     EMSL-LV  Contribution to Dallas Lead Study  EPA-600/4-84-012.
     Environmental Monitoring Systems Laboratory, Las Vegas, NV,
     1984.

38.   U.S. Environmental Protection Agency.  National Oil and
     Hazardous  Substances Contingency Plan.   Federal Register.
     Vol 47(137):31180-31243.  Washington,  D.C.  1982.

39.   U.S.  Environmental  Protection  Agency.   Handbook for
     Analytical Quality  Control in Water  and  Wastewater
     Laboratories.   EPA-600/4-79-019.  U.S.  Environmental
     Protection Agency.   Environmental Monitoring and Support
     Laboratory, Cincinnati, OH.  45268.   1979.

40.  U.S. Code of Federal Regulations, Land  Treatment, 40 CFR,
     Part 264,  Subpart M, July, 1983.


41.  Webster,  R.  and  T.  M.  Burgess.  Optimal  Interpolation and
     Isarithmic Mapping of Soil Properties.  III.  Changing Drift
     and Universal  Kriging.   Journal of Soil Science,
     31(3):505-524.  1980.

42.  Youden, W. J. and  E.  H. Steiner.  Statistical Manual  of the
     Association of Official Analytical Chemists.  Association of
     Official Analytical Chemists, Washington, D.C.  1975.   88p.
                              92

-------
                         APPENDIX A
       TOOLS  FOR  ESTIMATING NUMBER OF SAMPLES  TO ACHIEVE
         SPECIFIED LEVELS OF PRECISION AND CONFIDENCE


The  graph,  shown  as Figure  A-l,  can be used to determine  the
number of  samples  required to  estimate the standard deviation
within a stated  percent of its true value with confidence levels
of 90, 95 and 99  percent.  A series of tables are also presented
which provide estimates of the number of samples necessary to
achieve a specified level of  confidence for both single-tailed
(75-99.5%)'and two-tailed (50-95%) t tests.   The precision is to
be stated and it  is  assumed that the coefficient of variation is
known.
                               93

-------
iJO!
9
a

*
s
4
3
I


- -







6
;
*
a
i
B
O
TJJ
2 io
b«
•
afi

0) 6
^ 5
fl) 4
3
I




•
















































, 	 ,































—














-

























|



















































































































	
5














i

-,

































I1:














—










— '































~

;
^
4-t—

_
it,
- —
tt.



























i
a=
^
•V
A
• i *
— —
Mil








1 ' i :







1 1




i— • — -

1 :

I

— —
V '
*
jg

























na
i!




i
L



L
^




















^
v






i


1
'




















,
s







\
1
^
1
















:i

— , —



i

E
<




N

*c











_





-fJ4r









\

+\
L %
I
— V
•;;•
1 ! ' '
| i| ,


— -






""" 	 '

:

-uiii.





tin






^ .
\
V*
>s
— V
J








-*-«
— ,


.















'
s
1
S
\
1



























:
1
-
S

\
\




























\


1
s












--
' -












\

\
V,

\
f







i

—





I'M










^
• , \


k
^


>
^
-1 — —



!
r~~
^

--L...

1










1 	



s
\
^

s
s
^
A~


ni

i

±

	



















-V
s
1

X^
£
^





—























I
1
i
j
V-



:


























i -
i
-!
=
V







X






















1
i
3
V


•
— r— —
^7

— ' — T~
— J —






















— — -
t=
bi^
-r-r*
^
^
\
XT



























--
, -
A
>'
\





























s































S'






-































— i —
g
























V
*v


....




M











*• ••"











i i.
!



, — . — »



























9
Figure A-l. Number of degrees of freedom required to estimate the Standard
  Deviation within P% of its True Value with confidence level Y«

                                   94

-------
    TABLE A-l.   ESTIMATED NUMBER OF SAMPLES TO ACHIEVE SPECIFIED
    LEVELS OF PRECISION AND CONFIDENCE WHEN COEFFICIENT OF VARIATION
    (CV) IS KNOWN.
                         Confidence Level
                     50(%): two-tailed  t  test
                     75(%): one-tailed  t  test
                          Precision p(%)

CV (%)      1      25   10    15    20    25    30   40   50   75   100
1
2
4
6
8
10
20
30
40
50
100
1
3
8
17
30
47
182
410
728
1138
4550
1
1
3
5
8
12
47
104
182
285
1138
1
1
1
2
2
3
8
17
30
47
182
1
1
1
1
1
1
3
5
8
12
47
1
1
1
1
1
1
2
3
4
6
21
1
1
1
1
1
1
1
2
3
4
12
                                              11111     1
                                              11111     1
                                              11111     1
                                              11111     1
                                              11111     1
Confidence Level
60(%): two-tailed t test
SOU): one-tailed t test
Precision p(%)
CV (Z)
1
2
4
6
8
10
20
30
40
50
100
1
2
4
13
27
47
72
284
638
1134
1771
7084
2
1
2
4
8
13
19
72
160
284
443
1771
5
1
1
2
2
3
4
13
27
47
72
284
12
1
1
1
1
2
2
4
8
13
19
72
J_5
1
1
1
1
1
1
2
4
6
9
33
20
1
1
1
1
1
1
2
3
4
6
19
25_
1
1
1
1
1
1
2
2
3
4
13
                                                  30   40   50   75    100
                                                   1111      1
                                                   1111      1
                                                   1111      1
                                                   1111      1
                                                   1111      1
                                                   1111      1
                                                   1111      1
                                                   2111      1
                                                   2221      1
                                                   3221      1
                                                   9642      2
                                                             (Continued)

-------
                             TABLE A-l.
                         Confidence Level
                     70(%): two-tailed t test
                     85(%): one-tailed t test
                          Precision p(%)

CV (Z)      1      25   10   15   20   25   30   40   50   75   100
     1      2      11111111111
     2      6      21111111111
     4     19      62111111111
     6     42     11     321111111     1
     8     74     19     422111111     1
    10    115
    20    430
    30    967
    40   1719
    50   2686
   100  10742
1
2
6
11
19
28
115
242
430
672
2686
1
1
2
3
4
6
19
42
74
115
430
1
1
1
2
2
2
6
11
19
28
115
1
1
1
1
2
2
3
6
9
13
52
1
1
1
1
1
2
2
4
6
8
28
1
1
1
1
1
1
2
3
4
6
19
1
1
1
1
1
1
2
2
3
4
13
Confidence Level


80(%)
90(%)
: two-tailed
: one-tailed
t
t
test
test


Precision p(%)
2
2
3
8
16
28
43
165
370
657
1027
5
1
2
2
4
6
8
28
61
107
165
10
1
1
2
2
3
3
8
16
28
43
15
1
1
1
2
2
2
4
8
13
20
20
1
1
1
1
2
2
3
5
8
12
25
1
1
1
1
1
2
2
4
6
8
30
1
1
1
1
1
1
2
3
4
6
CV (%)      1      25   10   15   20   25   30   40   50   75   100
     1      3      21111111111
     2      8      32111111111
     4     28      82211111111
     6     61     16     422111111     1
     8    107     28     632211111     1
    10    165
    20    657
    30   1479
    40   2628
    50   4106
   100  16424   4106   657  165   75   43   28   20   12
                                                            (Continued)
                                    96

-------
                             TABLE A-l.
                         Confidence Level
                     90(%): two-tailed  t  test
                     95(%): one-tailed  t  test
                          Precision p(%)

CV (X)      1      25    10    15    20    25    30    40   50   75   100
                                                                  1     1
                                                                  1     1
                                                                  1     1
                                                                  1     1
                                                                  1     1
                                                                  1     1
                                                                  2     2
                                                                  2     2
                                                                  2     2
1
2
4
6
8
10
20
30
40
50
100
5
13
46
100
174
271
1083
2435
4329
6764
27055
2
5
13
27
46
70
271
609
1083
1691
6764
2
2
4
6
9
13
46
100
174
271
1083
1
2
2
3
4
5
13
27
46
70
271
1
1
2
2
3
3
7
13
21
32
121
1
1
2
2
2
3
5
8
13
19
70
1
1
1
2
2
2
3
6
9
13
46
1
1
1
2
2
2
3
5
7
10
32
1
1
1
1
2
2
2
3
5
6
19
1
1
1
1
1
2
2
3
3
5
13
                          Confidence Level
                      95(%):  two-tailed t test
                    97.5(%):  one-tailed t test
                           Precision p(%)

cv  (%)       i    	2   	5   J^_152025_30405025_iPJ)


                                                                  1     1
                                                                  1     1
                                                                  1     1
                                                                  1     1
                                                                  2     1
                                                                  2     2
                                                                  2     2
                                                                  3     2
                                                                  3     3
                                                                  4     3
                                                                  9     6
                                   97                        (Continued)
1
2
4
6
8
10
20
30
40
50
100
6
18
64
139
246
385
1537
3458
6147
9604
38416
3
6
18
37
64
99
385
865
1537
2401
9604
2
3
5
8
12
18
64
139
246
385
1537
2
2
3
4
5
6
18
37
64
99
385
1
2
2
3
3
4
9
18
30
45
171
1
2
2
2
3
3
6
11
18
27
99
1
1
2
2
2
3
5
8
12
18
64
1
1
2
2
2
2
4
6
9
13
46
1
1
2
2
2
2
3
5
6
9
27
1
1
1
2
2
2
3
4
5
6
18

-------
                             TABLE A-l.
                         Confidence Level
                     98(%): two-tailed t test
                     99(2): one-tailed t test
                          Precision p(%)

CV (%)   	1   	2     5   10   15   20   25   30   40   50   75   100
                                                  2111     1
                                                  2221     1
                                                  2222     2
                                                  2222     2
                                                  2222     2
    40   8659   2165   347   90   42   25   17   13    9    7    4
    50  13530   3383   542  136   63   37   25   18   12    9    5
   100  54119  13530  2165  542  241  136   90   63   37   25   13
1
2
4
6
8
10
20
30
8
25
90
195
347
542
2165
4871
4
9
25
52
90
136
542
1218
2
3
6
11
17
25
90
195
2
2
3
5
7
9
25
52
2
2
3
3
4
5
13
25
2
2
2
3
3
4
9
15
2
2
2
2
3
3
6
11
                         Confidence Level
                     99(%): two-tailed t test
                   99.5(Z): one-tailed t test
                          Precision p(2)

CV (2)   	1   	2     5   10   15   20   25   30   40   50   75   100
1
2
4
6
8
10
31
110
239
425
5
10
31
64
110
3
4
8
13
21
2
3
4
6
8
2
2
3
4
5
2
2
3
3
4
2
2
3
3
3
2
2
2
3
3
2
2
2
3
3
2
2
2
2
3
1
2
2
2
2
1
2
2
2
2
    10-   664    166    31   10    6    5    4    4    3    3    2     2
    20   2655    664   110   31   16   10    8    6    5    4    3     3
    30   5973   1493   239   64   31   19   13   10    7    6    4     3
    40  10618   2655   425  110   51   31   21   16   10    8    5     4
    50  16590   4148   664  166   78   46   31   22   14   10    6     5
   100  66358  16590  2655  664  295  166  110   78   46   31   16    10
                                 98

-------
                         APPENDIX B
 TABLES FOR USE IN  CALCULATING CONFIDENCE AND TOLERANCE LIMITS
           AND  JUDGING  THE VALIDITY OF MEASUREMENTS


A series  of  tables are presented which can be used  to calculate
confidence intervals  for averages,  percentiles  of  the
t-distribution, the standard  deviation,  the  tolerance  interval
for individuals and critical  values  for discarding  invalid
measurements.
                             99

-------
TABLE B-l.  PERCENTILES OF THE t DISTRIBUTION
Confidence Level (%)

20
30
60
Confidence Level (%)
df
1
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
40
60
120
oc
60
.325
.289
.277
.271
.267
.265
.263
.262
.261
.260
.260
.259
.259
.258
.258
.258
.257
.257
.257
.257
.257
.256
.256
.256
.256
.256
.256
.256
.256
.256
.255
.254
.254
.253
70
.727
.617
.584
.569
.559
.553
.549
.546
.543
.542
.540
.539
.538
.537
.536
.535
.534
.534
.533
.533
.532
.532
.532
.531
.531
.531
.531
.530
.530
.530
.529
.527
.526
.524
80
1.376
1.061
.978
.941
.920
.906
.896
.889
.883
.879
.876
.873
.870
.868
.866
.865
.863
.862
.861
.860
.859
.858
.858
.857
.856
.856
.855
.855
.854
.854
.851
.848
.845
.842
:l-a/2 for two-tailed test
80
:l-a for
90
3.078
1.886
1.638
1.533
1.476
1.440
1.415
1.397
1.383
1.372
1.363
1.356
1.350
1.345
1.341
1.337
1.333
1.330
1.328
1.325
1.323
1.321
1.319
1.318
1.316
1.315
1.314
1.313
1.311
1.310
1.303
1.296
1.289
1.282
90
one-tailed
95
95
test
97.5
6.314 12.706
2.920
2.353
2.132
2.015
1.943
1.895
1.860
1.833
1.812
1.796
1.782
1.771
1.761
1.753
1.746
1.740
1.734
1.729
1.725
1.721
1.717
1.714
1.711
1.708
1.706
1.703
1.701
1.699
1.697
1.684
1.671
1.658
1.645
4.303
3.182
2.776
2.571
2.447
2.365
2.306
2.262
2.228
2.201
2.179
2.160
2.145
2.131
2.120
2.110
2.101
2.093
2.086
2.080
2.074
2.069
2.064
2.060
2.056
2.052
2.048
2.045
2.042
2.021
2.000
1.980
1.960
98

99
31.821
6.965
4.541
3.747
3.365
3.143
2.998
2.896
2.821
2.764
2.718
2.681
2.650
2.624
2.602
2.583
2.567
2.552
2.539
2.528
2.518
2.508
2.500
2.492
2.485
2.479
2.473
2.467
2.462
2.457
2.423
2.390
2.358
2.326
99

99.5
63.657
9.925
5.641
4.604
4.032
3.707
3.499
3.355
3.250
3.169
3.106
3.055
3.012
2.977
2.947
2.921
2.898
2.878
2.861
2.845
2.831
2.819
2.807
2.797
2.787
2.779
2.771
2.763
2.756
2.750
2.704
2.660
2.617
2.576
                      100

-------
TaMfB-2.  CI=]T± A&R). Confidence Interval for Averages'
Prob.
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.0 1
0.05
0.01
0.05
0.01
k
1
2
3
4
5
6
7
8
9
10

2
6.36
31.9
0.879
2.11
0.360
0.660
0.210
0.350
0.140
o.::6
0.102
0.157
0.079
0.117
0.063
0.094
0.053
0.076
0.044
0.064

3
1.30
3.00
0.316
0.474
0.156
0.273
0.096
0.142
0.066
0.095
0.050
0.070
0.039
0.055
0.032
0.044
0.027
0.036
0.023
0.031

4
0.719
1.36
0.206
0.312
0.104
0.150
0.065
0.092
0.046
0.063
0.034
0.047
0.027
0.037
0.022
0.030
0.018
'0.025
0.016
0.021

5
0.505
0.865
0.154
0.227
0.079
0.112
O.OSO
0.070
0.035
0.049
0.027
0.036
0.021
0.029
0.017
0.023
0.014
0.019
"0.012
0.016
n
6
0.402
0.673
0.125
0.179
0.065
0.091
0.042
0.057
0.030
0.040
0.022
0.030
0.018
0.024
0.014
0.019
0.012
0.016
0.010
0.014

7
0.336
0.514
0.106
0.150
0.056
0.077
0.036
0.048
0.025
0.034
0.019
0.026
0.015
0.020
0.012
0.016
0.010
0.014
0.009
0.012

8
0.291
0.430
0.093
0.131
0.049
0.068
0.032
0.043
0.022
0.030
0.017
0.023
0.013
0.018
0.01 1
0.014
0.009
0.012
0.008
0.010

9
0.256
0.379
O.OS4
0.116
0.044
0.060
0.028
0.038
0.020
0.027
0.015
0.020
0.012
0.016
0.010
0.013
0.008
0.011
0.007
0.009

10
0.232
0.338
0.076
0.105
0.040
0.054
0.026
0.035
0.018
0.025
0.014
0.019
0.011
0.015
0.009
0.012
0.007
0.010
0.006
0.008
  •Given * subgroups of n numbers, the confidence interval is X ± A(1R).




   From Bauer, 1971.
                                  101

-------
Table B-3.  Single Classification Factor (cx) to Estimate Standard
Deviation from Range, and Equivalent Degrees of Freedom (/)a
                                  s = —


k
1
2
3
4
5
6
7
8
9
10


/
1.0
1.9
2.8
3.7
4.6
5.5
64
7?
8.1
9.0

I
Ct
.41
.28
.23
71
19
18
.17
.16
.15
14


/
2.0
3.8
5.7
7.5
9.1
11.1
12.9
14.8
16.6
IF 4

I
C|
.91
.81
.77
.75
.74
.71
.72
.71
.70
69

t
f
2.9
5.7
8.4
11.2
11.9
16.6
19.4
22.1
24.9
776

1
ct
2.24
2.15
2.12
2.11
2.10
2.09
2.08
2.08
2.07
707


/
3.8
7.5
11.1
14.7
18.4
22.0
25.6
29.3
32.9
1fi5

S
C|
2.48
2.40
2.38
2.37
2.36
2.36
2.35
2.35
2.34
714

(
/
4.7
9.2
13.6
18.1
22.6
27.1
31.5
36.0
40.5
449
n
6
Cl
2.67
2.60
2.58
2.57
2.56
2.56
2.56
2.55
2.55.
7S5


/
5.5
10.8
16.0
21.3
26.6
31.9
37.1
42.4
47.7
S?9

7
Cl
2.83
2.77
2.75
2.74
2.73
2.73
2.73
2.72
2.72
2.72


/
6.3
12.3
18.3
24.4
30.4
36.4
42.5
48.5
54.5
60.6

g
c,
2.96
2.91
2.89
2.88
2.87
2.87
2.87
2.86
2.86
2.86


/
7.0
13.8
20.5
27.3
34.0
40.8
47.6
54.3
61.1
67.8

»
C|
3.08
3.02
3.01
3.00
2.99
2.99
2.98
2.98
2.98
2.98

1
/
7.7
15.1
22.6
30.1
37.5
45.0
52.4
i9.8
67.3
74.8

0
Cl
3.18
3.13
3.11
3.10
3.10
3.10
3.09
3.09
3.09
3.09
From Bauer, 1971,
                                  102

-------
Table B-4. TI=JP ± I(LR). Tolerance Inlenalfor Individuals
Prob.
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
0.05
0.01
k
1
2
3
4
5
6
7
8
9
10

2
8.99
45.1
1.76
4.22
O.SS2
1.62
0.594
0.990
0.443
0.715
0.353
0.544
0.296
0.438
0.252
0.376
0.225
0.322
0.197
0.286

3
2.25
5.20
0.774
1.16
0.486
0.819
0.332
0.492
0.256
0.368
0.212
0.297
0.179
0.252
0.157
0.216
0.140
0.187
0.126
0.170

4
1.44
2.72
0.583
0.882
0.360
0.520
0.260
0.368
0.206
0.282
0.166
0.230
0.143
0.196
0.124
0:170
0.108
0.150
0.101
0.133

5
1.13
1.93
0.467
0.718
0.306
0.434
0.224
0.313
0.175
0.245
0.148
0.197
0.172
0.108
0.145
0.094
0.127
0.085
0.113
it
6
0.985
1.65
0.433
0.620
0.276
0.386
0.206
0.279
0.164
0.219
0)32
0.180
0.117
0.156
0.097
0.132
0.088
0.118
0.077
0.108

7
0.889
1.36
0.397
0.561
0.257
0.353
0.190
0.254
0.148
0.201
0.123
0.168
0.105
0.1-40
0.090
0.120
0.079
0.111
0.075
0.100

8
0.823
1-22
0.372
0.524
0.240
0.333
0.181
0.243
0.139
0.190
0.118
0.159
0.097
0.135
0.088
0.112
0.076
0.101
0.072
O.OS9

9
0.768
1.14
0.356
0.492
0.229
0.312
0.168
0.228
0.134
0.181
0.110
0.147
0.095
0.127
0.085
0.110
0.072
0.099
0.066
0.085

10
0.734
1.07
0.340
0.470
0.219
0.296
0.164
0.221
0.127
0.177
0.108
0.145
0.092
0.124
O.OSO
0.108
0.066
0.095
0.050
0.080
From  Bauer,  1971.
                          103

-------
Table B-5.   Critical
Values for  Discarding
Invalid Measurements
a
3
4
5
€
7
8
9
10
11
12
13
14
15
20
It
1.53
1.05
0.86
0.76
0.69
0.64
. 0.60
O.S8
0.56
0.54
0.52
0.51
0.50
0.46
  The probability is approximately
 0.95  that if 't, = | X - X \ IK is
 greater than tabulated t, the value
 being investigated is invalid.
From Bauer|  1971t
             104

-------