APTD-1132
      QUALITY CONTROL
                  PRACTICES
            IN PROCESSING
AIR POLLUTION  SAMPLES
U.S. ENVIRONMENTAL PROTECTION AGENCY
     Office of Air and Water Programs
  Office of Air  Quality Planning and Standards
    Research Triangle Park, N. C. 27711

-------
                                  APTD-U32
QUALITY CONTROL PRACTICES

           IN PROCESSING

    AIR POLLUTION  SAMPLES
                Prepared by

      PEDCo-Environmental Specialists, Inc.
           Suite 13, Atkinson Square
           Cincinnati, Ohio 45246
           Contract No. 68-02-021]
         EPA Project Officer: Neil Berg
                 Prepared for

        ENVIRONMENTAL PROTECTION AGENCY
        Office of Air and Water Programs
   Office of Air Quality Planning and Standards
  Research Triangle Park, North Carolina  27711

                 March 1973

-------
The APTD (Air Pollution Technical Data) series of reports is issued by
the Office of Air Programs, Environmental Protection Agency, to report
technical data of interest to a limited number of readers.  Copies of
APTD reports are available free of charge to Federal employees, current
contractors and grantees, and non-profit organizations - as supplies
permit - from the Air Pollution Technical Information Center, Environ-
mental Protection Agency, Research Triangle Park, North Carolina  27711,
or may be obtained, for a nominal cost, from the National Technical
Information Service, 5285 Port Royal Road, Springfield, Virginia  22151.
This report was furnished to the Environmental Protection Agency by
PI-DCo-Environmental Specialists, Inc., Cincinnati, Ohio, in fulfillment
of Contract No. 68-02-0211.  The contents of this report are reproduced
herein as received from the contractor.  The opinions, findings, and
conclusions expressed are those of the author and not necessarily
those of the Environmental Protection Agency.
                         Publication No. APTD-1132
                                     u

-------
                     ACKNOWLEDGMENT






     This report was prepared under the direction of Mr.



David W. Armentrout.  Principal authors were Mr.  George A.



Jutze, Mr. Charles E. Zimmer, and Mr.  Richard W.  Gerstle.



Mr. Robert J. Bryan of Pacific Environmental Services was



principal author of Chapter 3.






     Project Officer for the Environmental Protection Agency



was Mr. Neil Berg, Jr.  The authors appreciate Mr. Berg's



contributions to the concepts presented in this project.






     Several federal agencies and private organizations were



visited to obtain information used in preparing this report.



The authors particularly appreciate the assistance given by



the Center for Disease Control in Atlanta, the American



Society for Testing and Materials, the National Bureau of



Standards, and Armco Steel Corporation.






     Mrs. Anne Cassel was responsible for editorial review,



and Mr. Chuck Fleming reviewed the graphics.
                          111

-------
                     TABLE OF CONTENTS
                                               Page Number
Acknowledgment 	    iii
List of Tables	    viii
List of Figures	    ix

1.0  INTRODUCTION . . .	    1-1

1.1  General Considerations 	    1-1
1.2  Importance of 'Quality Control 	    1-2
1.3  Objectives and  Scope of These
      Guidelines 	    1-5

2.0  THE MAJOR ELEMENTS OF QUALITY CONTROL ....    2-1

2.1  Controlling Physical Parameters 	    2-3
2.2  Analysis of the Total Measurement
      System 	    2-4

     2.2.1  Precision and Accuracy 	    2-4
     2.2.2  Control  Chart Techniques 	    2-5
     2.2.3  Review and Control Procedures 	    2-11

3.0  QUALITY CONTROL IN ATMOSPHERIC SAMPLING ..    3-1

3.1  Controlling the Physical Parameters 	    3-2

     3.1.1  Environment 	    3-2
     3.1.2  Reagents and Supplies 	    3-4
     3.1.3  Calibration Materials 	    3-6
     3.1.4  Calibration and Maintenance
             Procedures 	    3-8
     3.1.5  Design and Maintenance of Probes
             and Manifolds 	    3-14

3.2  Analyzing and Controlling the Total
      Measurement System . . .	    3-15

     3.2.1  Functional Analysis .	    3-16
     3.2.2  Sensitivity Analysis  	    3-18
     3.2.3  Compiling and Using Data
             Histories	    3-21

3.3  Documentation and Demonstration of Quality
      Control 			    3-28

     3.3.1  Records	    3-28
     3.3.2  Reports  	    3-31
                             v

-------
                    TABLE OF CONTENTS (Continued)

                                            Page Number

4.0  QUALITY CONTROL IN SOURCE SAMPLING 	    4-1

4.1  General Considerations 	    4-1
4.2  Controlling the Physical Parameters ...    4-2

     4.2.1  Interferences 	    4-2
     4.2.2  Sampling Rate and Sample Volume.    4-3
     4.2.3  Equipment Maintenance and
             Calibration 	    4-3
     4.2.4  Conducting the Emission Test ...    4-6

4.3  Analyzing the Total Measurement System.    4-10

5.0  QUALITY CONTROL IN THE ANALYTICAL
      LABORATORY 	    5-1

5.1  Introduction 	    5-1
5.2  Controlling Physical Parameters 	    5-2

     5.2.1  Laboratory Support Services ....    5-2
     5.2.2  Chemicals and Reagents 	    5-5
     5.2.3  Instruments 	    5-9
     5.2.4  Analytical Technique 	    5-18

5.3  Statistical Methods 	    5-20

     5.3.1  Defining Performance Levels ....    5-20
     5.3.2  Choosing Statistical Techniques.    5-21
     5.3.3  Constructing Range Control
             Charts 	    5-25
     5.3.4  Constructing Coefficient of
             Variation Charts 	    5-28
     5.3.5  Determining Accuracy 	    5-31
     5.3.6  Control Charts for Accuracy ....    5-34

5.4  Interlaboratory Proficiency Testing ...    5-39
5.5  Tracing Errors 	    5-42

6 .0  DATA HANDLING AND REPORTING 	    6-1

6.1  General 	    6-1
6.2  Data Recording 	    6-1
                           VI

-------
                    TABLE OF CONTENTS (Continued)

                                              Page Number

     6.2.1  Data Errors in Intermittent
             Sampling	    6-2
     6.2.2  Data Errors in Continuous
             Sampling 	    6-3

6. 3  Data Validation 	    6-6

     6.3.1  Data Validation for Manual
             Techniques	    6-6
     6.3.2  Data Validation for Computerized
             Techniques 	    6-7

6.4  The Statistical Approach to Data
      Validation	    6-9

     6.4.1  Maintaining Data Quality in
             Manual Data Reduction Systems ...    6-9
     6.4.2  Acceptance Sampling Applications .    6-10
     6.4.3  Sequential Analysis                  6-14

7.0  REFERENCES                                  7-1

Appendix A - Statistical Formulae and
             Definitions                         A-l

Appendix B - Tables of Factors for Con-
             structing Control Charts            B-l

Appendix C - Calibration Curves from
             Regression Analysis by the
             Method of Least Squares             C-l
                           VI1

-------
                     LIST OF TABLES

Table                                       Page Number

3.1   Environmental Influences and Effects .     3-3
3.2   Effects of Housekeeping Practices on
       System Performance	     3-5
3.3   Span Drift for Analyzer A for 26
       Two-Week Periods	     3-25

4.1   Elements for Control of Interferences.     4-2
4.2   Flow Meter Types and Accuracy 	     4-4
4.3   Example Determination of Pitot Tube
       Calibration 	     4-5

5.1   Techniques for Quality Control of
       Laboratory Support Services 	     5-3
5.2   Guidelines for Quality Control of
       Chemicals and Reagents	     5-6
5.3   Restandardization Requirements 	     5-7
5.4   Techniques for Quality Control of
       Analytical Instruments 	     5-10
5 . 5   S02 Calibration Data	     5-15
5.6   Problems in Assessing Analyst
       Performance 	     5-19
5.7   Data Used to Construct Scatter
       Diagrams 	     5-25
5.8   Computation of Control Limits for Range
       Control Charts	     5-26
5.9   Computation of Control Limits for
       CV-Charts 	     5-30
5.10  Techniques for Determining Accuracy ..     5-33
5.11  Accuracy Data for Percent Nitrogen ...     5-36
5.12  Percent Recovery Data		     5-37
5.13  Variations in Method Procedures 	     5-41
5.14  Classification of Error Sources 	     5-43

Appendix B:  Table BI - Factors for Con-
      structing Control Charts for
      Averages 	     B-l
             Table BII - Factors for Con-
      structing Control Charts for Range        B-2

             Table Bill - Factors for Con-
      structing Control Charts for Coefficient
      of Variation 	     B-3

             Table BIV - The t Distribution
      (Two-Tailed Tests)	     B-4

Appendix C:  Table C-I - Calibration Data
      for SO? Determination 	     C-2

                         viii

-------
                    LIST OF FIGURES
Figure                                        Page Number

2.1  Quality Control in Total Analytical
      System 	   2-2

3.1  Functional Analysis of High Volume
      Suspended Particulate Sampling 	   3-19
3.2  Control Chart for Span Drift 	   3-27
3.3  Instrument Maintenance Form -
      California ARB	   3-30

4. 1  Field Data Sheet 	   4-8
4.2  Field Data Sheet 	   4-9
4.3  Expected Errors Incurred by Non-
      Isokinetic Sampling	   4-11

5.1  Spectrophotometer Weekly Function Check  .   5-12
5.2  Standardization Check at 420 nm	   5-13
5.3  Calibration Curve for S02 Determination  .   5-16
5.4  Scatter Diagrams for Determining Control
      Charts	   5-23
5.5  Control Chart for Range 	   5-26
5.6  Interpretation of Control Charts 	   5-29
5.7  Coefficient of Variation Chart  	   5-32
5.8  Accuracy Control Chart fronv Analysis of
      a Primary Standard 	 	   5-36
5.9  Accuracy Control Chart for % Recovery ...   5-38
5.10 Procedures for Tracing Sampling Errors ..   5-45

6.1  OC Curve 	   6-13
6.2  Sequential Test for the Error Rate of
      a Data Analyst 	   6-17
6.3  Operating Characteristic Curve  	   6-24

Appendix C:  Figure C-l - Calibration Curve
     for S0? Determination 	   C-3
                            IX

-------
1.0  INTRODUCTION




1.1  General Considerations




     State,  regional and local air pollution control labo-




ratories perform an array of technical services involving




sample collection, sample analysis and data validation.




To demonstrate that reliable data are reported, the technical




services group must establish a quality control program.




We define a quality control program as:




     "...  THE PROGRAM APPLIED TO ROUTINIZED SYSTEMS




     (I.E.,  SYSTEMS COMPOSED OF METHODS, EQUIPMENT AND




     PEOPLE) IN" ORDER TO EVALUATE AND DOCUMENT THE




     ABILITY OF A FUNCTION, ACTIVITY, OR PERSON TO PRO-




     DUCE RESULTS WHICH ARE VALID WITHIN PREDETERMINED




     ACCEPTANCE LIMITS".




     It must be emphasized that we are speaking of "quality




control" not "quality assurance", which we consider as en-



compassing both quality control and methods standardization.




Further, quality control programs should be successfully




implemented prior to involvement of a technical services




group in cooperative-laboratory standardization efforts.




     A quality control program is developed to minimize




sources of variation inherent in all analytical and technical




functions.  Through the use of standard operating proce-




dures and statistical techniques, items such as determinate




errors are identified and  controlled.  The effects of  random
                         1-1

-------
errors are measured and used to express the degree of con-




fidence to be placed in the analytical data and to deter-




mine when the process that generates the data is not func-



tioning properly.




     The quality control program should be designed so that




the supervisor, through his technicians, can set up a pro-




tocol for performing each operation involved in sample



collection, analysis, and data handling.  Implementation of




the quality control program requires that descriptions of



each procedure be readily available to the laboratory staff.



This document is intended to provide the administrator or



supervisor with guidelines for establishing a detailed



quality control program that is consistent with his specific



needs and objectives.



1.2  Importance of Quality Control



     The functions of a technical services group in an air



pollution control program encompass both field and labora-



tory operations, including surveillance of pollution sources,



acquisition of ambient air quality data, episode criteria



monitoring and performance of special studies.  The technical



services group provides qualitative and quantitative data to



be used at all levels of program operation.  Consequently,



each sample procured must be adequately representative of



emissions from the pollutant source or of the atmosphere



sampled.  In addition, analysis of the sample, performed in
                         1-2

-------
the field or in the laboratory, by automatic instrumentation



or by wet chemical means, must provide data that accurately



describe the qualitative or quantitative characteristics of



the sample.   In some instances, incorrect data could lead



to faulty interpretations, and this could be worse than no



data at all.



     Important and far-reaching decisions may be based on



air quality and emission data which sometimes may be pre-



sented as evidence in courts of law.  Aerometric data will



be used to determine whether standards are being met.  If



results indicate violation of a regulation, the appropriate



enforcement group is required to take action.  With the



current emphasis on legal action and social pressures to



abate pollution, personnel in the technical services group



must be made aware of their responsibility to provide re-



sults that present a reliable description of the sample.  In



addition, the analyst should know that his professional



competence, the procedures he uses, and the values he reports



may be presented and challenged in court.  To meet such a



challenge, all data must be supported by a detailed program



that documents the proper control o.f all factors affecting



the reported result.



     In testing of pollutant sources, the economic implica-



tions alone are sufficient reason for exercising extreme



care in sampling and analysis.  Plant operators-may use
                         1-3

-------
these data as a basis for decisions to change a process,



install control devices, or even to construct new facilities




     Special projects and short-range development studies



in air pollution control must be based on sound laboratory




data.  The value of a development effort will depend on the



validity of laboratory results.   The progress of a special



study and alternative experimental pathways,  especially,



are evaluated on the basis of accumulated data; the final



results and recommendations are  usually evaluated by pre-



sentation of data such as averages, standard  deviations,



ranges, frequency distribution,  and confidence limits.



     For these reasons and many  more, a quality control



program to assess and document the reliability of data



is essential.  Although most chemists, engineers, and



technicians practice some personal form of quality control,



they do so at varying levels and degrees of proficiency,



depending on such factors as professional integrity, back-



ground and training, and understanding or awareness of the




scope and importance of the work they are engaged in.  Un-



fortunately, these informal efforts at quality control are



often inadequate and usually fail to provide  adequate docu-



mentation.



     Because of the routine nature of the normal workload,



or, perhaps, the pressures of occasional high-priority



"rush" projects, quality control can be neglected easily.
                         1-4

-------
Therefore, in order to assure validity and reliability of

the final results,  it is imperative that each agency re-

quire a specific control program for every sampling pro-

cedure and analytical test.

1.3  Objectives and Scope of These .Guidelines

     The objective of this document is to provide guidance

or instruction for agency personnel at all levels to assist

in the development or modification of quality control pro-

grams .  The diverse functions of the many types of agencies

that will use these guidelines necessitate some generali-

zation; whenever possible, however, we provide examples to

illustrate application of the principles cited.  We attempt

to answer such questions as:

     1.   "Are the data valid?"

     2.   "Are they good enough for the intended use?"

     3.   "Is the technical services group performing
          consistently?"

     4.   "How can we be sure that our equipment is
          capable of providing correct results and
          is operating so that it does?"

     5.   "How should we document or demonstrate our
          level of performance to others?"

     In the following pages Section 2 describes the major

characteristics  of quality control, indicating the types of

activities available for incorporation into  a quality con-

trol program.  Sections 3, 4, and  5 deal with atmospheric

monitoring, source monitoring, and laboratory operations,
                          1-5

-------
respectively, considering for each of these functions three

major phases of quality control:
     0  Control of physical parameters
     0  Analysis of the measurement process
     0  Documentation and demonstration of quality control.
     Section 6 describes the basic techniques of data handl-

ing and data evaluation, particularly as they are applied in

air pollution control laboratories.
                         1-6

-------
2.0  THE MAJOR ELEMENTS OF QUALITY CONTROL


     The major elements of a quality control program can


be considered broadly in two categories:  control of the


physical parameters and analysis of the total measurement


process.  We differentiate these two aspects of quality


control chiefly as an aid to understanding.  Control of


physical parameters entails such functions as calibration,


maintenance, and standardization of materials.  Analysis of


the total measurement process is a management tool for con-


tinuous evaluation of the performance capability of a


technical services laboratory and of the data the laboratory


produces.  Analysis encompasses statistical monitoring


techniques in conjunction with such evaluation techniques


as analyzing spiked samples, replicate analysis, and inter-


laboratory comparisons.  We may say, then, that control of


physical parameters is designed to reduce the frequency of


occurrence of errors in laboratory operation and that pro-


cedures analysis is applied to determine the effectiveness


of the physical control measures.  An effective quality con-


trol program requires both kinds of effort, aimed at one


goal:  the generation of valid data:  Figure 2.1 illustrates

                                  /
how each category relates to each of three phases of the


total sample analysis system:
     0  Sample collection
     0  Sample analysis
        Data acceptance and performance evaluation
                         2-1

-------
             • PHASE I: SAMPLE COLLECTION
                                     Collect
                                     Sample
                                                                                                                                 Yes
                        PHASE III:  PERFORM EVALUATION
                                                                                                              PHASE  II:  SAMPLE ANALYSIS
                                                                                                                                     V
> Yes fr
) ^
Review
Analyst
Performance
W,
P<
                                                                                          Review
                                                                                          Control
                                                                                         Criteria
 Check Procedures

or  Physical Parameters
Decision Block
                                                          Figure  2.1   Quality  control in total analytical  system
Measurement System Checks
                              Procedure

-------
 2 .1  Contro.llin.g, :Phy,si,cal , Parameters

      Control- measures to assure .the production, ot valid

 data , can, also be ..thoughjt,. fpf s lmp.ly^ as jrgpp.d ppe^ating practices,

 in  both-, sampling r,an,d laJDora;tpr,y . anaJLys^s . ^Although details

 vary cwi.th ^each, ,labpratpry .and,. application,,,, the, practices

 usual ly.-, considered -are these.:.
 .,     °rt  .£a liberation,,
      0  Functidh'al checks'" vof "component s". in. the  measure-'
    ... ;ment ,s.ystem
      0  Scheduled preveh't'ative" maintenance   ...... "
    .  °  Nonscheduled. .maintenance^  ....
    • n   <-^; -,.'<..._- -J'-».\  - -:v. -Maif. 1': .:. J-^J-.< ...J.^j .
      0  Standardization ;of materials
    .  ° . .Control of, interferences. .  , ,.r
   •••• .-'V- *Vr~    -, ."-'-' -v-L*-t^-  r».v.-^'.. -A ji i,- ,../:.,. if •'..
      0  Recording procedures : .  ,   .-
      0 , Batch--checking_procedures:1 , ..„
     "•:•.-- -i--M; 'i,-'^ i'.w -Ljp' .~~-;*.r ,> i- -t-.-.K&ttl?..
      °  Go'od  housekeeping  technicjues
      °. ^.-Control, prpceduresr for  auxil-liary  services

      Examples of, .these., procedures, and, criteria  for applying

 them are given  in the sections, dealing with.^field. and labo-

 ratory .-.operations,., ...Management .of  each, .technical .services

 group wi>ll, require. decisions, .regarding jwhich. of these measures

 should be .-applied , how .often , ..and ..specifically  in what  way

 ( detai led ^p rptp.cp.l ),.. A.irAmqng _the_many elements that will in-

, f luence these decisions., are ^ such practical matters as

• ayailabilitv,  and  cost, J-pf^mater;ialsr,,.^ayailability and cost  of

 manpower , _. logistics, ,,_,s chedulinq ,. ,and , legal requirements .

 Further,  the.. choice ,of .quality .control ..methods  will be  in-

 fluenced  by  final application  of the data  generated and by
                     '• ' •   • /     ..-••'             . •   • .
 the results  of the continuing  second step  in quality control:

 analysis  of  the total measurement system.



                            2-3;.   .

-------
2.2  Analysis of the Total Measurement System



     The, effectiveness of the types of physical controls



just discussed -- calibration, maintenance, standardization,



and the like — can be determined by a number of statistical



and other techniques, including analysis of data and review



of laboratory records and reports.   Procedures for this type



of detailed analysis are amenable to a systematic approach,



we are speaking not about occasional, random spot-checks of



laboratory data and staff performance, but about a continuous,



orderly process of administrative analysis.  Techniques of



such analysis are discussed in detail and exemplified in later



chapters concerning field and laboratory operations.  At this



point we consider only a few fundamentals.



     2.2.1  Precision and Accuracy



     The terms "precision" and "accuracy" denote specific,



measurable characteristics of laboratory analysis.  They



are key concepts of quality control, defined as follows:



     0 Precision is a measure of the reproducibility of



       results.   It is determined by replicate analyses,



       and it represents the variability of results among



       those replicate analyses.  Precision can be expressed



       as standard deviation, variance, or range.
                         2-4

-------
     0  Accuracy is the difference between a measurement


                              2
        and an accepted value.   Accuracy represents the



        magnitude of error in a measurement.  It is ex-



        pressed either as relative error in percent or



        in concentration terms such as parts per million.



        Accuracy is determined by comparing analytical re-



        sults from analysis of unknown samples to results



        from analysis of reference materials.



     Most of the critical parameters in an analytical system



should be evaluated in terms of precision and accuracy.  If



they are not obvious, these critical parameters can be



identified through sensitivity testing.



     2.2.2  Control Chart Techniques



     Several techniques are available for plotting control



charts of accuracy and precision.  Choice of a technique



for computing precision depends primarily on the change in



reproducibility as a function of change in the parameter



being measured.  Chapter 5 represents a method for determin-



ing the type of precision control chart to be used in a



given situation.



     Choice of a technique for expressing and for plotting



accuracy data on a control chart depends on the nature of



the sample, interferences, and the sensitivity range of



the method.  The techniques used to determine accuracy are:
                         2-5

-------
        Analysis of primary or working standards.
        Spiked samples.
        Percent recovery calculations.
     The major difference between precision control charts

and accuracy control charts is that precision charts show

variability between sets of measured values, whereas

accuracy charts show variability between measured values

and known values.

     To enhance the reader's understanding of the statistical

terms that are used in discussing accuracy and precision in

later chapters, we provide some important definitions:'  '  '  '

     0  Measures of Central Tendency - Parameters such as

the arithmetic mean, geometric mean, median, mode, etc.

which are used to describe the point about which the data

tend to cluster.

     0  Arithmetic Mean - The most commonly used measure of

.central tendency is the sum of the values of the observations

divided by the number of observations.


                         1  N
     Population Mean y = rr  £   X,  i = l,2,3,...,N
                         N i = l   L

     where N = Number of observations in total population

           X.= Observed values
            i •

     0 . Frequency distribution.- Grouping observed values

into specific categories.

     0  Normal distribution - The bell-shaped probability

distribution which is determined by two parameters, i.e. the

mean and the standard deviation.

                         2-6

-------
     0  Variance - A measure of the variation of individual



observations about the mean.  The variance of the popula-


                          2      2
tion and of a sample are a  and S  respectively.



      91              ?
     a  = ±  Z   (Xi - y)    i = 1,2,3,...,N

            i = l

            2
     where a   = variance of population


           X.  = observed values



           y   = population mean



           N   = number of observations in total population



      2   1   N         - 2
     S^ = ±   Z    (X. - xr  i = 1,2,3,... ,N

             i=l

            2
     Where S   = variance of sample


           X.  = observed values


           X   = sample mean



           N   = number of observations in the sample



     0  Standard Deviation - A measure of  the variation of


individual observations about the mean.  The unit of measure-



ment for the standard deviation is the same as that for the



individual observations.  The standard deviation  (equal to


the square root of the variance) is referred to as a and S



for the population and a sample respectively.


     0  Random Error - Repeated analyses to determine the



concentration of a contaminant in a sample will usually re-


sult in different  values.  These values, which tend to


cluster about the  true value,  are caused by indeterminate
                          2-7

-------
errors.  The distribution of random error is generally

assumed to be normal with a mean equal to the true value

and a variance of a .

     0  Bias - The result of a determinate  (but possible

unknown) source of error which causes the result of an

analysis to be above or below the true value.  Typical

sources of bias include; improper calibration, human error

in reading a meter or a color change.

     0  Range - The difference between the maximum and

minimum values for a sample of observed values.  When the

number of observed values is small, the range is a

relatively sensitive measure of general variability.  As

the number of observations increases the efficiency of

the range (as an estimator of the standard deviation)

decreases rapidly.

     0  Coefficient of Variation - The ratio of the

standard deviation to the mean, also referred to as the

relative standard deviation.  It is usually expressed as

a percentage and is given by


     CV = §  (100) %
          X

     where S  = standard deviation of a sample

           X  = mean of a sample

     0  Confidence Interval - A statistic (e.g. the mean X)

is computed from the data for a sample.  The statistic is
                         2-8

-------
then used as a point estimate of the population para-

meter (.e.g. the mean u) .  It is recognized that the

statistic computed from a second sample would not be

identically .equal to that for the first sample.  Because

of this points A and B are determined such that it can be

said with a specified probability that the true value of

the population parameter lies within the interval de-

scribed by A ahd'B.

     For example the probability statement for the 95%

confidence interval estimate of the population mean  is

given by;

     P   (X - t _,S  <  y < X + tn_1S) = 0.95

             •\fn~~             -y[rT~

     where X  = sample mean

           S  = sample standard deviation

         t _,  = Student "t" value for n-1 degrees
         n      of freedom

           n  = number of observations in the  sample.

     The probabilities usually  associated with confidence

interval estimates are 90%, 95%  and 99%.  For a  given

sample size the width  of the confidence  interval  increases

as the probability increases.

     0   Confidence Limits - The end points of  the  confi-

dence interval A  and B as discussed above, where:
                          2-9

-------
     A  - X - t  ,S
               n-1
     B  = X + t  , S
               n-1
     0  "t" Distribution - A probability distribution de-



veloped by W. S. Cosset (writing under the pseudonym



"Student") used in the computation of confidence interval



estimates when a (the population standard deviation) is



unknown.  In such a case S (the sample standard deviation)



is used as an estimate of a.   When the sample size is



small the value of "t" for a given probability level



differs significantly from the "2" value for the normal



distribution.  For example in determining the 95% con-



fidence interval estimate of the mean when the sample



size was 10, the value of t is 2.262 whereas the value of



3 from the normal distribution is 1.96  (regardless of



sample size) .



     0  Adjustment Factors - Adjustment factors as applied



in the manual are multipliers used to calculate statistical



control limits for control charts .  They provide a method



of approximating the distribution of all the values in the



universe when calculating statistical limits.  This is
                                   i'

necessary because the distribution of sample values differs



from the distribution of universe values.  The factors



used in this manual are D.,, D  , B_, B , and A,,.  Definitions
                         2-10

-------
of these -factors and formulae for computing them are in



Appendix A.  Tables with the factors used for the 99%



confidence interval are in Appendix B.  The application



of each of the factors is:



     0  D_ -  Compute the lower control limit for a



              range control chart.



     0  D. -  Compute the upper control limit for a



              range control chart.  In some cases,



              D. can be used for the lower control



              limit also  (see Section 5.3.6).



     0  B- -  Compute the lower control limit for



              standard deviation or coefficient of



              variation control charts.



     0  B. -  Compute the upper control limit for



              standard deviation or coefficient of



              variation control charts.



     0  A_ -  Compute upper and lower control limits



              for  average control charts.



     2.2.3  Review and Control Procedures



     Non-statistical review procedures are also required



for determining validity of data.  These include procedures



for evaluating personnel performance, instrument performance,



and quality of reagents and materials.



     Procedures for checking the  performance of laboratory



analysts  include round-robin tests, both intra-laboratory
                          2-11

-------
(between individual analysts or groups) and inter-



laboratory (between members of different organizations).




Inter-laboratory proficiency testing can be applied to



both field sampling and laboratory analysis.  For deter-



mining trends in individual performance over a long



period of time, control charts showing the accuracy or



precision of the results obtained by a single analyst



using a specific analytical method can be useful.



     Procedures for review and control of instrument per-



formance include review of maintenance logs and function



check data and the establishing of control limits for



calibration curves.  Procedures are also required for



checking automatic recording and transmission equipment.



     Review and control procedures for checking quality



of reagents and materials include scheduled batch check-




ing and restandardization.
                         2-12

-------
3.0  QUALITY CONTROL IN ATMOSPHERIC SAMPLING

     Attaining the highest possible quality of data from

an atmospheric monitoring program requires good operating

procedures and methods of determining the level of quality

achieved.  Quality control procedures for atmospheric

monitoring in the field are different from those for

laboratory analyses, because of different operating con-

ditions.  Further, air monitoring does not involve uniform

samples, the concept of sample replication is not easily

applied, and the occasional use of unattended automatic

instruments in remote locations may complicate the appli-

cation of quality control techniques.  In field operations

the major application of statistical techniques is for

determining that the operational.variables of analyzers

and samplers are within acceptable limits.  The most im-

portant control variables for which criteria should be

adopted are:

     0  Sample and reagent flow rate
     0  Span and zero drift
     0  Output noise
     0  Instrument response time          '
     0  Recorder dead band
     0  Sample conditioning
     0  Environmental conditions
     0  Power supply
     0  Response to functional tests

     Standard procedures should be established to control

these parameters.  Also, output data may  be checked for

extreme values, and the total performance of an analyzer
                          3-1

-------
may be measured in terms of its output of good data as a




proportion of total in-place operating time.



     This chapter discusses quality control procedures



common to all phases of air quality monitoring, including



monitoring systems and integrated or static monitoring




techniques.



3.1  Controlling the Physical Parameters



     3.1.1  Environment



     Air sampling is conducted under a variety of environ-



mental conditions: monitoring sites range from fixed, ground-



level, enclosed (interior)  stations controlled for tempera-



ture and relative humidity and having adequate, well-




regulated power supplies to outdoor stations located at



various altitudes and exposed to widely ranging tempera-



tures, possibly mobile platforms mounted in aircraft or



vans.  These environmental factors can affect the reliability



of the sampling device and the quality of the data obtained.



Table 3.1 lists some of the more important environmental in-




fluences cited by the International Electrotechnical



Commission  and some possible effects of variation.  In



atmospheric sampling, each sampler and instrument used



should be examined for these and other possible environmental



effects with particular attention to those mentioned in



methods descriptions and listed by manufacturers.  Every



attempt should be made to adhere to absolute limitations
                         3-2

-------
                        TABLE 3.1

          ENVIRONMENTAL INFLUENCES AND EFFECTS
Influencing Factors
         Effect
Ambient Temperature
Sample and reagent flow
rates, life of electronic
components, amplifier
drift, motor performance,
ink drying, reagent
evaporation
Barometric Pressure
Mass flow rate of air
sample, requires correc-
tion to m.s.l. pressure
in reporting mass concen-
trations of contaminants
Vibration and Shock
Alignment of optical
components, joint leakage,
microphony  in NDIR analy-
zers, damage to fragile
components, ..recorder pen
movement
Electric. Fields
!Interference in electronic
signal processing
Voltage
Performance of voltage
sensitive  electronic
components, motor  speed
Frequency
 Speed  of  synchronous
 motors (chart  speed for
 example), performance  of
 frequency-dependent ampli-
 fiers
                      3-3

-------
and, where necessary, to make operating adjustments and



calculation corrections that will account for non-



controllable variations.  Examples of such practices in-



clude the calibration of rotameters with regard to tempera-



ture (particularly those used to metier reagent flow) and



the correction of contaminant concentrations reported in




mass units for pressure variation.



     Good housekeeping is as important at air monitoring




sites as it is in laboratory operations in that it provides



the proper setting for an effective quality control program.



Although some of the effects of poor housekeeping may seem



indirect or may seem more related to occupational safety




and health than to system performance, they usually entail



poor maintenance and so a reduction in the quality of



data.  Some elements of poor housekeeping practices and



possible adverse effects are given in Table 3.2.



     3.1.2  Reagents and Supplies



     Most of the relevant information on quality control as



related to reagents and supplies is presented in Chapter 5.



Several important points relating to air monitoring are



emphasized here.  These include provision for proper trans-



port of reagents in the field and periodic checking of re-



agents that are held in instrument reservoirs or that are



regenerated in situ.   Some important reagent checks are pH



tests, visual clarity, and efficiency of dye or color re-
                         3-4

-------
                          TABLE 3.2

             EFFECTS  OF. HOUSEKEEPING  PRACTICES:
                   ON SYSTEM  PERFORMANCE
     Element
     Possible Effects
Excess atmospheric or
accumulated dust
 Failure of electrical
contacts and switches,
excessive wear of mechani-
cal components, excessive
soiling of optical components
Reagent spillage or
leaks
Corrosion, hazardous
vapors, electrical
hazards, insecure
footing
Improper maintenance
of air conditioning
equipment
Air conditioning failure,
operation outside of
designated limits, equip-
ment damage, freezing,
inking pen failures,
excessive reagent evapo-
ration
Improper use of
extension cords or
overloading of
circuits
Poor voltage control,
excessive circuit failures,
electrical hazard
Improper cleaning
of glassware and
reagent containers
Reagent contamination
Non-systematized storage
of parts and tools
Loss of tools, absence
of tools and parts when
required, subsequent
system failure
                        3-5

-------
moval by carbon columns.  Procedures for checking reagents

and schedules for their replacement or regeneration should

be documented and followed.

     Filter media can be classified as reagents.  Random

sampling schemes for checking new lots of filter media can

be established to measure flow characteristics, surface uni-

formity, presence of pinhold leaks, pH, ion blanks, and light

reflectance or transmittance characteristics.  Acceptance

criteria should be related to the use of the media.  An

example of an acceptance sampling scheme is presented in

Chapter 6 for data checking.  The same kind of sampling de-

sign can be applied to media.  As a practical consideration

such tests as checking for pinhold leaks in glass fiber

filters should be performed in the laboratory on every filter

to be used in the field.  Acceptance testing becomes practical

only when the number of acceptance criteria is large enough

to make testing of each item costly or time consuming.

     3.1.3  Calibration Materials           '
                                 •,   i
     Calibration materials, primarily gases, are introduced

into air monitoring devices in known quantities so that the

analyzer signal output can be related to the concentration

of the contaminant in the sample stream.  Calibration gases

may be obtained commercially or generated on site.  In

either case the major concern in quality control is the

degree of accuracy obtainable with the calibration.
                         3-6

-------
Secondary considerations are stability of output, effect of

storage, effect of operating conditions, and verification

of standards.

     Commercial calibration gases are usually obtained in

pressurized cylinders.  They are most often used "as delivered",

or they may be diluted on site.  Because statements concern-

ing preparation tolerance and analysis accuracy vary among

suppliers, it is useful to define the term "calibration gas"

and to use the definition in procurement and validation.  One

useful statement follows:

          "For purposes of spanning operations, these
          gases represent conventionally true values
          against which indicated values are compared.
          Therefore,  the calibration gases should be
          traceable to standards agreed on by the user
          and the supplier or to national standards,
          and the uncertainty of the conventionally
          true values should be stated."

New certified cylinders of calibration gases should be

obtained before depletion of existing cylinders so that

cross-checks can be made.  If results indicate uncertainty,

additional interorganization cross-checks or comparison

testing may be required.

     Experience has shown that true "zero" gases can be as

difficult to obtain as true "calibration" gases, and their

use requires the same precautions.  Increasingly, calibra-

tion gases are being generated on site.  One rapidly develop-

ing technique is the .use of permeation or diffusion sources.
                         3-7

-------
This technique usually involves the permeation of a pure




gas through a semi-permeable material : (such as Teflon) from



a liquefied quantity of the gas encapsulated under its own




vapor pressure in a container having' at least one surface



of the semi-permeable material.  The pure gas is diluted by



a carrier gas flowing at a known and precisely controlled




rate.  The rate of permeation is temperature-dependent and



must be known.  One advantage of the technique is that the



rate of permeation can be verified by determining the weight



loss of the permeation source.



     Ozone must be generated on-site, by passing oxygen or



air over an ultraviolet source.  At,the present state of



development, a referee analysis must be performed.



     Regardless of the type of calibration material,  an




effective quality control program requires accuracy levels



with these materials that are consistent with the method of



analysis.  Methods specified by the Environmental Protection



Agency and most other published methods will include  state-




ments of the accuracy and replicability to be expected.



Obviously, the accuracy of the calibration materials  must be



greater than the overall accuracy expected.  The stated



analysis accuracy of calibration sources should be +^2% of the



true value.



     3.1.4  Calibration and Maintenance Procedures



     Calibration - Specific calibration procedures are
                         3-8

-------
                                                    789
described in both official and unofficial documents. '  '

Procedures specified for the method or analyzer in use  should

be followed.  The frequency with which calibrations are per-

formed will affect the confidence limits of the data, depend-

ing on the degree to which calibration curves change with

time.  Initially the maximum time intervals between cali-

brations may be set by determining what regulatory require-

ments apply, as given, for example, in the surveillance

portion of an implementation plan.  This minimal schedule

will, of course, be modified by resource and logistical con-

straints such as total available manpower, manpower time re-

quired for each calibration, station siting  (e.g. distance

from central laboratory), and required laboratory assistance.

     Each technical services organization should develop a

history of the replicability of calibration runs and of the

expected change in the calibration curve with time.  On the

basis of such records, operators can determine the accept-

ability of a single calibration and perhaps modify the cali-

bration schedule.

     Generally, calibration records should include space

for the following information:

     0  Instrument identification  (serial or other
        identification number)
     0  Date
     0  Operator identification
     0  Calibration technique
        Description and identification of standard
        material used
                         3-9
o

-------
     0  Test or other code - for use in identifying
        samples for referee analysis
     0  Operator comments
     0  Data

     Example - In calibration of a nondispersive infrared

carbon monoxide analyzer, the following procedure might be

used:

     0  Conduct functional checks described in the operator's

manual to be certain the analyzer performs within specifi-

cations .

     0  Obtain five cylinders of carbon monoxide zero and

calibration gas covering the range of 0-90% of full scale

and including the zero gas and the 90% of full concentration.

The stated concentrations should be known within +2% of the

true concentration.

     0  Span and zero the analyzer using the zero gas and

the calibration gas closest to 60% of full scale concentration,

Adjust the zero and span controls so that the recorder values

correspond to the specified cylinder values.

     0  Introduce the other calibration gases sequentially,

allowing the analyzer to reach a stable value before changing

cylinder sources.

     0  Plot the recorder scale values against the stated

values.

     0  Repeat steps 3 through 5.

     0  Inspect the calibration plots.  If both plots indicate
                         3-10

-------
that any one span gas deviates from an otherwise smooth




curve that could be drawn between the remaining points , re-



check that calibration gas.  If a smooth curve can be drawn



through at least one plot, proceed to analyze the difference



between duplicates.



     0  The acceptability of the calibration precision can



be determined with a control chart for Range.  For this in-



strument the control limits are R+3o/^jn~, where R = average




value of the Range (in the case of duplicates this is equal



to the difference between the two values).  Charts have been



tabulated to give the estimate of a in terms of R.    The



upper and lower control limits are D.R and D.,R respectively.



D3 and D. for duplicates are found in these tables to be



D  = 3.267 and D., = 0.  The central line is R.  After the con-



trol chart has been prepared the values of the Range are



plotted by successive sub-groups  (each set of duplicates).



If the values for R fall within the control limits the



replicability or precision would be judged to be within con-



trol.  As technical personnel gain experience with a par-



ticular instrument, they may desire to establish new limits



for the calibration curve.  The decision will be based on



results of periodic recalculation of the standard deviation



of the Range.  An example of how control limits for a  cali-



bration curve are calculated is shown in Appendix C.



     0  Record all data pertinent to the calibration.  Take
                         3-11

-------
action to implement the use of a new calibration curve if




changed.



     0  Introduce calibration gas to be used in routine span



operation and note scale value.  This value will be necessary



for proper setting of the span control in future routine




operations until a new calibration is performed.



     Other types of calibration may be performed such as in



those other dynamic systems utilizing on-site generation of




calibration gases through use of permeation tubes,  ultra-



violet generation of ozone, and other physical or chemical



processes.  Still other forms of calibration include those




which are designed to calibrate the detector portion of the




analyzer only.  These are sometimes referred to as  "static"



or "reagent" calibration techniques.  Regardless of the



form of calibration, a history of calibration data should be



developed such as described in the preceding paragraphs.



The data collected on replicate sampling during calibration



can be used to determine whether or not the analyzer or



sampler is performing according to expected or required



specifications and to determine the frequency with which



calibrations must be performed in order to keep within per-



formance specifications of the method.



     Maintenance - From the standpoint of quality control,




proper maintenance of air monitoring equipment should improve



the quality of data and increase the recovery rate of valid
                         3-12

-------
data.  Servicing and maintenance schedules should relate to

the purpose of the monitoring, the environmental influences,

the physical location of analyzers, and the level of operator

skills.  Data provided by instrument manufacturers and the

EPA "Field Operations Guide for Automatic Air Monitoring

Equipment"  are useful in developing an initial maintenance
                                                    ••i
program.  Most such manuals present instructions for general

service and for service oriented to specific instruments.

They further indicate the recommended time intervals for

various types of service, such as tasks to be. performed

daily, semi-weekly, bi-weekly, monthly, quarterly, and semi-

annually.

     Example.  Schedule for Daily Servicing - General

     0  Upon arrival at the monitoring site observe all
        recorders for indication of normal operation.
     0  Check liquid traps in instruments and probes.
     0  Check for broken sample lines; check, condition
        of probe or sample line filters.
     0  Check all connections for possible leaks.
     0  Check for chart supply and replace charts where
        needed.
     0  Check for timing of all charts, timers and clocks.
     0  Check all flow-rate indicators for proper flow.
        Check all rotameters for dirt and water, particu-
        larly adjacent to ball.  Clean and dry if necessary.
        Check sample conditioning apparatus.
        Service inking pens where required.
        Check for reagent supply on all wet chemical
        analyzers.
     0  Check level of reagent in all lines, reservoirs,
        etc. on wet chemical analyzers.  Check for evidence
        of carryover.
     0  Check gas cylinder pressure.
     0  Check for noisy or leaking pumps.  Lubricate where
        required.
     0  Check recorder dead zone on all recorders.  Note
        where checked.
                         3-13

-------
     0  Check sampling schedule for static or intermittent
        sampling devices and operate as required.
     0  Record .all maintenance and adjustments on log
        books or forms as specified.
     0  Record jpar.ts and supplies used on reorder form.

     Service and maintenance should be performed by personnel

according to skills and staffing pattern.  In general, station

operators perform routine servicing and trouble shooting

tasks; they should not attempt repairs for which they lack

the proper training or equipment, or for which the time re-

quired would interfere with other scheduled operations.

     3.1.5  Design and Maintenance of Probes and Manifolds

     The sampling of gaseous or particulate air contaminants

range from exposure of a static  (passive) device  (such as

a lead peroxide candle) through sampling from individual

probes and shelters, to sampling with common probes and

manifolds using an auxiliary air-moving device.  Various

probe designs in common use have been summarized and recommen-

dations given to assure that the air sample reaching the

analyzer or sampler is altered minimally.  After an

adequately designed sampling probe and/or manifold has been

selected and installed, the following steps will help in

maintaining constant sampling conditions:

     0  Conduct leak test - Seal all ports and pump down

        to approximately 0.5 inch water gauge vacuum as

        indicated by a vacuum gauge or manometer connected

        to one port.  Isolate the system.  The vacuum

        measurement should show no change at the end of a

        15-minute period.


                         3-14

-------
     0  Establish cleaning techniques  and schedules  - A




        large-diameter manifold may be cleaned by pulling



        through it a cloth on a string.   Otherwise the



        manifold must be disassembled  periodically and



        cleaned with soap and water.  Visible dirt should



        not be allowed to accumulate.



     0  Plug the ports on the manifold when sampling lines



        are detached.  This will help  to maintain the de-



        sired mass/velocity ratio.



     0  Maintain a flow rate in the manifold at 3 to 5 times



        the total sampling requirements or at a rate equal



        to the total sampling requirement + 5 ft /min.  This



        will help to reduce sample  residence time in the



        manifold and insure adequate gas flow to the



        monitoring instruments.



     0  Vacuum in the manifold should  not exceed 0.25



        inch water gauge.  Keeping  the vacuum low will help



        to prevent the development  of  leaks.



3.2  Analyzing and Controlling the  Total Measurement System



     Having applied all practical measures for control of



the physical components of a sampling/analysis operation,



the supervisor or manager proceeds  with thorough analysis



of the total measurement system in  terms of quality con-



trol.  He does this by applying several analytical



techniques, which are described and .illustrated throughout
                         3-15

-------
this manual.  In this chapter the emphasis is on atmospheric




sampling; the techniques presented here, however, such as




the functional analysis described next, usually can be



applied in other phases of technical services operations.



     3.2.1  Functional Analysis



     In application of quality control measures, the total



measurement system can be viewed as a complex consisting of




the analytical method, the instrument or analyzer, and the




operator.  The critical components of this complex are



identified by functional analysis.  We exemplify this




technique as it applies to continuous monitoring, because



continuous monitoring instruments do not provide a discrete



sample with which to perform conventional statistical pro-



cedures, such as use of replicate analyses, spiked samples,



and control samples.



     A useful first step in functional analysis is to pre-



pare a schematic diagram of the analyzer, identifying all



major components and controls.  Briefly summarize the



theoretical principles of the measurement.and indicate all



algorithms and transfer functions relating the measured



quantity to the final data statement of air quality, such



as pollutant concentration.  With this information one can



identify the primary variables directly affecting instrument



performance.  Some of these variables include:
                         3-16

-------
     0  Source output (active devices)
     0  Detector sensitivity
     0  Contactor efficiency (where reagents are used)
     0  Sample air flow rate
     0  Reagent flow rate
     0  Cell pressure (nohdispersive infrared analyzers)
        Optical path length and alignment (colorimeters
        and spectrophotometers)
        Amplifier gain and stability
        Reagent quality
o
     Next, extend the analysis to determine means for control-

ling these critical parameters and for detecting off-specifi-

cation performance.  Some parameters (such as flow rates and

pressures) can be observed directly, but often one must

monitor performance by observing secondary parameters that

may indicate malfunction or off-specification performance.

Such secondary parameters include:

     0  Signal output noise
     0  Span or zero drift
     0  Repetitive transient signals >
     0  Time response to step input .of .contaminant
     0  Atypical appearance of charts
     0  Response to built-in function tests
     0  Physical appearance of reagent lines, reagent,
        and other components

     Of these observable indicators of performance, the ones

most adaptable to statistical monitoring are noise, span

drift, and zero drift.

     In contrast to continuous monitoring, batch or sequential

monitoring (accomplished by collection of filter samples,

use of impingers, or similar procedures) is much more

closely related to laboratory procedures.  Batch monitoring

operations, like those in the laboratory, can be evaluated
                          3-17

-------
by use of replicate sampling, standard or spiked samples,




and cross-checking among samplers or operators.  Beyond



these, however, a thorough functional analysis of the



sampling arid analysis procedures should be performed to



identify critical operating parameters and sources of



error, and to define what data histories are needed, how



they shall be collected, and procedures for manipulating



and interpreting the data.



     An initial procedure in this functional analysis of



batch monitoring might be to construct a flow chart of the



sampling and analysis processes, identifying all performance



criteria and control and measurement points.  Calculations



used in obtaining final results would be shown, as would



information on expected repeatability and accuracy.



Figure 3.1 illustrates a 'flow chart constructed for the



determination of suspended particulate matter by the high-



volume filter sampler method.  The illustration indicates



that flow rate, time of exposure, sample conditioning, and



initial and final weights represent the critical parameters



in the hi-vol method.  Opportunities for cross-checking in



the weighing operations also arise, and parallel sampling



can be conducted.



     3.2.2  Sensitivity Analysis



     Following the functional analysis of the measure-



ment system the technique of sensitivity analysis may be
                         3-18

-------
                  Obtain
                   and
                  Inspect
                  Filter
                Equilibrate
                  Filter
K     Record
      Time
      Flow
                    I
                          /Specified \
                          \Conditionsr
                   Weigh
                   Filter
           Load Filter
              and
             Start
            Sampler
                             Balance \
                             Spec's. I
                 SamplerX
                   and
                Operatingy
                 Spec's.
  A
    \
Record
 Time
 Flow
Terminate
Sampling
  and
Unload
                Equilibrate
                   Filter
                       ^	/Specified
                           ^Condition!
     Record
    ^Weight
Weigh
Filter
fcfc-
p
                             To Other
                             Analyses
                               or
                              Store
Calculate
and
Document
tot
IP
Report
Results
Figure 3.1
       Functional analysis of high volume
       suspended particulate sampling.
                     3-19

-------
used to determine the effect of variations in either primary

measurement control variables or in secondary influences or

interferences.  Sensitivity is defined as the partial de-

rivative of a system output with respect to input.   In a

complex system a more practical definition of sensitivity,

often employed for analytical purposes,  is the incremental

change in output resulting from an incremental change in

input.

     If the system can be described in terms of a mathematical

model, an analytical approach may be taken usually utilizing

a factorial design.  In the control of air monitoring systems

an empirical approach to sensitivity testing is more likely.

     Example

     Excessive zero drift has been, detected in a continuous

air monitoring instrument, but no component failure is indi-

cated.  Knowing the analyzer, the operator suspects that

temperature or voltage variations may be influencing the

instrument behavior.  The operator wants to know how the

voltage and temperature interact.  Because the number of

variables is small he could select a full factorial design

of an experiment designed to test each variable at three

levels -  (1) within specified normal range,  (2) above

normal range, and  (3) below normal range.  The total number

of experiments required would be p  where

             n = no, of factors = 2
             p - no. of levels = 3


                         3-20

-------
    2
or 3  = 9.  The experiments can be identified as T..V,,  T^V^



T1V3' T2V1' T2V2' T2V3' T3V1' T3V2' T3V3 where T and V  refer


to temperature and voltage respectively, and the subscripts


refer to the level.  Other tests, such as Youden ruggedness


testing, could be used if the number of factors and/or


levels is large.  Fewer tests would be required than with the


above technique, but the results would be nearly the same.


     If an analysis of the data shows a relationship between


one or more of the influence parameters and the dependent


variable, in this case zero drift, a case would be made for


better control of that parameter.


     3.2.3  Compiling and Using Data Histories


     In a quality control program, positive steps should


be taken to provide methods, equipment, facilities, staff,


and operational procedures that make possible the production


of high quality data.  Means of measuring performance and


detecting promptly any deviation from acceptable performance


should be adopted. . Corrective actions in the event that


data quality falls outside acceptable limits should be


defined.  Adequate data histories must be available with


which to verify statements concerning the quality of the•


data.


     In addition to the calibration data described earlier


data histories should be collected on all of the critical


measures of performance identified in the functional analysis,
                         3-21

-------
Data histories,most commonly kept are those for sample air



and reagent flow rates, results of span and zero checks,




results of use of other standards or function tests, noise,



response time,;and recorder dead zone.  For example, if the



specification for zero drift for a continuous analyzer is




that it not exceed 1% of full scale/24 hours, and if the



data acceptance criteria are set to reject all data recorded



during a period in which the total drift exceeds 2% of



scale, then an initial schedule might be set for zero check-




ing once every two days.  Experience may require adjusting



the checking interval.



     Use of data histories in maintaining quality control is



based on the recognition that no analytical method or instru-



ment yields perfect results; when repeated determinations



are made on the same sample, some variation in results is



inevitable.  Even if all identifiable o'r assignable causes



of variation are removed, some indeterminate sources remain.



These indeterminate errors should be randomly distributed.



Since the behavior of these random events can be predicted



statistically, limits within which repeated measurements



should fall can be computed.  The design and use of control



charts and other uses of data histories rely on the com-



putation of ,these 'control limits'.  It should be understood



that even though application of statistical techniques



indicates that an analysis .or air monitoring process is in
                          3-22

-------
a state of control, this does not mean that every analysis




or element of data is acceptable.  Nor does it mean that the




control limits are rigid; they can be adjusted to produce



higher quality data.



     As discussed earlier, replicate analysis cannot easily



be applied to automatic continuous analyzers, and quality con-



trol of the data they produce must usually be based on main-



taining performance characteristics within limits.  One



important indication of low-quality data from a continuous



analyzer is excessive zero or span drift.  Drift affects the



quality of data because the calibration curve has been



altered without the data reduction process having been



automatically altered at the same time.  Within pre-set



limits it is possible to interpolate between the expected



and the altered calibration curve if the altered curve can



be identified.  This is done by field operations known as



zeroing and spanning, during which a known zero gas and a



single calibration gas are introduced to the analyzer.  The



scale value is noted and compared with the value expected



on the basis of the most recent full calibration.  In effect



two calibration curves are available, one representing the



start of the time period immediately following the previous



standardization and one for the time of the current standardi-



zation.  If the change is not too great, a linear interpola-



tion between the two curves is performed for data points
                         3-23

-------
between the two standardization operations.  The difference

between the succeeding span and zero readings is called span

and zero drift and is expressed in terms of concentration

per unit time, e.g. yg/m  per 24 hours.

     One method of monitoring the state of control for span

drift would be to construct a control chart for the mean

span drift over some reasonable time period, perhaps two

weeks.  The center line for such a chart is drawn at the

expected mean drift, y, and the upper and lower limits will

be y + 3 o/^n.  If the distribution of the span drift is

normal, these limits should contain approximately 997 values

out of every 1,000.  If the subgroup span drift means are

expressed as x, the best estimate of y is the overall

sample mean, x, which may be computed from the subgroup

means.  If we have less than ten span operations during the

two-week period representing the subgroup, then R/d9 is an

unbiased estimate of a.  R is mean of the subgroup ranges.

The factor A,, = 3/pr- d,., has been compiled and listed in

standard texts on statistics.  Appendix A defines these

factors.  Table 3.3 lists the span drift data for an

analyzer over 26 two-week periods.  Since span checks were

made every other day, seven checks were made every two-

week period.  Therefore

             x = E x = 35.0 = 1.35
                  n      26
                          3-24

-------
                        TABLE  3.3

    SPAN DRIFT FOR ANALYZER A FOR 26 TWO-WEEK PERIODS
    Period
      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
Mean Sp_an
Drift (x)

   1.4
   1.5
   1.3
   1.4
   1.2
   1.3
   1.2
   1.4
   1.3
   1.2
   1.8
   1.3
   1.4
   1.2
   1.5
   1.2
   1.1
   1.3
   1.2
   1.4
   1.6
   1.4
   1.5
   1.3
   1.2
   1.4
Range (R)

  0.8
  0.9
  0.6
  0.8
  0.9
  0.6
  1.1
  1.3
  0.4
  0.3
  0.6
  0.9
  1.1
  0.4
  0.7
  0.8
  0.8
  0.7
  0.2
  0.5
  0.7
  0.9
  0.8
  0.9
  1.2
  0.4
Mean span drift (x) = average span drift for a
                      two-week period.

Range =  Max-min values for a two-week period.
                      3-25

-------
             R = £ R  = 19.3  = 0.74
                  n      26



            A2 = 0.419

            A7 is based on n = 7,  the number of

            checks per two-week period.

            A2R  = 0.419 (0.74)  -  0.31

            x + A2R = 1.35 + 0.31  = 1.66

            x - A2R = 1.35 - 0.31  = 1.04

The information developed from Table 3.3 is plotted on the

control chart shown in Figure 3.2   It is obvious that,

except for only one two-week period, the analyzer operated

satisfactorily.

     Control charts for duplicate  testing, such as might be

developed for use with batch or intermittant monitoring,

are similar to those described in  detail in Chapter 5 on

laboratory methods.  Parallel sampling, such as might be

done with high-volume particulate  samplers and bubblers,

can be analyzed by this technique.
                                   i
     A special control chart may be/ used for duplicates
                                     i
obtained in comparison testing.  If it is assumed that this

type of testing will be performed  only at intervals, it may

be difficult to obtain an unbiased estimate of the true

expected difference, d, between duplicates on a control

chart on which the central line is 0.  Existence of bias can

be determined from several indicators by using the method of

extreme runs:


                           3-26

-------
1.8
1.7
1.6
1.5
1.4
... 1 ^S
IX -L.-JJ
, 1.3
PT-
U) 
-------
     0  8 successive points are on the same side
        of the central line
     0  10 out of 11 are on the same side
     0  12 out of 14 are on the same side
     0  14 out of 17 are on the same side
     0  16 out of 20 are on the same side

     This type of chart is used, for example, in analyzing

data collected by a mobile unit that regularly checks the

air monitoring conducted at fixed sites.  In this type of

operation it is imperative that the mobile unit and the

stationary instruments are sampling the same air mass.

3.3  Documentation and Demonstration of Quality Control

     3.3.1  Records

     Administration of a quality control program requires

adequate records needed to validate data prior to process-

ing, to provide data histories, to aid in modification of

service procedures and schedules, to develop records of

parts usage, and to guide trouble shooting and repair.  The

general categories of these records are:

     0  Instrument service records - These may be in the

form of lab books or prepared multiple copy forms.  The

California Air Resources Board has determined that the

latter type better serves management functions.  These

forms provide space for entry of routine data such as flow

rates, filter or span checks, reagent checks  (such as pH

of potassium iodide reagent for total oxidant), and main-

tenance performed.  Similar forms could also include
                         3-28

-------
acceptable performance criteria for convenient reference



by the station operator.



     0  Calibration records -  These would contain all data



applicable to calibration including certified analyses of



cylinder gases; flow rates; results of referee or side



stream analyses; recorder output data, including scale



values; lag time, rise time, fall time; copy of recorder



chart; gravimetric determinations, if any; temperatures;



barometric pressure; and description of calibration pro-



cedures .



     0  Station log books - To include data on all station



changes,  station conditions, dates of supply delivery, sub-



jective statements on air quality conditions, visibility



observations, and notes of unusual events that might influence



analyzer readings.



     0  Data validation and reduction notes - These would in-



clude all entries necessary for chart editing, including



results of span and zero checks, flow rates adjustment,



function tests, down times for service, and notes on known



bad data.  These entries may be made either on the chart or



on special report forms.   They are normally made on the



chart for manual data reduction and on a special form for



automated data reduction.  In the latter case, the notes



form a basis for supplementary computer instructions cover-



ing baseline data, span constants, and data to be excluded.
                         3-29

-------
LOCATION
  C H i'c 0
 MODEL No.
A.R.3. No.
                   OPERATOR
                                                             PAGE
                                                          . 7O,
                                                         100
FDLTCR CHECK
Number of Filter
OX
NOX
NO2
Chart Reading
Sample Energy
Ref. Energy
Chart Reading
Sample Energy
Ref. Energy
Chart Reading
Sample Energy
Ref, Energy
Date: Ol - u\- - 7 0
0
O
So
50
O
55
55
O
4^
46
i
i"5
44
5o
^
4X
^5
^4<
•34
46
2
2.5
40
60
51
26
«#
51
2*
4fe
3
H3
35
5o
Bl
20
£5
•62
(6
4t
4
£7
3i
^o
tlo
13
55
l\5
l (
46
5

•7
55
0^?
^3
^fc
51
^2
•7.0
64
32.
16
^
4
6
7
S'f
f'7
6
'.V,
    SOLUTION
   FLOW CHECK
    Check
Date  Rate   Set
                                       Reset to
Date  Rate   Set
                     Check
                               Date
                     Rate
                          Set
                                   Reset to
Date  Rate
 Oxidant
                                    3-0
 NOX
    /-OO
HO
                                      it
 NO2
Ki Solution
|>M Check
Date
^•*^
pH 1 Date
^.s-p"'2
pH
6,?
Date
?-/9
pH
7'^>
Date
9-Z7
PH
7,/
Date
Y-ZS'
pH
(.3
Dalo

;~-'

  DATE
             WIAINTENANCG PiS
                               fttfo
                                          5 ML
                                                            L. J
                                                     TO
 °('20
                            lAQVZO ZWL  Vou/i<:i»
-------
     In addition to the records mentioned, provision should



be made for rapid notification concerning out-of-limits



analyzer performance as detected by the station operator



or, with telemetry, by the control center supervisor.



     3.3.2  Reports



     Reports prepared as part of a quality control program



are usually summaries designed to inform administrative



personnel as to the quantitative and qualitative level of



performance of the air monitoring activity.  These reports



must be tailored to the size and complexity of the operation.



Several types of reports that might be employed are described



briefly.



     0  Calibration Report - This report should summarize



calibration activities during the report period, including



equipment calibrated, date calibration performed, time re-



quired for calibration, source of standards, techniques used,



problems encountered, and indication of any change in cali-



bration.  Quality control charts are useful here.



     0  Maintenance Summary - This report should describe



significant maintenance, not routine servicing.  It includes



such activities as replacement of major components and



changes of the equipment.  This report will aid in develop-



ing a history of parts used and operations performed, to



serve as input for a preventative maintenance program.



     0  Data Recovery Report - This report describes the
                          3-31

-------
  performance of the air monitoring activity in terms of use-

  ful data recovered.  For each separate analyzer or batch

  sampler one can determine a theoretical maximum amount of

  data that should be available, based on time in place,

  sample schedule, and scheduled downtimes for calibration,

  span and zero checking, and preventative maintenance.   This

  report would list the percent of data recovered for each

  analyzer and sampler, by location, for the report period

  involved.  A monthly or quarterly report should suffice.

       A control chart may also be developed to indicate

  whether instrument or station performance is within expected

  limits on the basis of valid data, expressed as a percent of

  data theoretically available.  This is known as a control
                             10
  chart for percent valid data.  The following parameters

  are calculated:

       p = mean percent valid data (in this case the over-
           all mean determined from past experience on per-
           cent of valid data).

100 -llpq = standard deviation of percent valid data.
     »n

       q = 1-p

       n = number of hours in sample period, e.g. in one
           month.

       For example, if a station is yielding 85% valid data

  out of a possible 100%, the following calculations would

  apply if the station had operated 720 hours during the

  period of interest:
                           3-32

-------
     p = 0.85




     q = 0.15




     n = 720




       W= 0.013






     UCL (upper control limit) = 0.85 + 3 (0.013) =0.88




     LCL (lower control limit) =0.85-3 (0.013) = 0.82




     The control chart in this case is plotted with p as the




central line, and p +_ 3 U *-S.  as the upper and lower control




limit lines.  The percent valid data  (p) for the monthly sub-




group is plotted on the y-axis, and the subgroup number is




plotted on the x-axis.  If n is reasonably large, as for




example the number of hours in a month, the normal distribution




is a good approximation of the binominal distribution.  In




this case the probability is only three in a thousand that




a value of p will fall outside the control limits by chance.
                          3-33

-------
4.0  QUALITY CONTROL IN SOURCE SAMPLING



4.1  General Considerations



     Currently, application of basic quality control elements



to source or emission testing is almost nonexistent.



Although operators are usually careful to clean and calibrate




equipment before testing, the equally or even more important




phases of sampling, such as statistical test design, control



of interferences from unknown compounds, and other steps to



insure the efficiency of sampling procedure are often



ignored.  The primary reasons for this are lack of knowledge



pertaining to emissions from specific sources and lack of



background data on various source sampling procedures.



     In spite of these difficulties, however, the technical



services supervisor who is oriented toward quality control



will work toward developing a set of systematic procedures



that will assure the highest possible quality of source-



sampling data.  He will find that many of the basic con-



cepts of quality control can be applied or adapted to



source sampling.  As an example, consider the instrumental



sampling of emissions.  Although most emission testing is



currently accomplished by manual techniques, instrumental



methods are being developed and in special cases have



proved successful for determining a specific gaseous com-



ponent in a 'clean1 dry gas stream.  For most instrumental



methods, the same decisions and criteria for maintaining
                         4-1

-------
acceptable operation of atmospheric monitoring (see Chapter

3)  will also apply to source sampling.

4.2  Controlling the Physical Parameters

     4.2.1  Interferences

     The goal of a source sampling team should be to collect,

store, and transport to the analytical laboratory a sample

that is as free as possible from interferences.  The inter-

ferences most commonly affecting field samples are related

*.  4-u    *  4-    11,12,13,14
to these factors:

     0  Composition of probes
     0  Composition of collection media and filters
     0  Cleaning procedures
     0  Standardization of reagents
     0  Storage and transport of samples.

     These factors are analyzed more fully in Table 4.1.

                         TABLE 4.1
           ELEMENTS FOR CONTROL OF INTERFERENCES
  Element
          Control Consideration
Probes
Media and
filters
Cleaning
procedures

Standard-
ization of
reagents


Storage and
transport
of samples
Inert to gases sampled.
Trial runs to test inertness in undefined en-
vironment.
Temperature control.

Filter requirements.
Non-reactivity of filters.
Requirements for distilled or deionized water
Blank requirements.
Inertness of cleaning solutions.
Elimination of residues.

Normality.
Stability.
Frequency of exposure.
Storage requirements.

Inertness of containers.
Elimination of atmospheric influences.
Time required for transport and storage.
                         4-2

-------
     4.2.2  Sampling Rate and Sample Volume



     The sampling rate must be known accurately in order



to relate the amount of sample collected to the stack gas



concentration.  Direct displacement or totalizing meters



such as dry or wet test meters are preferred, since they



can accurately measure the gas volume even if the gas flow



rate varies, and they do not require constant attention.



Rate meters require careful measurement of total sampling



time and constant observation to make sure the rate does



not vary because of pressure changes in the sampling train.



Table 4.2 compares the accuracies of various types of flow



meters.



     Evacuated tanks or flasks can also be used to determine



the volume of gas sampled if the initial and final pressure



and temperature in the tank are carefully measured and the



tank volumes are known.



     4.2.3  Equipment Maintenance and Calibration



     The accuracy with which one can measure various emission



parameters depends greatly on the accuracy of the instruments



used in sampling procedures.  Some of the common sampling



components that require maintenance and calibration to



assure maximum accuracy are Pitot tubes, manometers, thermo-



meters, flow meters, and gas meters.



     Rules for calibration of the various instruments used



in source testing are not available in the literature.  In
                         4-3

-------
                             Table  4.2   FLOW METER TYPES  AND ACCURACY
       Type             Fluid
Direct Measurement
 Gas Prover               G
 (inverted bell type)
 Frictionless Piston      G
 Bubble Meter             G
 Mass                     L
Fluid Dynamic Type
 (Roots type)
 Piston Pump
Area Meter
 Rotameter
                          G,L
                                     Principle

                                Batch displacement

                                Batch displacement
                                Batch displacement
                                Mass measurement
                                Application
                          Calibration


                          Calibration
                          Calibration
                          Calibration
                                  Accuracy Range


                                     +0.2%


                                     +0.2%
                                     + 0.2%
                                     + 0.1%
Orifices2
Venturi Meter
Direct Displacement
Wet Test Meter
Dry Gas Meter
Cycloidal
G,L
G,L
G
G
G,L
Head loss
Head loss
Continuou
Continuou
Continuou
                                                           Continuous 6 Intermittent Sampling    +0.5% (max.)
                                                           Continuous & Intermittent Sampling    +_0.5%
                                 Continuous displacement    Calibration
                                 Continuous displacement    Intermittent sampling
                                 Continuous displacement    Calibration

                                 Continuous displacement    Continuous sampling
Head loss
Continuous sampling
 (1) Obtainable with NBS traceability
 (2) Commonly used in field sampling work
                                                                + 0.5%
                                                                +1%
                                                                + 1%


                                                                + 1%


                                                                +1% to 10%

-------
practice, the measuring devices are calibrated against known

quantities or against devices known to provide a higher

degree of accuracy (see Table 4.2), and are then adjusted

to read the correct value.  If the device cannot be adjusted,

it is replaced or used as a spare with an appropriate

correction factor.

     Following are guidelines to the effective maintenance

and calibration of source sampling equipment:

     0  Pitot Tubes - Compare with a standard type pitot
        tube by inserting both Pitot tubes into a duct
        and measuring the velocity at a specified point.
        A correction factor is then calculated as shown
        in Table 4.3

                          TABLE 4.3

       EXAMPLE DETERMINATION OF PITOT TUBE CALIBRATION
Standard
Pitot reading
Ho -\[H7"
0.3 0.5477
0.5 0.7071
1.0 1.000
Type S Pitot reading
HI ^T
0.415 0.642
0.700 0.837
1.44 1.200
Ratio -
0.
0.
0.
C = 0.
P
NH~o
>Pl = Cp
853
844
833
843
H0  = Velocity head  ("H20)

EI  = Velocity head  ("H20)

C   = Pitot tube coefficient


     0  Manometers - Inclined and U-tube manometers give
        direct readings and do not require calibration.
        The manometer must be clean, air-tight, and filled
        with the liquid specified on the scale.  Where
        transducers or gauges are used to measure pressure,
        they should be connected in parallel with a manometer
        and adjusted to read the same value.
                         4-5

-------
     0   Thermometers - Bi-metallic dial-type thermometers
        are commonly used.   These should be checked against
        a mercury-in-glass  thermometer and adjusted to read
        correctly.   All thermometers used in a sampling
        program should be checked prior to use and should be
        adjusted to the following limits:

                      150°  F.    + 2°

                      150-500°  + 5°

                      500°       + 10°

     0   Dry Gas Meters - A dry gas meter is calibrated by
        connecting it in series with a bell-prover, or wet-
        test meter.  Meters should be calibrated before
        every test series and should be adjusted to read
        within 1% of the true value.

     0   Orifices, Rotameters,  etc. - These devices are cali-
        brated by connecting them in series with a more
        accurate volume meter, such as a wet-test meter.
        Calibration should be performed before every 10
        months, depending on frequency of usage.  Calibration
        curves with deviations of no more than 1% should be
        established for each device.

All sampling equipment must be cleaned carefully to prevent

sample  contamination.  Dichromate cleaning solutions are

recommended for cleaning glassware prior to beginning a test

series.  Complete rinsing with tap and then distilled water

is suggested.  For metal analyses, certain types of glass-

ware should not be used and cleaning with nitric acid is

recommended.  The method description should include guide-

lines for choice of glassware.

     4.2.4  Conducting the Emission Test

     Before starting an emission test, the sampling crew

should follow a series of preparatory steps designed to re-
                         4-6

-------
duce or eliminate interferences from various sources.   Check-

lists enable the crew to follow the standardized procedures

consistently.  The operators should maintain complete notes

in the field.  Checklists can include such data as the

following:

     0  Preparation of water or reagent blanks.
     0  Cleaning of impingers,  probes, collection vessels,
        etc.
     °  Confirmation of reagent grades specified in the
        method.
     0  Preweighing of filters.
     0  Checkout of heating or cooling systems for probes.
     0  Leak check of sample train.
     0  Identification or visual check of Pitot tubes,
        meters, thermometers.

     Typical data sheets used in the field are shown in

Figures 4.1 and 4.2.

     After the preparatory measures, the principal quality

control effort is directed toward preventing measurement

errors.  Stack sampling involves a number of physical para-

meters, and the errors of measurement associated with each

parameter combine to produce an error in the calculated

emission rate.

     Measurement errors are of two types: bias and random.

In bias errors, which usually result from poor technique  or

faulty equipment, the measured value tends to differ  from

the true value in one direction.  Errors of this type can

be minimized by proper calibration and adequate training.

Random errors result from a variety of sources that cause

the measured value to deviate in either direction from the
                         4-7

-------
                    GAS SAMPLING FIELD DATA
Material Sampled For

Date
Plant                       Location
Bar. Pressure 	" hg
              o
Ambient Temp.  F	 Stack Temp.,  °F 	

Run No. 	

Power Stat. Setting	

Filter Used: 	 Yes 	 No

Operator 	
                      .,        Flow Rate    Meter Temperature
Time	Meter (Ft. )          CFM              °F	
                            Contents

Impinger No. 1 	

Impinger No. 2 	

Impinger No. 3 	
                   Figure 4.1   Field data sheet

                         4-8

-------
                 GAS SAMPLING - EVACUATED FLASK
TEST NO.
        DATE
   LOCATION
Type of Operation



Sampling Flask No.
Volume of Reagent, V
                    R
                      Leg 1



Initial Flask Vacuum, Leg 2



                      Total
P .
 i
                     Flask Volume,  V
                                                ml
                                       ml, Type of Reagent
 P. = P, - p. =
-   i     b   ^i


"Hg
                      Leg 1



Final Flask Vacuum,   Leg 2



            f         Total
                       - Pf = pb - pf =


                        "Hg
Initial Flask Temperature,  T. =
Final Flask Temperature, T~ =
Barometric Pressure, P,  =
                      b
Clock Time
                               °F + 460 =




                               °F + 460 =
                              "Hg
                         °R




                         °R
CALCULATIONS:
vs - (vo - V 
-------
true value.  They are caused by inability of the operator



to read scales precisely, by the quality and sensitivity



of the measurement device, and by uncontrolled environmental



variables .



     Application of standard statistical techniques has shown



that the maximum error associated with determining an emission



rate (product of concentration and total stack gas flow) is


          12 13
about 15%.   '    Under normal sampling conditions, with no



bias in the readings due to faulty equipment or operator



technique,  an error of less than 10.4% can be expected 99.6%



of the time.  The most significant error associated with any



one measurement involves reading the inclined draft gauge



used to measure stack gas flow velocity head.



     Another significant error can occur in measuring



particulate concentrations.  Particles are segregated at



the sampling nozzle when the velocity of approach to the



nozzle does not equal the velocity of the stack gas at the



sampling point.  This error varies with the size of the



particles,  as shown in Figure 4.3.



4.3  Analyzing the Total Measurement System



     Although, as mentioned earlier, many statistical tech-



niques of quality control cannot be applied to emissions



testing, a basic and detailed analysis of the total sampling



operation can aid the technical services staff in providing



high-quality samples that yield reliable data.
                         4-10

-------
     2.5
w
w
o
EH

Q
W
EH
U
W
O
U
O
H
EH
U
     2.0
   1.5
     1.0
       0
                    80-100  micron
               5-25
               micron
o

o
H
EH
       0        0.5       1.0       1.5       2.0

      RATIO OF NOZZLE VELOCITY TO ACTUAL STACK VELOCITY IN DUCT


 Figure 4.3  Expected Errors Incurred by Non-Isokinetic
                        Sampling

             (These data should not be used to correct
              concentrations obtained under non-isokinetic
              conditions since a wide variety of particle
              sizes is usually present.)


     The emissions testing duties of a technical services


group can involve a variety of sampling procedures related


to the variety of sources that are monitored.  Selecting the
                                 ;   i
sampling method most appropriate is the first step in quality


control.  Among the factors to be considered are the major


variables of the process to be sampled, location of sampling


points, and size of the sample.  Experience and judgement


will influence method selection; review of method descrip-


tions and consultation with others are often helpful, as


are relevant literature sources such as those given in


references 11 through 15.
                            4-11

-------
     Working within the framework of the method selected




for a specific emissions test, one can proceed with analysis




of the total measurement system.  Prepare a schematic



diagram of the sampling train; list the critical components



of the system and note how their operation affects results.



Set forth, as briefly and simply as possible, the theoretical



principles on which the system is based.



     These items should be prepared on paper for use by the



technical services staff.  The schematic diagram expresses



relationships in visual form, showing how one element of a



sampling/analysis system affects another.  It displays



opportunities for error, and therefore opportunities for



control.



     Systems analysis for quality control of emission test-



ing should incorporate as many as possible of the elements



described in Chapter 3 concerning atmospheric sampling and



Chapter 5 concerning laboratory operations.  Measures to



assure quality of reagents, for example, are appropriate



here.  Compilation and analysis of data histories (such as



calibration records) can lead to improvements in data



quality.   Documentation, by way of records, log books,



summaries, and reports, can provide the information base



required for quality control.  Although performed in the



field, often under conditions less formal than laboratory



operations, competent source sampling requires the dis-
                          4-12

-------
ciplined application of technical skills.  In terms of



quality control, 'stack sampling1 represents a challenge



to the technical services group:  to develop attitudes,



techniques, and detailed procedures that lead to consistent



production of high-quality emissions data.
                          4-13

-------
5.0  QUALITY CONTROL IN THE ANALYTICAL LABORATORY

5.1  Introduction

     Laboratory quality control programs should include

systematic procedures for performing analyses and for check-

ing the level of performance for each method used.   Statisti-

cal procedures are usually required for specifying perfor-

mance standards, for recognizing analytical results that do

not meet the standards, and for interpreting historical data.

Standard operating procedures provide a base for achieving

and maintaining a consistent level of analytical performance

and for tracing errors when results do not meet the expected

level.  Standard operating procedures should be developed

for each of the four major sources of analytical variation:
     0  Support services
     0  Reagents and materials
     0  Instrumentation
     0  Analytical technique
These may be considered as the chief physical parameters of

laboratory operations that are subject to quality control.

Each is treated in detail in later sections of this chapter.

     In the discussion that follows it is assumed that

methods have been standardized; analytical methods approved

by EPA should be used whenever possible.

     Techniques for controlling the major sources of

analytical variability are presented in this chapter, along

with the criteria for choosing the techniques best suited

to a specific laboratory situation.  Certain general
                         5-1

-------
criteria are applicable in nearly all laboratory appli-



cations; e.g., cost, requirements of the analytical methods,



experience of other laboratories, and effects of inter-



ferences as determined through sensitivity testing.




5.2  Controlling Physical Parameters



     5.2.1  Laboratory Support Services



     Laboratory support services include laboratory gases,



water, and electricity.  The parameters that affect the



quality of laboratory support services are given in



Table 5.1 with suggested control techniques.



     Decisions concerning the frequency with which generat-



ing and storage equipment should be maintained and reagents



checked need to be made.  Manufacturers' recommendations



provide a reasonable starting point, but schedules can be



adjusted as experience dictates.  Initially, checks should



be made more frequently than is recommended to provide



data for decisions on scheduling.  Maintenance contracts



with the manufacturer provide a convenient, and often



economically justified, method of meeting maintenance require-



ments.    When a laboratory reagent such as water or air



has been purchased commercially, it should be subject to



procedures, like conductivity tests for water, that will



verify manufacturers' quality statements and verify that the



reagent is not changing over a period of time in the labora-



tory.  The results of all verification procedures should be
                         5-2

-------
         TABLE 5.1  TECHNIQUES FOR QUALITY CONTROL
                    OF LABORATORY SUPPORT SERVICES
    Support
    Service
       Parameters
    Affecting Quality
       Control
      Techniques
 Laboratory
    Gases
Purity specifications -
vary among manufacturers

Variation between lots
                Atmospheric interferences
Develop purchasing
guides

Overlap use of old
and new cylinders

Adopt filtering and
drying procedures
 Reagent
  Water
Commercial source varia-
tion
                Purity requirements
                Atmospheric interferences
                Generation and storage
                equipment
Develop purchasing
guides - Batch test
for conductivity

Redistillation,
heating, deionization
with ion exchange
columns

Filtration of exchange
air

Maintenance schedules
from manufacturer
recommendations
Electrical
  Service
Voltage fluctuations
Battery power

Constant voltage trans-
formers

Separate lines

Motor generator sets
Ambient
Conditions
Temperature


Humidity
Heating and air
conditioning systems

Humidity controls
                          5-3

-------
recorded on standard format check sheets, which could in-

clude the following data:

     0  Date of receipt of container
     0  Container identification
     0  Manufacturer identification
     0  Lot number, if available
     0  Date of verification test
     0  Name and purpose of test
     0  Results of test
     0  Signature of the analyst

     Procedural guidelines for purchasing laboratory gases

and water can be developed by compiling definitions for

purity specifications used by different manufacturers.

Purchasing personnel should refer to such information to

avoid confusion when they are attempting to maintain a con-

stant quality of reagent, but must purchase from different

manufacturers.

     Procedures for testing the quality of support media

should be developed with the aid of professional publications

and manufacturers' literature.  References 17 through 22

will be useful in designing procedures for testing quality

of laboratory water and gases.

     Considering the final application of the data generated

will help in the decision of which control techniques to

apply.  The cost and effort required for some control

techniques may not be justified in terms of the purpose of

the analyses.  If the data are used for internal preliminary

survey work, for example, to determine the relative
                         5-4

-------
         TABLE 5.1  TECHNIQUES FOR QUALITY CONTROL
                    OF LABORATORY SUPPORT SERVICES
    Support
    Service
       Parameters
    Affecting Quality
       Control
      Techniques
 Laboratory
    Gases
Purity specifications -
vary among manufacturers

Variation between lots
                Atmospheric interferences
Develop purchasing
guides

Overlap use of old
and hew cylinders

Adopt filtering and
drying procedures
 Reagent
  Water
Commercial source varia-
tion
                Purity requirements
                Atmospheric interferences
                Generation and storage
                equipment
Develop purchasing
guides - Batch test
for conductivity

Redistillation,
heating, deionization
with ion exchange
columns

Filtration of exchange
air

Maintenance schedules
from manufacturer
recommendations
Electrical
  Service
Voltage fluctuations
Battery power

Constant voltage trans-
formers

Separate lines

Motor generator sets
Ambient
Conditions
Temperature


Humidity
Heating and air
conditioning systems

Humidity controls
                          5-3

-------
recorded on standard format check sheets,  which could in-

clude the following data:

     0  Date of receipt of container
     0  Container identification
     0  Manufacturer identification
     0  Lot number, if available
     0  Date of verification test
     0  Name and purpose of test
     0  Results of test
     0  Signature of the analyst

     Procedural guidelines for purchasing laboratory gases

and water can be developed by compiling definitions for

purity specifications used by different manufacturers.

Purchasing personnel should refer to such information to

avoid confusion when they are attempting to maintain a con-

stant quality of reagent, but must purchase from different

manufacturers.

     Procedures for testing the quality of support media

should be developed with the aid of professional publications

and manufacturers' literature.  References 17 through 22

will be useful in designing procedures for testing quality

of laboratory water and gases.

     Considering the final application of the data generated

will help in the decision of which control techniques to

apply.  The cost and effort required for some control

techniques may not be justified in terms of the purpose of

the analyses.  If the data are used for internal preliminary

survey work, for example, to determine the relative
                         5-4

-------
efficiency of a combustion process, control measures may

be considerably less important than they would be if the data

were used for setting emissions standards.

     5.2.2  Chemicals and Reagents

     A laboratory quality control program should include

standard procedures for choosing chemicals, preparing

standard solutions, storing and handling chemicals and

reagents, and choosing and handling standard reference

materials.  Some of the parameters affecting each pro-

cedure are listed in Table 5.2 with some appropriate control

techniques.

     The control techniques for choosing chemicals include

the development of purchasing guidelines.  These guidelines

should clarify the difference in grade designations used

by various manufacturers.  The American Chemical Society

classifications can serve as the reference for most

definitions.
     i
     Standard solutions may require occasional restandardi-

zation.  If the analytical method does not indicate the re-

quired frequency of restandardization, the frequency can be

established by considering any or all of the following

criteria:

     0  Normality of the solution
     0  Frequency of exposure to the  atmosphere
        Availability of inert storage containers
        Availability of storage that prevents temperature
        and light effects
                         5-5

-------
      0  Cost of restandardization as  opposed to making a
        new solution -  depends  on frequency of use

                          TABLE 5.2

  GUIDELINES FOR QUALITY CONTROL OF CHEMICALS AND REAGENTS
Procedure
  Control Parameter
    Control Technique
 hoice of
Chemicals
Manufacturer designa-
tions
Method purity specs.
Develop purchasing
guides
Use American Chemical
Society designations as
a base
Develop purification or
treatment procedures
specified by method
Preparation
of Standard
Solutions
Calibrated glassware

Standard reference
materials (SRM)

Stability
Purchasing guidelines
                                    Schedules for
                                    ardization of
              restand-
              solutions
Storage and
Handling
Container composition
Filtering or pre-
treatment
Environmental sensi-
tivity
Design a labeling system

Purchase single lot
numbers

Rotate stock

Control temperature,
light, atmospheric
exposure
Standard
Reference
Materials
Availability
Stability
Store in temperature
controlled atmosphere
Desiccate when necessary
Replace if instability
is suspected
Weigh to determine loss
or degradation
      Storage and restandardization requirements for several
                                          Q
 standard solutions as shown in Table 5.3.
                           5-6

-------
                         TABLE 5.3

              RESTANDARDIZATION REQUIREMENTS
Solution

0.02-1.0N
0.02-1.0LM
0.02-1.0N
0.1N

0.1N
0.1N
0.1N
0 . IN
0.1N
Sodium Hydroxide
Hydrochloric Acid
Sulfuric Acid
Iodine

Sodium Thiosulfate
Ammonium Thiocyanate
Potassium Dichromate
Silver Nitrate
Potassium Permanganate
Storage
Require-
ments

Pplyolef in
Glass
Glass
Amber Glass
Refrigerate

Glass
Glass
Glass
Amber Glass
Amber Glass
Frequency of
Restandard-
ization

Monthly
Monthly
Monthly
Open Bottles-
Weekly
Sealed
Bottles-
Monthly
Weekly
Monthly
Monthly
Monthly
Weekly
     Storage and handling procedures for chemicals and re-

agents should include the use of a labeling system.  Labels

on chemical bottles should include the following:

     0  Chemical Name
     0  Formula
     0  Manufacturer
     0  Lot Number
     0  Date Received
     °  Expiration Date

     Labels on standard solution bottles should include the

following:
                          5-7

-------
     0  Chemicals used
     0  Manufacturers
     0  Lot numbers
     0  Date of Preparation
     0  Date .of next standardization
     0  Standardization data
     0  Analyst identification
     0  Conditions of analysis (temperature, pressure,
        humidity)

     Standard reference materials are available from the

National Bureau of Standards and from commercial manufacturers

They are used for standardizing solutions, calibrating equip-

ment, and monitoring accuracy and precision of analytical

technique.  NBS classifies chemical standards as (1) primary

standards, (2)  working standards, and (3) secondary standards,

                                 22
having the following definitions:

     Primary Standard - A primary standard is a commercially
          available substance of purity 100 +_ 0.02% accuracy

     Working Standard - A working standard is a commercially
          available substance of purity 100 + 0.05% accuracy

     Secondary Standard - A secondary standard is a substance
          of lower purity that can be standardized against a
          primary grade standard

     The availability of primary standards may be limited.

Since analytical results rely on the accuracy of solutions

made with standard reference materials, it is important that

steps be taken in the laboratory to eliminate errors due to

mishandling of standard reference materials.  This is best

accomplished by providing the analysts with written pro-

cedures for handling of standard reference materials.  The

vendors usually recommend storage and handling procedures

(see Table 5.2) .

                          5-8

-------
     5.2.3  Instruments

     Standard procedures for operation, calibration, and

maintenance of analytical instruments are important for in-

terpreting results as well as for preventing errors.  Table

5.4 shows the important control parameters for analytical in-

struments and the control techniques that apply to those

parameters.

     Function checks can be used to indicate whether a sub-

system within an instrument is functioning properly within

predefined limits.  The limits are usually defined by the

instrument manufacturer, and the user should determine

whether the limits are acceptable for his requirements.  The

frequency with which function checks are performed depends

on how the instrument is used.  If the instrument is used

for short periods each week, function checks may be per-

formed with each use or even monthly.  If the instrument is

operated daily, function checks may be made daily or before

each run.

     The following example illustrates a function check of

linearity of a spectrophotometer.

     0  Prepare a standard data sheet to record relative
        concentration of standards, readings at various
        wavelengths, date of test, analyst identification,
        remarks.
     0  Prepare four standard solutions that absorb strongly
        at selected wavelengths in the range of the analytical
        methods.  Commercially prepared solutions were used
        for this example.
                         5-9

-------
                          TABLE 5.4

  TECHNIQUES FOR QUALITY CONTROL OF ANALYTICAL INSTRUMENTS
      Control Parameter
   Control Technique
Instrument Operating Range
Interferences
Environmental Conditions
Associated Equipment
(cuvettes, volumetric
ware, dilutors, etc.)
Normal System Drift

System Component Functions
Response Readout
Coordinate instrument selection
with method requirements.

Sample conditioning  (drying,
separating, mixing, etc.)

Use of blanks

Use of spiked samples

Monitor and control temperature,
humidity, pressure, any atmos-
pheric parameter that can
affect system response.  Con-
sult manufacturer instructions
and method descriptions.

Proper handling procedures

Standard procedures for clean-
ing

Standardization or calibration

Zero adjust

Apply function tests

Plot response to changing
concentrations

Perform maintenance when in-
dicated

Use calibration curve, adjust
using blanks and zero -
span controls
                          5-10

-------
     0  For each relative concentration, record the absor-
        bance at the specified wavelengths.
     0  Using Cartesian graph paper .(rectangular coordinates)
        plot a graph of absorbance vs. relative concentra-
        tion.  Prepare a separate graph for each wavelength.
        The graph should be a straight line if the instrument
        is functioning properly and if the absorbing solution
        follows Beer's Law.
     0  At each successive testing interval, plot the points
        as above.
     0  If the points for a wavelength appear to be on a
        straight line or to be equally distributed around
        it, the operator can assume that the instrument is
        functioning satisfactorily.
     0  If the points for a wavelength begin to fall con-
        sistently above or below the baseline and the
        distance between the points and the baseline increases,
        the operator should expect an impending problem
        that will require maintenance.

For the data obtained at 420 nm (Figure 5.1), the graph (Figure

5.2) indicates that the response is beginning to change.  The

manufacturer's literature should indicate whether the amount

of drift is acceptable.  If the drift is not within defined

tolerance limits, maintenance should be performed according

to manufacturer's recommendations.

     Calibration provides a technique for translating in-

strument response into meaningful concentration units.  A

calibration curve is constructed by analysis of materials

containing varying known concentrations of the element or

compound of interest.  Each time a curve is constructed, a

test is applied to determine whether instrument response

is within predefined limits.  If the calibration curve is

acceptable, i.e., if it lies within the statistical limits,

then the analyst can apply it to translate instrument output
                           5-11

-------
      SPECTROPHOTOMETER WEEKLY FUNCTION CHECK

             FOR LINEARITY AND PRECISION
Relative
Concentration
                 ABS.at ABS at ABS at
                 670 mu 520 mu 420 mu Date Analyst  Remarks
.l_._0_0_

i). 75
                         0.71*
                                      Lr/3
0.50
0.25
0.00
                                0,00
                                      \
1.00
                                             LtiE
0. 75
0.50
0.25
0.00
                  0,00
                                      \
\
1.00
                         0.7/0
0.75
0. 50
0. 25
0.00
                                       \
1.00
                                  ,6/J
0. 75
0.50
0.25
                                0, t
0.00
  .00
0.75
0.50
0.25
0.00
Figure 5.1
             Spectrophotometer weekly function  check  for
                linearity and precision

                         5-12

-------
         -0;8
                                          STANDARDISATION CHECK

                                             -AT'420 nm
ui
I
                                                                      Mfr.
                                                                       Drift
 :*e commended
   Limit
                                                                                              :...,Pirocedure
                                                                                          Requ'ired    .
                                0.25                 0.50

                                       RELATIVE CONCENTRATION
0.75
1.0
                              Figure 5.2  Standardization check at 420 nm

-------
into the appropriate concentration units.   The example for

constructing an S0_ calibration curve, presented in

Appendix C, will show how to establish control limits.

     The standards used to construct the calibration curve

can be primary or working standards, or spiked standards

designed to eliminate the effects of interferences en-

countered in real tests.  In calibration the standard is

exposed to the same preparatory steps that are applied to

unknown samples.  For example, if samples  are subjected

to a separation procedure, then the standard should also

be subjected to that procedure.  With this technique the

analyst can introduce the same net interferences, as closely

as possible, that will occur in normal application of the

analytical methods.  The choice of standard will depend on

several criteria:
     0  Method requirements
     0  Expected concentration range of unknown samples
     0  Known interferences determined by method
        standardization and sensitivity tests
     0  Availability of standard reference materials
     0  Cost of standard reference materials
        Stability of standards
o
     Many laboratories use a non-statistical approach to

constructing calibration curves.  The concentrations of the

standard that are analyzed to provide data for calibration

curves should cover the working range of the method.  The

data in Table 5.5 were used for constructing the calibration

curve (Figure 5.3) for S02 determination using the modified

pararosaniline method.

                          5-14

-------
                          TABLE 5.5
                    SO   CALIBRATION DATA
Concentration
0.20
0.40
0.60
0.80
1.00
1.20
1.40
Average Absorbance
0.095
0.240
0.315
0.440
0.565
0.660
0.780
     The steps used to construct the curve are:

     0  Analyze at least three concentrations of the
        standard.   The concentrations should cover
        the working range of the method.  Three
        replicates should be analyzed for each con-
        centration .

     0  Plot the average absorbance values (y-axis)
        for each set of replicates against the con-
        centrations (x-axis) on rectangular coordinate
        graph paper.  If percent transmittance is
        plotted, semi-log graph paper should be used.

     0  Establish a calibration line by drawing a line
        of best fit through the plotted points.  This
        line will not necessarily pass through the
        origin of the graph.  The y-intercept can vary
        depending on such variable parameters as equip-
        ment sensitivity, environmental influences, or
        degradation of the standard.

     Since the line .of best fit can shift, the analyst

should periodically analyze the standard at one of the con-

centration levels used to construct the calibration curve.

The result can be plotted on the curve to indicate the
                          5-15

-------
1/1
I
                                                      .6        .8        1.0

                                                   Concentration yg/ml
1..4
                              Figure  5.3   Calibration Curve for SO- Determination

-------
degree of curve shift, if any.  The frequency with which the


curve should be checked depends on the reproducibility of the


instrument, the ruggedness of the method, and changes in


environmental influences.  Control lines cannot be rigorously
                                                      i

determined with statistical techniques for the type of


calibration curve in this example.  The control lines in


Figure 5.3 merely illustrate that limits can be placed on the


calibration curve if the method description specifies the


limits.  Each time the standard is analyzed, the result should


lie within the established control lines.  If it lies outside


the control lines, a change in one or several parameters in the


analytical system has occurred, and either corrective action


or a new calibration curve is necessary.  The technique


just described is useful when the analytical method is to be


used infrequently.  However, when a method is being used


routinely, the preferred technique for constructing the


curve and its limits is regression analysis by the method


of least squares.


     To obtain maximum precision in determining the line of


best fit, the EPA suggests in the Federal Register using


regression analysis by the method of least squares.  If


this technique is used, statistical control limits for the


calibration curve can be established that relate to different


confidence intervals.  We suggest that if an analytical


method that requires a calibration curve is used continuously
                         5-17

-------
in the laboratory,  regression analysis by the method of

least squares is a  better technique for constructing the

calibration curves than is the previously explained technique.

An example of how this statistical technique is applied

appears in Appendix C.

     5.2.4  Analytical Technique

     The quality of analytical technique is a function of the

analyst's experience and training.  Basic operations, such

as dilution procedures and handling of analytical weights,

contribute significantly to indeterminate errors, i.e.,

errors that cannot be traced.

     Maintaining good analytical technique in the laboratory
                               23
is primarily a management task.    The laboratory supervisor

can use several methods to encourage good technique and to

insure a continuing effort to maintain good technique.

     0  Proper allocation of manpower
     0  Periodic review of analyst performance
     0  Implementation of maintenance schedules and
        procedures for all laboratory instruments and
        materials
     0  Design and review of error histories

     If he is applying one or all of these methods to maintain

good technique among the analysts, the supervisor can use a

form of statistical analysis to evaluate and compare the

performances.  A control chart can be constructed for each

analyst to show the variability of results he obtains with

any given analytical method.  Data used for such charts are

from analysis of replicate samples or of known standards

over a specified period of time.  The analyst's performance

                         5-18

-------
can be compared with his performance over another time period

to show trends in his proficiency, or it can be compared

with the performances of other analysts using the same

analytical methods.

     The major problems with designing a program to monitor

the analyst's performance are concerned with design of the

sampling system.  The problems are:
     0  What kinds of samples to use.
     0  How to prepare and introduce samples into the
        run without the analyst's knowledge.
     0  How often to check the analyst's proficiency.

     The problems and their suggested solutions or criteria

for decision are given in Table 5.6.

                          TABLE 5.6

        PROBLEMS IN ASSESSING ANALYST PERFORMANCE
   Problem
       Solutions and Decision Criteria
Kinds of Samples
Introducing the
 Sample
Frequency of
 Checking
 Performance
Replicate samples of unknowns or refer-
ence standards.
Consider cost of samples.
Samples must be exposed by the analyst
to same preparatory steps as are normal
unknown samples.
Samples should have same labels and
appearance as unknowns.
Because checking periods should not be
obvious, supervisor and analyst should
overlap the process of logging in
samples.
Supervisor can place knowns or replicates
into the system/occasionally.
Save an aliquot' from one day for analysis
by another analyst.  This technique can
be used to detect bias.

Consider degree of automation.
Consider total method precision.
Consider analyst's training and attitude.
                          5-19

-------
In this discussion of quality control in an analytical

laboratory we have considered the four major sources of

variation or error; that is

        the laboratory facilities
        the reagents and materials
        the instruments, and
        the analysts.

We have described some measures for quality control in each

category.  Now we turn to the second phase of quality con-

trol that is required for effective laboratory operations -

the application of statistical techniques and other review

and control practices that together constitute in-depth

analysis of the total measurement system.

5.3  Statistical Methods

     Statistical techniques provide a means of defining

acceptable levels of analytical performance and determining

whether those levels are being achieved and maintained.  The

major steps in developing a statistical evaluation system

are:

     0  Defining the performance levels.
     0  Choosing the statistical techniques.
     0  Constructing control charts.

     5.3.1  Defining Performance Levels

     Before a system for evaluating analytical performance

can be initiated, acceptable performance levels must be

defined.  Laboratories normally work in the 99% confidence

interval  (see Section 2.2.2 discussion of adjustment factors
                          5-20

-------
 and  confidence  intervals).   Statistical  formulae  can be

 applied  to  determine  confidence  limits.   No formula can  be

 applied  to  determine  confidence  level, but several practical

 criteria can  affect  the  choice:

      0 Method specifications
     . ° EPA  recommendations
      0 Ultimate use  of the  data
      0 Method standardization precision  and accuracy data
      0 Intralaboratory and  interlaboratory test results

      5.3.2  Choosing  Statistical Techniques

      Statistical techniques are  applied  to determine whether

the errors associated with analytical data are within

operational limits designated for the method.  Precision con-

trol charts are  prepared from results of replicate analyses

and are used to  monitor the degree of variability among

laboratory results.

     Precision control charts indicate the level of precision

in two ways:  (1) in the unit of measurement of the variable;

 (2) in percent.   Precision in units is calculated in terms

of range   (R-Chart) or standard deviation  (S-Chart).  Pre-

cision as percent is calculated in terms of the coefficient

of variation  (CV-Chart),  also referred to as relative

standard  deviation.

          R = Max - Min
          S =llZ(x-x)2          R = Range
             »  Z_i            S = Standard Deviation
                  •"•           CV =' Coefficient of Variation
              S_nnn.        ''   x = Individual Value
              _^1UUJ           x = Mean Value
              x                N = Number of Replicates
                          5-21

-------
      The use of R-Charts and S-Charts is based on the

assumption of homogeneity of variance (i.e., the variation

between replicate analyses on a single sample is constant

over the range of concentration being measured).  The

CV-Chart is used when precision is dependent on the con- •

centration being measured (i.e. the standard deviation

changes, usually increasing, with concentration).  The co-

efficient of variation, expressed as a percent, is indepen-

dent of concentration from the standpoint that the percent

is constant over the concentration range.

      Before an analytical method is adopted for routine use

in the laboratory, replicate analyses should be run on known

standards.  Prepare the standards to represent both the low

and the high concentrations expected and at least one, but

preferably two, intermediate concentrations.  As a good

"rule of thumb", between 5 and 10 replicate analyses should

be run at each known concentration.

      Compute the mean (x) and standard deviation (s.) for

each concentration,  x, s,,  x2 s-, x_ s.,, x. s..  Plot

the mean and standard deviation on a scatter diagram

(Figure 5.4).  Generally the scatter diagram will follow

the pattern of one of the two cases shown.

      Case A:  The standard deviation is independent
               of the mean.   Use either an R-Chart or
               S-Chart.
                          5-22

-------
                                            CASE B:

                                            Standard Deviation
                                            Increases with Concentration
   45678
    Concentration Mean
                                            CASE A:
                                            Standard Deviation is
                                            Independent of Concentration
   45678
    Concentration Mean
Figure 5.4  Scatter Diagrams for Determining Control Charts

-------
     Case B:  The standard deviation is dependent on changes
              in concentration.   Use the CV-Chart.

     Choice of an R-Chart or an S-Chart depends on the

number of replicate analyses to be run routinely to monitor

precision.  If the number of replicates  (n) is small

(n < 12), the R-Chart is the most efficient.  When n is

large (n > 12) the S-Chart provides a more efficient control

of precision.  Since precision is usually determined on the

basis of a small number of replicates, S-Charts for

precision are not discussed in this manual.  Table 5.7

presents data used to plot Case A and Case B in Figure 5.4.

     5.3.3  Constructing Range Control Charts

     The procedure for constructing a control chart for

range follows (see Table 5.8):

     0 List the absolute value of the range (R) for each
       set of replicates.
     0 Compute the average range  (R) for all sets of
       replicates.
     0 Compute the upper control limit by UCL = D.R.  D
       is from Table BII, Appendix B.
     0 Compute the lower control limit by LCL = D~R.  D_
       is from Table BII, Appendix B.
     0 Plot .the line for R on the control chart.
     0 Plot the values for ranges of each set of replicates
        (Figure 5.5).

     The control limits computed for this control chart are

for the 99% confidence interval  (see Section 2.2.2).  There-

fore 99%of the calculated range values should be between

these control limits.
                          5-24

-------
               TABLE 5.7
DATA USED TO CONSTRUCT SCATTER DIAGRAMS
CASE A
X
2.3
2.2
2.3
2.4
2.1
x =
3.0
3.1
3.2
2.9
3.2
x =
5.1
5.0
5.2
5.1
5.0
x =
9.5
9.6
9.4
9,7
9.6
v —
x-x
0
-.1
0
1
-.2
2.3
-.1
0
.1
-.2
.1
3.1
0
-.1
.1
0
-.1
5.1
-.1
0
-.2
.1
0
9.6
(x-x)2
0
.01
0
.01
.04
S = .12
.01
0
.01
.04
.01
S - .13
0
.01
.01
0
.01
S = .08
.01
0
.04
.01
0
S = .12

2
2
2
2
2
x
3
3
3
2
3
x
5
5
5
4
5
x
9
9
9
9
9
x
X
.3
.2
.3
.4
.1
=
.0
.1
.2
.9
.2
=
.1
.2
.4
.9
.3
=
.5
.1
.6
.0
.2
=
CASE B
x-x
0
-.1
0
1
-.2
2.3
-.1
0
.1
-.2
.1
3.1
-.1
0
.2
-.3
.1
5.2
.2
-.2
.3
-.3
-.1
9.3
(x-x)2
0
.01
0
.01
.04
S = .12
.01
0
.01
.04
.01
S = .13
.01
0
.04
.09
.01
S = .19
.04
.04
.09
.09
.01
S = .26
                5-25

-------
            TABLE 5.8  COMPUTATION OF CONTROL
             LIMITS FOR RANGE CONTROL CHARTS
Sample (n)            x,             x?               R
1
2
3
4
5
6
7
8
9
10

_
R
21
39
14
8
59
88
7
88
38
22

R 68
~ n 10 6'8
29
47
18
10
71
96
9
98
46
28
R Total =


8
8
4
2
12
8
2
10
8
6
68


     UCL = D4 R = 3.267 x 6.8  - 22.22
     LCL =D3R=Ox6.8=0
     D., and D  are for n =  2
 OJ
 tn
               UCL  =  22.22
                                Average Range = 6.
                         Order of Replicates
                  Figure 5.5  Control Chart for Range
                           5-26

-------
     Data plotted on precision control charts should be




approximately evenly distributed around the mean value line.




The Theory of Runs dictates that, for the 99% confidence




level, if eight successive values appear on one side of the




mean value line, then the process is judged to have bias.




If this occurs, stop the process and follow a standard




routine for error tracing to find the source of error that




is causing the bias.  Note the problem and the error source




on the control chart for easy future reference.  A system




for tracing errors is described later in this chapter.




     If any of the results plotted on the control chart fall




outside of the control limits, the process should be judged




out of control and the analysis stopped.  Whenever possible,




reanalyze all affected samples when a bias or an out-of-




control situation is observed.  The feasibility of identi-




fying and reanalyzing affected samples depends on the




frequency with which replicate samples are analyzed.  For




example, if few samples are analyzed by a method each day,




and only one set of replicates is analyzed each day, bias




would not be confirmed on the control chart until at least




the end of eight consecutive days.  It would be impractical,




and in many cases impossible, to save an aliquot of each




sample for at least eight days for the possibility of reruns




Keep in mind that biased results are still acceptable if the




control limits are not exceeded.  The most important step is
                          5-27

-------
to find the source of error and correct it before it causes




the analytical process to go out of control.



     inspection of the control chart will help to determine



which samples may have been affected if bias or loss of



control occur.  In a bias situation, all samples associated



with the eight sets of replicates that produced the plotted



control data can be considered to have been affected.  This



situation is illustrated by Case A in Figure 5.6.  When loss



of control is indicated, all samples analyzed between the



last set of replicates that showed control and the set of




replicates that indicated loss of control can be considered



to have been affected (Case B, Figure 5.6).  An exception to



Case B would be the appearance of an upward or downward trend



of the control data prior to loss of control (Case C,



Figure 5.6).  This could be a situation in which the



analytical process was subject to bias but went out of



control before enough replicates were analyzed to display



the bias.  In this case the control data points are traced




back to the last point appearing on the side of the mean



value line opposite the points that indicate suspected bias.



All samples beyond this point could have been affected.



     5.3.4  Constructing Coefficient of Variation Charts



     The CV-Chart is prepared from measurements obtained



from replicate analyses of routine samples.  Typically the



number of replicates is two.  The duplicate analyses are
                          5-28

-------
0)
Cn

Case B - Ou
:at last po
""••; 	 ";""••" 	 'UCL 	 ' ""
.: 	 : . - f- 	
' --; -I. J:-!/^; k • ..^.r;
y >1 ' J " '~ Mean V;i"l UP

^v iv : :
' i JLCL
: i " '
; . ' -Order of Result
- i , - Case A - Bi
: .. . _ [eight ..cons
i ; , ; on one .sic
.' ] ' '' \": :"' -' ----<- : -" line". "
— .! • .':':" "'".. j ;."
• ! l • -• . [ - -. IUCL •_ -•_ •:
. , . . !
. . o — if^- 	 — .._;Mean- Value
r - ;
:LCL •
• - ; ; j - ; v
t of contr
int
A
; -, -' •— 1 :
; . i . . ... . ; .!. .!
' "1 ', ' !
.. -i
-..._, -• ,
< - • ' , ' '
i • ' '•
'-'•',-
,_, _ : }. ...

s J : ;
as indicat
Ol
. • - "" . • i : ,
i " ' '••>)•
." ' ' "•';".
• -- -- ; :- ;-•!- '- 'oj;
: : . ' : . Cn
C
(3
: «.
ed by



A 7
\ ./
.: ^

ecutive points ^ ~
e 'of mean lvalue ! ' , - :
'..!.'. i - 'All samp
- have bia
: .:
ft



/



le
s
i ' : .
Figure 5.6 Interpre


i
•' 	 * 	
• •' ,
Ca
c
': " . - .'-b
-./"
! jt~ - -. '•• -'-



' - Grde
s past thi
tation of

se C - Out
on'tr.o.l and
ia|s ; i ;
• .- _i. . ^ i.
UCL i
Mean Value

LCL , , :

r jof Resul
s point ma
Control Ch
•

of
pps'sible
. — j.... . ,—
.._ -.^ -i ..
- :- --•; - - --;--
'
'
" "" . :::.\
'
. \
...... !.H...


ts: r:T"
y . • ' :.
• i
arts

                 Order of  Results

-------
run on 5-10 percent of the incoming samples, selected by a

random process.  The duplicates are sent to the analyst in

such a manner that the analyses are run "blind", i.e. the

analyst cannot determine which samples are being used to

monitor precision.

     It is difficult to rigorously determine the number of

duplicate analyses needed to establish the necessary control

limits, but a minimum of 15-20 is preferred.  Similarly, as

the duplicate analyses yield more data, control limits

should be revised periodically, perhaps monthly or quarterly

Cumulative data may indicate that control limits should be

revised annually.

     The procedure for constructing the CV-Chart for the

data in Table 5.9 is described below.  The term "sample"

            TABLE 5.9  COMPUTATION OF CONTROL
                    LIMITS FOR CV-CHART

Sample      x,     x_     R       x           CV
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
23
39
14
8
59
78
7
80
38
22
12
29
48
75
48
80
29
47
18
10
71
96
9
98
46
28
16
35
60
91
58
100
6
8
4
2
12
18
2
18
8
6
4
6
12
16
10
20
26
43
16
9
65
87
8
89
42
25
14
32
54
83
53
90
16.3
13.2
17.7
15.7
13.0
14.6
17.7
14.3
13.5
17.0
20.2
13.3
15.7
13.6
13.3
15.7
                                    I CV = 244.8
                          5-30

-------
refers to a set of duplicates.  The range is used here for
convenience, since it is an efficient estimate of standard
deviation for n = 2 .
     0 Compute the range R for each_sample .
     0 Compute the arithmetic mean x for each sample.
     0 Compute the coefficient of variation  for each
       sample , where :
       CV =   R
        n = sample size 2
     0 Compute the average coefficient of variation.
            16
       _   Z CVi
       CV = 1 _  = 244.8 =  15.3
              16      16
     0 Compute the upper control limit.
               UCL = B4 CV
       where B. = 1.552 (Table Bill, Appendix B)  for
              4   n = 16
               UCL = 1.552 (15.3)  = 23.75
     0 Compute the lower control limit.
               LCL = B3 CV
       where B3 = 0.448 for n = 16
               LCL = 0.448 (15.3)  - 6.85
     0 Prepare the CV-Chart (Figure 5.7) showing CV,
       UCL, and LCL.
    'The CV-Chart is now ready for use to monitor precision
of routine sample analysis.  For each sample compute CV and
plot new values sequentially on the control chart.
     5.3.5  Determining Accuracy
     Accuracy determinations involve the comparison of
results from analysis of unknown samples with results from
                          5-31

-------
20
                A
                            UCL = 23.75
                                = 15.30
v
10
                           v
                                  LCL =6.85
 0
                               Order of Results
                  Figure 5.7   Coefficient of  Variation Chart
   analysis of standards  of  known concentration.   Two



   techniques  for accuracy determination,  classified by type



   of reference material,  are  shown  in  Table  5.10  with  the



   criteria that affect the  choice of each.



        One problem with  using a  primary or a working standard



   for accuracy determination  is  that these standards do not



   approximate the quality of  actual samples.   Field samples




   are exposed to a variety  of sources  of  interference  during



   their collection and transport to the laboratory.  After



   arrival  at  the laboratory,  the samples  are subjected to more




   potential sources of interference during sample preparation.



   The effects of those interferences are  significant for many
                             5-32

-------
     TABLE 5.10  TECHNIQUES FOR DETERMINING ACCURACY
Analytical Technique
Criteria for Application
Pure standards
Spiked samples
(% recovery
 technique)
Often not available.
Advantage - composition is
well established.
Disadvantage - does not
duplicate the sample.
Interferences normally
found in samples are not
present, and effect on
method cannot be measured.
Can be used when stable
standards cannot be obtained,
Can be used when accuracy
determinations are very
infrequent.
Applicable for trace
analysis.
types of analyses, such as atomic absorption analysis for

trace metals.  Consequently, when accuracy is being deter-

mined it is often desirable to compensate for the net

effects of normal interferences.  The standard additions

technique does this by providing a correction factor that

can be applied to observed values to calculate true values

for the analytical results.
steps:
     The method of standard additions includes the following
      24
       Analyze an aliquot of the unknown sample.
       Add to another aliquot of sample a portion of
       a standard of the species of interest such that
       the total concentration of the resulting solution
       will be within the optimum detection range of
       the method.  Analyze the resultant mixture.
       Calculate the recovery of the added substance
       with the following formula:
                          5-33

-------
       C(s + a) -Cs x 100 = % RecoVery
            L,a

       where:  C(s + a) = Observed concentration value of
                          sample plus standard

               Cs = Observed concentration value of sample

               Ca = Actual value of standard

The percent recovery represents a correction factor for

interferences, and the true value of a sample concentration

is calculated by multiplying the observed value by the per-

cent recovery.  Another way of expressing the correction
                                  •p
factor is by use of the formula:  •=-  = CF
                                  £\
      R - Quantity Recovered
      A = Quantity Added
     CF = Correction Factor
               ov
     Then TV = ^

     CV = True Value
     OV = Observed Value
     CF = Correction Factor

     5.3.6  Control Charts for Accuracy

     Control charts on which to plot the results from

accuracy determinations can be constructed in much the

same manner as precision control charts.  Standard deviation

is convenient for measuring the variability among the

accuracy determination results.  Two situations are dis-

cussed in this section:  (1) a primary or working standard

is analyzed to determine accuracy; (2) the method of

standard additions is used to determine accuracy.
                          5-34

-------
     Figure 5.8 illustrates the first situation, in which

percent nitrogen has been determined by titration.   Table

5.11 presents the data used.   The control chart in  Figure

5.8 was constructed by the following steps:

     0 Analyze 15-20 aliquots of a primary standard over a
       period of time sufficient to insure that normal lab-
       oratory operating conditions are represented.  Com-
       pute the mean percent nitrogen value for all analyses
       This value will represent the nominal value  about
       which analytical values from later tests will be
       plotted.  Do not be confused by the expression of
       nitrogen concentration as a percent.  This is not a
       percent recovery technique.

     0 Calculate the standard deviation by:

           S ="\| £ (x - x^1
               1  n - 1

     0 Calculate a lower and upper control limit by:

           UCL = x + D4S

       and LCL = x - D S

       Use D4 = 1.652 for the 99% confidence level, when
       n = 15

     0 Construct the chart.

     0 The formula for computing standard deviation in this
       example is interchangeable with the formula  in the
       next example.

     In the second example percent recovery is used to

determine accuracy:  In the percent recovery technique the

control chart represents recovery efficiency, whereas in

the technique just described the control chart represented

analytical values.  In preparing a control chart for per-

cent recovery, follow these steps:
                          5-35

-------
       TABLE 5.11  ACCURACY DATA FOR PERCENT NITROGEN
 Sample

    1
    2
    3
    4
    5
    6
    7
    8
    9
   10
   11
   12
   13
   14
   15
% Nitrogen (x)

    25.89
    25.92
    25.87
    25.83
    25.79
    25.53
    25.39
    26.00
    25.53
    25.90
    25.83
    25.60
    25.65
    25.80
    25.40
x - x

 .17
 .20
 .15
 .11
 .07
-.19
-.33
 .28
-.19
 .18
 .11
-.12
-.07
 .08
-.32
S = 0.207
x = 25.72

D4 S = 1.652  (0.207)

     = .342
UCL = 25.72 + .342

    = 26.06



LCL = 25.72 - .342

    = 25.38
c
0)
Cr>
O
•H
-P
C

-------
       Perform the  recovery procedure 15-20 times and use
       the results  for calculations.
       Calculate the average (x)  for  all percent recoveries
       The average  will become  the mean nominal value for
       the control  chart.
       Compute the  standard deviation of the percent
       recoveries by:	
               S =
  |n £x  -  (Ex)

     n(n - 1)
     Convenience determines which formula to use for

standard deviation.

     0 Compute the upper control limit and the lower
       control limit by UCL = .x + D4 S and LCL = x - D4 S.
       D ,  as it was in the previous example, is based on
       n =  15.
     0 Plot the mean value for all the recoveries on the
       control chart and construct the upper and lower
       control limits.  The control chart is now ready for
       plotting percent recovery data as generated.  Table
       5.12 shows the data used to construct Figure 5.9.
    110
 OJ
 >

 8 100
 o\o
    90
                  UCL =  103.69 Recovery
99.74% Recovery
                  LCL = 95.79 Recovery
        Figure 5.9  Accuracy Control Chart  for  %  Recovery
                           5-37

-------
            TABLE 5.12  PERCENT RECOVERY DATA
                      % Recovery
Sample                _ x _

   1                      .982        £x2 = 14.93
   2                      .990
   3                      .993        D.  = 1.652 for n = 15
   4                     1.005         _
   5                     1.021         x  = .9974
   6                     1.030
   7                      .970         S  = .0239
   8                      .975
   9                      .985       UCL  = x + D  S
  10                       QQ fi
  11                     . nil            = .9974 + 1.652 (.0239)

  12                     1 020              1'°369
  ll                     I 015            = 103-69%

                            8        LCL  = * - °4 S
                                          = .9974 - 1.652  (.0239)
                                          = .9579 = 95.79%

The control chart constructed from the data in Table 5.12

has the mean nominal value line at the point representing

the 99.74% recovery level.  The upper control limit repre-

sents the 103.69% recovery level, and the lower control

limit represents the 95.79% recovery level.  Perfect

recovery would be at the 100% level.  Negative interferences

would be expected to keep the recovery below 100%, but

positive interferences can also influence an analysis .  The

positive interferences are stronger than the negative when

the percent recovery is found to be more than 100 percent.

The direction of each percent recovery from the 100 percent

level, then, can indicate the kinds of errors to look for

when control limits are exceeded.
                          5-38

-------
5.4  Interlaboratory Proficiency Testing




     The purpose of an interlaboratory proficiency testing



program is to determine the relative proficiency of a group



of laboratories and/or analysts in performing an analytical



method.  Laboratories participating in such programs are




provided with standard reference samples and instructions



for analysis of those samples.   A co-ordinating laboratory



prepares the samples and evaluates the results.  Partici-



pation in interlaboratory programs is recommended.



     Several problems can occur in proficiency testing



programs.  Some are procedural  or administrative problems,



and some are technical problems related to method standardi-




zation and ruggedness testing.



     How to obtain unbiased treatment of the sample is an



important problem in proficiency testing programs.  Samples



that are analyzed as part of a  proficiency testing program



should be placed in the sample  run as unknowns.  If the



analyst recognizes the sample as coming from the referee



laboratory, he could use extra  care in the analysis, and



the result would not represent  his normal proficiency with



the method.



     Stability of the sample is an important technical



consideration.  The referee laboratory should determine the



net effect of storage and transportation on the sample.  The



Center for Disease Control, Atlanta, Georgia, has employed
                          5-39

-------
a technique called "pigeon sampling".   A sample is prepared



and an aliquot is analyzed.  Another aliquot is shipped to a



designated point and returned unopened to the central labor-



atory.  The aliquot is analyzed, and the results are compared



with those obtained for the sample aliquot before shipment.



The difference is an indication of the net effect of trans-



porting the sample.



     Before a method can be used routinely to rate the pro-



ficiency of laboratories participating in the program, it



must be subjected to extensive testing to determine the



total effect of changes in operating parameters.  Disagree-



ment among analytical results from different laboratories can



indicate that the laboratories are not controlling to the



same extent the parameters that affect the final results.



     A case study involves the disagreement in results from



two laboratories using the SPADNS - Zirconium Lake Method


                                            25
for determination of fluorides in stack gas.    A statis-



tical analysis indicated that the difference between the



results of Laboratory A and Laboratory B was significant.



     Because of the significance of the difference in



analytical results,  steps were initiated to determine any



operational differences in the use of the method by the two



laboratories.  The investigation showed two major differences



     1) Laboratory B varied the steps of the method, but

        varied each step in the same manner each time;
                          5-40

-------
     2)  Laboratory A stayed within the operational ranges
        for each parameter, but varied inconsistently within
        those ranges.  The specific differences are shown
        in Table 5.13 below.

       TABLE 5.13  VARIATIONS IN METHOD PROCEDURES

Method Description        Laboratory A      Laboratory B

Charge still with        Same as method.     Charge still with
0.5 to 0.9 mg                               0.4 to 0.6 mg
fluoride                                    fluoride

Perform determina-       Same as method     Perform deter-
tions at 15° to                             minations at
30°C                                        24.8°C

Change H2S04 '  H_0       Same as method     Change H2SO.
mixture when recovery                       H,,0 mixture
check indicates                             every three runs
necessity

     A program was undertaken at Laboratory A to determine

the effects of varying the parameters outlined in the method

description.  The investigation showed that change in the

temperature parameter made a significant difference in the

analytical results.  The procedure follows:

     0 Data were obtained by using the method with
       temperature control  (SPADNS  (b)) and by using
       the method without temperature control  (SPADNS  (a)).

     0 Data were obtained by using the method with
       temperature control  (SPADNS  (b)) and by using a
       specific ion electrode  (c).

     0 The differences in results between  (a,b) and  (b,c)
       were analyzed to determine significance.

     Comparative data analysis gave the following results:
                      SPADNS(a)-SPADNS(b)  SIE(c)-SPADNS(b)
No. of samples                18                 18
Average difference          -0.79              -0.10
Standard deviation           0.98               0.67
of difference
Student  "t" value     -3.4  (Significant)   -0.66  (Not  Signi-
                                                  ficant)

                           5-41

-------
     NOTE:  For a sample of size n = 18, if t = -0.66, the



probability that the two methods yield similar results is



about 0.50.  However, if t = 3.4 the probability that the



two methods yield similar results is less than 0.005.  It is




generally accepted that a probability of 0.05 or less is




sufficient to allow one to conclude that the difference



between the methods is statistically significant.



     These data show that the temperature is more critical



than was indicated in the initial method description.  For



use in interlaboratory proficiency testing, it is important



that the operating parameters of the analytical method be



specifically defined.




5 . 5  Tracing Errors



     When precision or accuracy control charts indicate



that the analytical process is in error, it is necessary to



find the source of error and to correct it.  The histories



and records from calibration, function checking, maintenance,



and material quality checks can provide information with




which to trace the sources of analytical variability.



     To make optimal use of time and resources when tracing



an error, the analyst should perform the search in a standard



logical pattern designed so that the most obvious sources of



error are considered first.  If the search must then progress



to less obvious sources of error, the procedure becomes



more involved.  A review of reagent standardization charts
                          5-42

-------
may be required and perhaps such procedures as parallel

analysis using the original standard solutions before and

after restandardization,  if the restandardization results

differ.

     Table 5.14 illustrates one classification of error-

tracing procedures according to complexity.

       TABLE 5.14  CLASSIFICATION OF ERROR SOURCES
Source of Error
   Explanation
Input to Tracing
    Procedure
Variation of Method
Steps (Intentional)
Unintentional
Method Variation
Change in Sample
Analytical Error
Change in Reagent
or Materials
A procedure or
material speci-
fied has been
changed or
substituted.

A procedure has
been altered or
deleted uninten-
tionally.

The sample has
changed or an
interference
has been intro-
duced .
An error has
occurred in a
basic procedure
such as dilu-
tion, scale
reading, weigh-
ing, etc.

A change has
occurred in a
reagent or in a
material selec-
ted to use in
the preparation
or storage of
reagents or in
preparation of
sample.	
Control cha^rt - look
for trend on all
samples.  Method
description.


Method description -
note special require-
ments such as sample
preparation.
Field sampling
records. Sample
label data.
Calibration  data for
dilutors, volumetric
ware, etc.
Calibration curves.
Maintenace logs.
Function check
results for instru-
ments .
Standardization and
check lists for
reagents and labora-
tory services.
Method description -
look for storage
requirements, pre-
parative steps.
                          5-43

-------
     The flow diagram, Figure 5.10, shows the steps that




would be followed for the third level of complexity in error




tracing.  At this level the analyst suspects that inter-



ferences have been introduced into the sample during



collection, storage or transport, or during sample condition-




ing and preparation in the laboratory.  Pertinent input data




are shown, and logical alternatives are suggested.



     The specific classification of error sources and the



logical steps involved in tracing the errors differ in each



laboratory.  Application of the error tracing technique



depends on the methods used, complexity of the routine



operational checking and maintenance procedures, and the



training of the analysts and their understanding of the




methods.
                          5-44

-------
   SAMPLE ERROR
                              , Compare
                              . Sources
                               For all
                               Samples
                          No /  Samples
                               Compatible
                                                                        Adjustment
                                                                        Possible
          Field  Data
            Method
          Descriptio

	 .JUt
Yes
i
Identify
                              Interferences
                               I	I
                                 Std.  Pro-
                                 cedures
                                 Followed
                                                                                 Procedures
Figure  5.10   Procedures for Tracing
                Sampling Errors
                                                                       Data or Input Document
                                                                        Decision
Critical Steps

A Check Sample Source
B Check Field Handling
C Check Collection Process
                                           5-45

-------
6.0  DATA HANDLING AND REPORTING



6.1  General




     Measurements of the concentration of a contaminant,



either in the ambient atmosphere or in the exhaust gas from



an emission source, are assumed to be representative of the



conditions existing at the time of sample collection.  The



extent to which this assumption is valid depends on the



sources of error and bias inherent in the collection, handling



and analysis of the sample.  Methods that have been thoroughly



researched and evaluated should have minimal error and no bias.



     Besides the sampling and analytical error and bias,  human



error may be introduced anytime between sample collection and



sample reporting.  Included among the human errors are such




things as failureoof the technician to record pertinent in-



formation, mistakes in reading an instrument, mistakes in



calculating results, mistakes in transposing data from one



form to another.  Data handling systems involving the use



of computers are susceptible to keypunching errors and



errors involving careless handling of magnetic tapes and



other storage media.  Although human error cannot be com-



pletely avoided, it can be minimized.



6.2  Data Recording



     Methods for determining concentrations of air contami-



nants can be classified into two categories:  (1) intermittent,



(2) continuous.  In most intermittent methods a discrete
                         6-1

-------
sample is collected in some media and is sent to a labora-



tory for the analytical determination.  Both the field



technician and the laboratory analyst can make errors in



data handling.  Continuous methods involve an analytical



sensor that produces a direct readout of the pollutant con-



centration.  The readout may be an analog record traced on



a moving strip chart, or it may be a value punched on paper



tape or written on magnetic tape.  Some systems use tele-



metry to transmit data in real-time to a data processing



facility.



     6.2.1  Data Errors in Intermittent Sampling



     The field technician records information before and



after the sample collection period.  This information in-



cludes identification of the sampling location, start and



stop date and time, and data pertaining to flow rates, etc.



It is necessary to rely on the integrity of the field



personnel that such items as location, start and stop time



and data are recorded correctly.  Acceptability limits



should be set for data pertaining to flow rates, etc., and



the technician should invalidate the sampling data when



values fall outside of these limits.  Questionable measure-



ment results indicate the need for instrument maintenance or



calibration.



     The analyst in the laboratory reads measurements from



balances ;-. colorimetersv~spectrophotometers , and other
                         6-2

-------
instruments and records the data on standard forms or in




laboratory notebooks.   Each time he records a value he has




the potential for incorrectly entering the result.  Typical



recording errors are transposition of digits (i.e. 216



could be incorrectly entered as 126)  and incorrect decimal



point location  (i.e. .0635 could be entered as 0.635).



These kinds of errors  are difficult to detect.  The labora-



tory director must continually stress the importance of



accuracy in recording  results.




     6.2.2  Data Errors in Continuous Sampling



     Continuous air monitoring systems may involve either



manual or automated "data recording.  Manual data recording




is used to reduce data from strip charts.  Automated data




recording may involve  the use of a data logging device to



record data on paper tape or magnetic tape at the remote



sampling station, or the use of telemetry to transmit data



on-line to a computer  at a central facility.



     Manual reduction  of pollutant concentration data from



strip charts can be a  significant source of data errors.  In



addition to those errors associated with recording data



values on record forms, the individual reading the chart can



also err in the determination of the time-average value.



Usually the reader estimates by inspection the average con-



centration.  When the  temporal variability in concentration



is large, it is difficult to determine an average concen-
                         6-3

-------
tration.  Two persons reading the same chart may yield re-



sults that vary considerably.



     Individuals who will be responsible for reducing data



from strip charts should be given extensive training.  After



a trainee is shown how to read a chart, his results should



be compared with those of an experienced technician.  Only



after a technician has demonstrated that he is capable of



obtaining satisfactory results should he be assigned to the



data reduction activity.



     Periodically the supervisor or senior technician should



check strip charts read by each technician.  When it becomes



obvious that an individual is making gross errors it may be



necessary to provide additional training.



     Because manual chart reading is a tedious operation, a



drop in productivity and reliability can be expected after a



few hours.  Ideally an individual should be required to



spend only a portion of a day at this task.



     The use of a data logging device to automate data



handling from a continuous sensor is not a strict guarantee



against data recording errors.  Internal validity checking



is necessary to avoid serious data recording errors.  There



are two sources of error between the sensor and the recording



media:  (1) the output signal from the sensor; (2) errors in



recording by the data logger.



     A system recently installed by the Division of Air



Pollution Control, Cincinnati, Ohio, has validity checks




                         6-4

-------
for both sources of error.   In this system a number of air



quality and meteorological sensors are interrogated at 5-



minute intervals.  The data values are assembled into a



record and written on magnetic tape.  Two of the data



channels in each record are reserved for an electronic check of




the data logger.  One data channel is programmed to read



0000 + 0005 and the other to read 1600 + 0010.  If the



value recorded in either data channel is not within the pre-



scribed limits the recorded data values for all of the



sensors is of questionable validity.



     The second validity check, performed once each day, is



designed to test the electronics of each sensor.  This also



is a two point check in that each sensor transmits a signal



representative of 10 and 70 percent of scale.  For the



first 5-minute period the signal corresponding to 10 percent



of scale is transmitted from the sensor to the data logger



and then to the magnetic tape.  Similarly for the next 5-



minute period the signal for 70 percent of scale is written



on the magnetic tape.  Small tolerances are permitted for



both levels on the scale.  Should the value recorded on tape



for a particular sensor fall outside of the acceptable range,



all data for that sensor (since the prior sensor check) is



of questionable value.



     For a system involving the use of telemetry it is also



necessary to include a validity check for data transmission.
                         6-5

-------
6.3  Data Validation
     Data validation is the final step in handling raw
measurement data from air quality monitoring equipment or
emission source testing.  Data validation involves a critical
review of a body of data in order to locate spurious values.
It may involve only a cursory scan to detect extreme values
or a detailed evaluation requiring the use of a computer.
In either situation, when a spurious value is located, it is
not immediately rejected.  Each questionable value must be
checked for validity.
     6.3.1  Data Validation for Manual Techniques
     Both the analyst and the laboratory supervisor should
inspect intermittent air quality monitoring data and emission
source testing data.  At regular intervals, daily or weekly,
results should be scanned for questionable values.  This
type of validation is most sensitive to extreme values, i.e.
either unusually high or low concentrations.
     The criteria for determining an extreme value are de-
rived from prior data obtained at the particular sampling
site (or a similar site if no previous data is available
for a site).  The data used to determine extremes may be
the minimum and maximum concentrations for all prior data
from a site.  The decision criteria might also be based on
minimum and maximum for each season, each month., or each
day.
                         6-6

-------
     The time spent checking data that has been manually re-

duced by technicians depends on the time available and on

the demonstrated abilities of the technicians to follow in-

structions.  No agencies appear at this time to be using a

specific formula for determining how much data should be

checked for validity in a manual data reduction system.  One

air pollution control agency approached the problem in the

following manner:

     0  A senior technician or supervisor was assigned to
        check approximately 10% of the data interpreted by
        each of four or five technicians.  The 10% figure
        was arbitrary based on time availability and exper-
        ience in finding errors.

     0  Data was checked for obvious trends or unusual
        values indicating possible reader bias.

     0  No statistical formula was applied to determine
        the significance of differences between readings
        interpreted by the technician and readings inter-
        preted by the senior technician or supervisor.  If
        the two values differed by more  than two digits in
        the last significant figure, the data was judged
        unacceptable.

     0  Each analyst's technique of data interpretation
        was checked against written procedures describing
        the use of graphic aids to determine if those
        graphic aids had been properly used.  The most
        significant errors originated from the technician
        deviating from the written procedures — not from
        random error.

     6.3.2  Data Validation for Computerized Techniques

     A computer can be used not only to store and retrieve

data but also for data validation.  This will require the

development of a specialized computer program.  The system
                         6-7

-------
for checking extreme values in manual techniques also




applies here.  The criteria for extreme values can be refined




to individual hours during the day.  With this procedure



an hourly average concentration for carbon monoxide of 15



ppm may not be considered as an extreme value for 8:00 A.M.




but could be tagged as questionable if it appeared at 2:00



A.M.



     Another indication of possible spurious data is a large



difference in concentration for two successive time intervals.



The difference in concentration, which might be considered




excessive, may vary from one pollutant to another and quite



possibly may vary from one sampling location to another for



the same pollutant.  Ideally this difference in concentration




is determined through a statistical analysis of historical



data.   For example, it may be determined that a difference of



0.05 ppm in the S0? concentration for successive hourly



averages occurs rarely (less than 5 percent of the time).




But at the same station the hourly average CO concentration



may change by as much as 10 ppm.  The criteria for what con-



stitutes an excessive change may also be linked to time of



day.



     The extent of the decision elements to be used in data



validation can not be defined for the general case.  Rather,



the validation criteria should be tailored along the lines



suggested above for varying types of air monitoring networks.
                         6-8

-------
6.4  The Statistical Approach to Data Validation



     6.4.1  Maintaining Data Quality in Manual Data Reduction



            Systems



     Often the output from a continuous air monitoring device



is an analog trace on a strip chart.  Usually the strip



charts are cut at weekly intervals and are sent to the data




handling staff for interpretation.  A technician may estimate



by inspection the hourly average pollutant concentrations and



convert the analog percent of scale to engineering units,



e.g.  ppm.  He may also read daily maximum five minute or ten



minute concentrations from the chart.



     Reading strip charts is a tedious job subject to varying



degrees of error.  A procedure for maintaining a desirable



quality for data manually reduced from strip charts is im-



portant.  One procedure for checking the validity of the data



reduced by a technician is to have another technician or the



supervisor check the data.  Because the values have been



taken from the strip chart by visual inspection, some



difference in the values derived by two individuals can be



expected.  When the difference exceeds a nominal amount and



the initial reading has been determined to be incorrect, an



error should be noted.  If the number of errors exceeds a



predetermined number, all data for the strip chart are re-



jected and the chart is read again by a technican other than



the one who initially read the chart.  The question of how
                         6-9

-------
many values to check can be answered by applying acceptance




sampling techniques.



     6.4.2  Acceptance Sampling Applications




     Acceptance sampling can be applied to data validation




to determine the number of data bits (individual values on a




strip chart) that need to be checked to determine with a



given probability that all the data bits are acceptable.




The supervisor wants to know, without checking every data



value, if a defined error level has been exceeded.  From



each strip chart with N data values, the supervisor can




randomly inspect n data values.  If the number of erroneous



values is less than or equal to C, the rejection criteria,



the values for the strip chart are accepted.  If the number



of errors is greater than C the values for the strip chart



are rejected, and another technician is asked to read the



chart.



     The following discussion is meant only to explain how



acceptance sampling can be applied.  The procedure is



relatively complex, and several types of acceptance sampling



can apply.  A procedure for any specific application should



be developed by a competent individual who understands the



statistical derivation of acceptance sampling.



     Values of n (the number of data values to be checked)



and C (the maximum number of errors that is acceptable) are



selected to insure a high probability of acceptance of all
                         6-10

-------
the strip chart values if the error rate is P,  or less.   A




low probability of acceptance is insured if the error rate




is P? or greater.  The probability of acceptance of all the




strip chart values is 1-a for an error rate of P, and 3 for




an error rate of P2.   Typical values of a are 0.01, 0.05, and




0.10, and typical values of 3 are 0.05, 0.10, and 0.20.   The




error rate is the percent of erroneous values.




     The probability levels a and 3 are subjectively




chosen.  To determine the probabilities to use, the risk of




accepting bad data or of rejecting good data must be con-




sidered.  If the risk is small the value for the a (the




acceptance probability) and 3 (the rejection probability)




may be set at 0.10 and 0.20 respectively.  Decreasing the




values of a and 3 increases, the size :'of the sample required




to be checked.




     Another problem is determining values for P,  and P»




 (the acceptable and the non-acceptable error rates).  The




acceptable error rate could be ten percent.  If the number




of erroneous values exceeds ten percent of all values checked,




then the data for the entire strip chart could theoretically




be rejected.  From a practical standpoint, however, if the




error rate is eleven percent, the strip chart may  still  be




acceptable.  The P, value represents the error rate that the




supervisor is trying to maintain.  The P~ value represents




the maximum error rate that can be tolerated.  As  the
                         6-11

-------
difference between the acceptable error rate and the non-


acceptable error rate increases for small sample sizes, e.g.


ten or less, the acceptance and rejection probabilities


change significantly.

     A technique that can be applied to determine the effect


of varying n (the number of individual values checked) and

C (the maximum number of erroneous values allowed)  is that


use of an operating characteristic (OC) curve.  The OC


curve (Figure 6.1)  gives the probability of acceptance for

various error rates.   For the following example values of P


between 0.02 and 0.20 are used.

     Consider that the choices of sample size, n, are 10,


25, and 50.  Consider the rejection criterion to be 2, i.e.,


if more than 2 errors occur, the strip chart is rejected.

The next step is to compute the probability of 2 or fewer

errors for all P values from 0.02 to 0.20.


     The probabilities of 0, 1, or 2 errors for each sample

size n and each error rate P can be evaluated from the binomial

distribution.  Since the values of P being considered are

small, the Poisson distribution can be used to approximate

the binomial distribution.  The cumulative probability

curves for the Poisson distribution can be approximated by
                                            O £
use of a chart developed by Dodge and Romig.

     The OC curves for the values of n and C are shown in

Figure 6.1.  When n = 50 the probability of acceptance, if
                         6-12

-------
   Probability of Acceptance
Probability of(Rejection
1.0
                        0.0
 40
 20
                                                               0.20
                                                               0.40
                        0.60
                        0.80
                                                               1.00
        02  .04  .06  .08  .10  .12  .14  .16  .18  .20   .22
                               Error Rate  •
         Figure 6.1  OC Curve
                               6-13

-------
the error rate is P = 0.04, is approximately 0.68.  But  for



the same sample size of 50, if the error rate is P = 0.02, the




probability of acceptance if the error rate is P = 0.10  is



approximately 0.12.  If the sample size is n = 10  (and the




acceptance criterion remains at C = 2) the probability of



acceptance if P = 0.04 is approximately 0.99, and  if P = 0.10



it is still approximately 0.92.  So when the sample size



decreases from 50 to 10 the test cannot discriminate well



between error rates of 0.04 and 0.10.



     It is apparent that establishing an acceptance plan



for maintaining a defined quality of data is difficult.



Various types of sampling plans are available.  A  quality



control expert should be consulted before an acceptance



sampling plan is adopted.



     6.4.3  Sequential Analysis



6.4.3.1  Test Procedure - The typical approach used in



performing a statistical test of hypothesis requires the



collection of a sample of a fixed size.   A statistic is



then computed from the sample data and compared with some



critical values for that statistic.   A decision is then made



to accept the hypothesis (H )  or to accept some alternative



hypothesis (H,).   With such a procedure it is necessary to
                          6-14

-------
collect the specified sample of observations regardless of



the results that may be obtained from the first few observations,





     A procedure called sequential analysis requires that a



decision be made after each observation is made.  The possible



decisions to be made are;



     1. Accept H .
            c   o


     2. Accept H,.



     3. Continue the sampling process.





This procedure has  the advantage that, on the average, fewer



observations are required to reach a decision than would be



the case with a fixed sample size.





     Sequential analysis is readily adaptable to acceptance



sampling related to checking the error rate of an analyst



responsible for reducing data from strip charts.  Suppose



that a random sample of data values read by one analyst is



checked for validity by another (and probably more experienced)



analyst.  Each time the difference between two such readings



exceeds some nominal value an error is said to have occurred



in the original value.  Suppose further that it is desired to



reject a set of data only 5 percent of the time if the error



rate of the analyst is less than 0.05 and accept the set of



data only 10 percent of the time if the error rate is greater



than 0.15.
                         6-15

-------
     H   :  P = P  < 0.05
      o         o —




     Hl  :  P = Pl - °'15



     a = 0.05




     B = 0.10





     On the basis of the above information it is possible



to construct the graph shown in Figure 6.2.  The parallel



lines define the critical regions for the three possible



decisions,  one of which must be made after each observation.



The actual observational data is plotted on the graph as a



step function (i.e. a line is drawn one unit to the right if



a data value is determined to be valid and one unit up if



the value is declared to be incorrect).  The sampling process



is stopped when the step function crosses either parallel



line.





     Suppose the following results are obtained when data



values read by an analyst are checked for validity:
                         6-16

-------
I
t-1
•^1
                 e

                 i
 C
 O
•H
-P
 (0
(U
U)
X!
O

-P
O
Q)
J-i
In
O
U
C
M

M-l
O
Q)
       15
      10
                                      Accept  H,

                                      P = 0.15
                                                                         Continue

                                                                         Sampling
                                       15
                                                                  20
25
                           Number of Valid Observations  - gm
                                                                       Accept  HQ
                                                                          =  0 . 0 5
30
35
             Figure 6.2   Sequential Test for the  Error Rate of a Data Analyst

-------
                  Observation      Result
                     I               g
                     2               g
                     3               g
                     4               b
                     5               g
                     6               b
                     7               g
                     8               g
                     9               g
                    10               g
                    11               g
                    '12               b
                    13               g
                    14               g
                    15               g
                    16               g
                    17               g
                    18               g
                    19               g
                    20               b
                    21               g
                    22               g
                    23               g
                    24               g
                    25               b
                    26               g
                    27               g
                    28               b
                    29               g
                    30               b
                    31               g
                    32               g
                    33               g
                    34               b
                    35               g
The results can be plotted on Figure 6.2 by drawing a line

one unit to the right if the data value is good  (g) and one

unit up if the data value is bad  (b).  After only a few

observations it appears that the final outcome will be to

reject the hypothesis that the analyst true error rate is

equal to or less than 0.05.  Indeed after the 30th observation

(i.e. g  + d  =30, where g  = 23 and d  = 7) the sampling


                        6-18

-------
process would be halted and the decision would be to accept



H  (i.e. reject the hypothesis that the error rate is



equal to or less than 0.05) indicating that the error rate



of the analysis is not acceptable.





     The equations of the parallel lines shown in Figure  6.2



are:



     1.099 d  - 0.111 g  = 2.890                        (1)
            m         3m


     1.099 d  - 0.111 g  = -2.255                       (2)
            m          m


     where d  = number of errors out of m observations
            m


           g  = number of valid values out of m observations
            m




     The general form of equations  (1) and  (2) is given by




     dm ln F1> 9m ln 1^= ln ^                     (3)
            o            o



           P          1-P

and  d  In ^ + g  In T-T-i = In -=-§—                     (4)
      m    P    ym    1-P       1-cc
            o            o





     where P  = 0.05 = acceptable error rate



           P, = 0.15 = excessive error rate



           a  = 0.05 = probability of rejecting the  set of



           data when the true error rate is <_ 0.05



           6  = 0.10 = probability of accepting the  set of



           data when the true error rate is > 0;.15
                         6-19

-------
     In = natural logarithm of the term following
     Substituting the values shown above in equations  (3)




and  (4) yields;



     rl  In °-15 + „  In l~° ' 15 - In l~° ' 10

     dm ln 0705 + gm ln T^OTOS ~ ln TT05-
        i  °-15 j.    i  1-0-15   ,  0.10

      m ln 0705 + gm ln l^OTOS = ln I^OT
     The method of sequential analysis is based upon the



computation of the sequential probability ratio P, /pom-



The denominator of this ratio, P   is the probability that
                                om


the m observations would occur if hypothesis H  were true.



Similarly P,  is the probability that the m observations



would occur if hypothesis H-, were true.  The test procedure



followed is as follows:
             p

     -i   . ,   1m <    , B          . „
     1.  if —= --    :-TiL-  , except H

             Pom      l~a
            p


     2.  If ^  >• ^-  , except HI

             om
     3.  If T—  <  ^-- < —-  , take: another observation.
            1-a     P      a
                     om
                         6-20

-------
     If the error rate of the analyst is P,, the probability



of getting exactly d  errors and g  valid data values out of
                    m            3m


a set of m observations is
     Plm = pl m d-Pi)"1                                W


     On the other hand if the error rate of the analyst is
             d        g

     P   = P  m (1-P )  m
      om    o       o
     The sequential probability ratio is
     P,   I  P, \dm
      1m     1 \
      om
1-P.
Jm
                                    (9)
                                                        (10)
and the decision to accept H  shown above becomes
         m
            1-P.
            1-P
                  m
  - 1-a
                                                        (11)
Taking the natural logarithm of  (11) yields
                       1-P.
     dm ln
9m ln -I=W^  -
          <  In
                1-a
                              (12)
from which equation  (4) is obtained by using  the  equality



(i.e. the critical value).  Equation  (3)  follows  in  a



similar manner.
                         6-21

-------
6.4.3.2  Operating Characteristic - The operating characteristic


(OC) function is the function which defines the probability  of


acceptance when the error rate for the data analyst is P.  For


the example discussed above, four points on the OC curve  are


known;




               Error Rate     Probability of Acceptance


                   (P)            (OC Function)


                    0                  1


                  0.05          1-a = 0.95


                  0.15            8 = 0.10


                    1               0



     An additional point between P  and P,  can be plotted


for the proportion

                    1-P
                 In 	±
                    1-Po
                    1-P,       P,
                 In	  - In pi
                    1-P         o
                       o
and the corresponding probability of accepting P' as



                   In —^-
                       a .
           Pr(P') = 	=-3	5	
                   In ^_P _ In  P
                       a       1-a
                         6-22

-------
Substituting P ,  P,, a and 3 for the .example yields,





                n   1-0.15
           p, = _   1-0.05
                ,   1-0.15 _    0.15

                ln 1-0.05   ln oToT
              = 0.09


                   1-0.10
                In
       Pr(P') = -   °-°5
                ,   1-0.10   ,   0.10
                In —r—^-=— - In
                    0.05       1-0.05



              = 0.56








     The OC curve for the example problem is presented in



Figure 6.3-As can be seen if the actual error rate of the



analyst is equal to or greater than 0.15 the probability of



acceptance with this test procedure is small (<0.10).



If, however the actual error rate of the analyst is about 0.10



the probability is about 0.5 that this test procedure will



accept the set of data as having an acceptable error rate.







6.4.3.3  Average Sample Number (ASN) - It is obvious from



Figure 6.2, that the sampling process to determine if



the analyst is maintaining an acceptable error rate will



terminate after a few observations if the error rate is



much greater than P, or much smaller than P .  On the other
     J             1                       o


hand if the error rate is near or between P, and P  it may
                         6-23

-------
QJ
U

C
      1.00
      0.80
0)
o
o
0.60
0
•H
rH
•rH

•a
XI
o
0.40
     0.20
                   0.2
                        0.4
0.;6
0.8
                                                            1.0
                               Error Rate - P
           Figure  6.3   Operating Characteristic  Curve
                                6-24

-------
be necessary to check a large number of data values to



ascertain whether or not the analyst  is performing at



a satisfactory level.  Based on the example, if all of



the data values that are checked are incorrect, the check-



ing process will terminate after only 3 observations.



Likewise if all values are valid the checking process will



be stopped after 16 observations .





     The average sample number  (ASN) for the example problem
for P  = 0.05 is given by
                        In   - + a In
            ASN = 	      ^ "         a
                  Po ln P1 +  (1-po) ln
                         o
                = 40




Similarily for P, = 0.15 the ASN is given by
            ASN = B ln T^ +  (1-B) ln ^
                        P              1-P

                  P  In  ± +  (1-P ) In ,-
                         o                0




                - 33





6.4.3.4  Observations in Groups - There may be  situations  when



it is more efficient  (from the standpoint of  the  collection of



a sample of observations) to make several observations  at  a



time and then record  the results in  a manner  similar  to that
                        6-25

-------
                                                               (4)
for the example problem above.  It can be shown mathematically


that the ASN will be increased by an amount equal to the


number of observations in each group.  Thus the ASN's reported


for P  and P,,  above would be increased by 5 to 45 and 38


respectively.
                        6-26

-------
 7.0   REFERENCES

 1.    Bauer,  E.L.,  A Statistical Manual  for  Chemists.

 2.    Bennett,  C. A.,  and  Franklin, N. L., Statistical Analysis
      in Chemistry  and The Chemical Industry, John Wiley  and
      Sons, New York,  1954.

 3.    Duncan, A.  J., Quality  Control  and Industrial  Statistics,
      Richard D.  Irwin,  Inc.,  Homewood,  Illinois, 1959.

 4.    Dixon,  W. J., and Massey, F.J., Jr., Introduction to
      Statistical Analysis, McGraw Hill  Book Co.,Inc., New
      York,  1957.

 5.    Snedecor, G.  W., and Cochran, W. G., Statistical Methods,
      The Iowa  University  Press, Ames, Iowa, 1967.

 6.    International Electrotechnical Commission, Technical
      Committee 66, Working Group 6 on Electronic Measuring
      Devices of  Air and Water Pollution.

 7.    Office  of Air Programs,  "Field  Operations  Guide  for
      Automatic Air Monitoring Equipment", Office of Air
      Programs  Publication No. APTD-0736, EPA,  Research
      Triangle  Park, North Carolina,  1971.

 8.    Intersociety  Committee,  APHA, Methods  of Air Sampling
      and Analysis, American  Public Health Association,  1015
      18th Street,  N.W., Washington,  D.C., 1972.

 9.    "Part  50  -  National  Primary and Secondary  Ambient Air
      Quality Standards",  Federal Register,  36 FR 22 384,
      November  25,  1971.

10.    Grant,  E.L.,  Statistical Quality Control,  McGraw Hill
      Book Co., Inc.,  New  York, 1969.

11.    Morrow, N.L., "Sampling and Analyzing  Air  Pollution
      Sources", Chem.  Eng.,pp . 85-98, January 4, 1972.

12.    PEDCo-Environmental  Specialists, Inc., "Administrative
      and Technical Aspects of Source Sampling for Par-
      ticulates", Contract No. CPA-70-124, EPA,  1971.

13.    Shigehara,  R.T., et.al., "Significance of  Errors in
      Stack  Sampling Measurements", Paper 70-35, APCA, St.
      Louis,  Missouri, June 14-18,  1970.
                           7-1

-------
14.   American Society for Testing and Materials,  "Standard
      Method for Sampling Stacks for Particulate Matter",
      ASTM Method D 2928-71.

15.   Devorkin,  H., "Source Testing Manual",  Los Angeles
      County Air Pollution Control District,  Los Angeles,
      California, 1963.

16.   Price, L.W.,  "Maintenance of Laboratory Instruments",
      "Laboratory Practice",  summary of LABEX International
      1969 meeting, under leadership of L.W.  Price,  Bio-
      chemistry  Dept., University of Cambridge.

17.   U.S. Environmental Protection Agency,  Handbook for
      Analytical Quality Control in Water and Wastewater
      Laboratories, June, 1972.

18.   American Public Health  Association, Standard Methods
      for the Examination of  Water and Waste  Water,  13th
      Edition, 1971.

19.   American Society for Testing and Materials,  "Water;
      Atmospheric Analysis",  1971 Annual Book of ASTM
      Standards, Part 23, November, 1971.

20.   Nelson, G.O., Controlled Test Atmospheres  Principles
      and Techniques, Ann Arbor Science Publishers,  Inc.,
      Ann Arbor, Michigan, 1971

21.   Jacobs, M.B., The  Chemical Analysis of  Air Pollutants,
      Interscience  Publishers, Inc., New York, 1960.

22.   U.S. Dept. of Commerce, National Bureau of Standards,
      "NBS Special  Publication 260".

23.   Bayse, David, "Quality  Control in Laboratory Manage-
      ment", U.S. Department  of Health, Education  and
      Welfare, Center for Disease Control, Licensure and
      Development Branch, Proficiency Testing Section,
      Atlanta, Georgia.

24.   Varian Techtron Pty. Ltd., "Water Analysis by  Atomic
      Absorption Spectroscopy", Varian Techtron, Ltd.,
      1972.

25.   Decker, C.E., and  Smith, W.S. "Determination of
      Fluorides  in  Stack Gas: SPADNS - Zirconium Lake
      Method". U.S. Dept. HEW, National Center for Air
      Pollution  Control, Cincinnati, Ohio.
                            7-2

-------
26.    Dodge,  H.F.,  and Romig,  H.G.,  Sampling Inspection Tables,
      John Wiley and Sons,  Inc.,  New York,  1944.

27.    Reference 3,  pp. 365-366

28.    Reference 3,  p.  886

29.    Reference 3,  p.  367
                             7-3

-------
                       APPENDIX A

          STATISTICAL FORMULAE AND DEFINITIONS
ARITHMETIC MEAN
                   n
                   Z Xi
              X  = 1
                     N
              X = mean value
              Xi = individual value in the sample
              N = number of values in the sample
RANGE
              R = X max - X min

              X max = maximum value in the sample
              X min = minimum value in the sample
STANDARD DEVIATION
                 \Z  (X - X)2
                "
                   N -. 1

              X = individual value in the sample
              X = average of all values in the sample
              N = number of values in the sample

COEFFICIENT OF VARIATION   (also referred to as relative  standard
                           deviation)

              CV =  S  (100)
                    X
              S = standard deviation
              X = mean value
                         A-l

-------
CONTROL LIMITS FOR CONTROL CHARTS
     RANGE CHART
                (27)
              Upper Control Limit = UCL = D.R

                  R = average range for a set of replicates
                  D. = adjustment factor

                  o^ = standard deviation of the relative range
                  d_ = an adjustment factor used to estimate
                       the standard deviation of a universe

              Lower Control Limit = LCL = D_R

                  R  = average range for a set of replicates
                  D- = adjustment factor
                  D, = 1 - 3 o'w/d,
                   •J              £•
COEFFICIENT OF VARIATION CHART(28)
              Upper Control Limit = UCL = B^ CV
                  CV = average coefficient of variation for
                       a set of replicates
                  B. = adjustment factor
                  B4 = 1 + K
                  N = number of values in a sample
                  c  = an adjustment factor used to estimate
                       the standard deviation of a universe
                         A-2

-------
              Lower Control Limit = LCL=B CV


                  CV = average coefficient of  variation
                       for a set of replicates
                  N = number  of  values  in  a sample

                  c2 = an adjustment  factor used  to  estimate
                       the standard deviation  of  a universe
CONTROL CHART FOR AVERAGES

     Upper control limit =  UCL =  x  +  A-R

     Lower control limit =  LCL =  x  -  A»R

     *2 =  -^-

     N = number of values in a sample

     d = an adjustment factor used  to estimate  the  standard
         deviation of a universe.
                         A-3

-------
                       APPENDIX B
   TABLES OF FACTORS FOR CONSTRUCTING CONTROL  CHARTS*


                          TABLE BI
              FACTORS FOR CONSTRUCTING CONTROL
                     CHARTS FOR AVERAGES
Observations in                   Factors  for  Control Limits
Sample, n                                      A.,
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
1.880
1.023
0.729
0.577
0.483
0.419
0.373
0.337
0.308
0.285
0.266
0.249
0.235
0.223
0.212
0.203
0.194
0.187
0.180
0.173
0.167
0.162
0.157
0.153
                          B-l
  *The  factors  in  these  tables  are for the 99% confidence interval.

-------
                         TABLE BII

              FACTORS FOR CONSTRUCTING CONTROL
                      CHARTS FOR RANGE
Number of Observations
                                Factors for Control Limits
in Sample, n
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
"3
0
0
0
0
0
. 0.076
0.136
0.184
0.223
0.256
0.284
0.308
0.329
0.348
0.364
0.379
0.392
0.404
0.414
0.425
0.434
0.443
0.452
0.459
"4
3.267
2.575
2.282
2.115
2.004
1.924
1.864
1.816
1.777
1.744
1.716
1.692
1.671
1.652
1.636
1.621
1.608
1.596
1.586
1.575
1.566
1.557
1.548
1.541
                          B-2

-------
                        TABLE Bill
            •FACTORS FOR CONSTRUCTING CONTROL CHARTS
                 FOR COEFFICIENT OF VARIATION
Number of Observations
in Sample, n
 Factors  for  Control Limits

i,                       B,
2 0
3 0
4 0
5 0
6 0.030
7 0.118
8 0.185
9 0.239
10 0.284
11 0.321
12 0.354
13 0.382
14 0.406
15 0.428
16 0.448
17 0.466
18 0.482
19 0.497
20 0.510
21 0.523
22 0.534
23 0.545
24 0.555
25 0.565
3.267
2.568
2.266
2.089
1.970
1.882
1.815
1.761
1.716
1.679
1.646
1.618
1.594
1.572
1.552
1.534
1.518
1.503
1.490
1.477
1.466
1.455
1.445
1.435
                         B-3

-------
              TABLE BIV
THE t DISTRIBUTION  (TWO-TAILED TESTS)
Degrees of
Freedom
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Probability of
0.050
12.707
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
a Larger Value, Sign Ignored
0.010
63.657
9.925
5.841
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
             B-4

-------
                       APPENDIX C

     CALIBRATION CURVES FROM REGRESSION ANALYSIS BY THE
                 METHOD OF LEAST SQUARES



E.I  Constructing the Calibration Curve

     The calibration curve presented in Figure C-l shows

the relationship between the concentration of sulfur dioxide

in the sample and absorbance (i.e. the instrument response).

The calibration curve was determined on the basis of a least

squares regression analysis in which the relationship between

concentration (X) and absorbance  (Y) is assumed  to be of  the

form;

          Y = a + bX  .

  where   a = intercept  (i.e. the point at which the line

                         crosses the Y axis)

          b = slope  (i.e. the change in absorbance per unit

                         change in concentration)

     Equations  for determining a and b are:
                x =
                     N         N     N
                    NZ X.Y. -  EX.  ZY.
                    NZ X  -   (Z
                a = Y - bX
                     N       N
                     Z Y.   bZ X.
                  "   N       N

  where N = 12  (the number of samples analyzed)  as  shown in

  Table C-l.
                            C-l

-------
                        TABLE C-l
         CALIBRATION DATA FOR S02 DETERMINATION
X
.20
.20
.20
.60
.60
.60
1.00
1.00
1.00
1.40
1.40
1.40
ZX =
X =
Y
.095
.080
.123
.305
.329
.355
.559
.560
.590
.780
.810
.790
9.6 ZY
.8 Y
x2
.04
.04
.04
.36
.36
.36
1.00
1.00
1.00
1.96
1.96
1.96
= 5.376
= .448
Y2
.009025
.0064
.015129
.093025
.108241
.126025
.312481
.313600
.348100
.6084
.6561
.6241
EXY =

XY
.0190
.0160
.0246
.1830
.1974
.2130
.559
.560
.590
1.092
1.1340
1.106
5.694

        ZX2= 10.08  ZY2= 3.220626
     Substituting the appropriate summation from Table C-l:
                b = 12 (5.694) -  (9.6) (5.376)
                       12(10.08) -  (9.6)2
                b = 0.5805
                    5.376    (0.5805) (9.6)
                a =
                      12          12
                a = -0.0164
     The equation for the calibration line is:
                  Y = -0.0164 + 0.5805X
     To plot this straight line (Figure C-l) two points  in
the XY plane are necessary.  One point is the Y intercept,  that
is when X = 0, Y = -0.0164.  A second point can be selected  for
any other value of X, for example, X = 1.2.  Then
                Y = -0.0164 + 0.5805 (1.2)
                Y =  0.6802
                           C-2

-------
                                          LU_L_U
                                          I  | !! I  I ' i
                                          I  | i i I  | I
                                       ! ! !  I M I  i I
                                            I i I  i i
                                                       I I I  I I !
                                                       L i LUJ.
                                                        L.L ' i_ JjJ
                 l.LLLLL! i_
                 III   I  ! 1 I
                 I i r!" -"i -i-i
                             :-  u i :L:J:
                 L!_J_!_UJlU
                 i n~n
                 M	
                 i i i i_n TTI
                 rrrm—
                 M  I  II i I
                 ! I  I I ! M I I
                 " '    ~  i i i
                    i  | l  | i_i
                  i  i r i n i i
                 i I  I I I !  I I i
                 _-_LJ_LJ_J_L!
                 LLDlLLLLi
                    i i I I  I
                    m
                                                                I  | I t I  i ! M
                                                                         "
                 LLLIJ
                 MIM
             ...I !I.!M _j_U
             ±!_LJTl | H _LLJ

                         UCL

                         99%
                                                                                !_U_LU_!J_i
                                                                                i  MJ_!_M_!_I
                                                                                l  I M i  i i i  i
                                                                                  I II M i M
                                                                                  rQIuirn
                              M M i i I  i
                              i u  U
                                                                               _LJlLJ- iZLLL!
                                                                               _L_LlJL.Li.J_!_!
                                                                                J !_:_!_]_!_! Jj
                                                                               JZ1 i ! !  i !' I -1
                                                                                 i i~T"i i  1 ! i  i
   .LLLJ
  .I J_L.L
                                               i  i h i—h-Y =  -0.0164 +  .5805X
                        jj
                        J.JJ
                    1  IJ_LI_!
                              I  Ml
                             / II l  I I M !
                _i_L 'L.L- _L>J./
__:	i  |_ |   ! M
                                                j i 1
                                        CONCENTRATION
                                            | , i .  ! ! i i  I I !  l I |._!  I i I   I ! i  • ' l I  !
               Figure  C-l
               Calibration Curve for S02 Determination
                                ri1 !~n  ri n
  12 D705

' O THE INCH
                                                             KEUFFEL. ft ESSER CO.

                                                                 MADE IN U .E A.

-------
E.2  Constructing Control Limits for the Calibration Curve

     The control limits for the calibration curve are deter-

mined from the standard error of estimate Syx.  The standard

error of estimate is a measure of the spread of the responsive

variable (absorbance) about the regression line.  Its value

can be determined from:
          Sy-x =
                  N-2
N      N      N
EY. - aZY. - bEX.Y.
          Syx = -T~3- 220626-(-0. 0164) (5. 376)-(0. 5805) (5
                                   .694)1
               = 0.00585

     Regression theory is based on the assumption that, for

each value of X, the values of the response variable Y are

normally distributed about the point on the straight line.

Then a region of ±2.58 Syx units on either side of the regres-

sion line should contain 99 percent of all of the measured values

of Y.  The limits of this region are then the UCL and LCL for

determining whether or not future calibration values are in

control .

     For any specific methodology, a clear plastic overlay could

be constructed showing the control lines for the calibration

curve.  Each time a control sample is run, and the analytical

value is plotted against the calibration curve, the process is

judged to be in control if the point lies between the control

lines of the overlay.
                          C-4

-------