Tbe procedures described in these standardsare intended fr widest possible application, hut when chemical sludgesor slurries
the examination of waters of a wide range of quality, including or other sampiesof highly unusual composition are encountered,
wateTsuitable fr domestic or industrial supplies, surface wateT, the methods of this manual may require modification or may be
ground wateT, cooling or circulating wateT,bOiler wateT, bOiler inappropriate.
feed wateT,treated and untreated municipal or industrial waste Most of the methodsincluded hefe have been endorsedby reg
wateT,and saline wateT.Tbe unity of the fjelds of wateTsupply, ulatory agencies. Procedural modification without formal ap
receiving wateTquality, and wastewatertreatmentand disposal is proval may be unacceptableto a regulatory body.
recognizedby presenting methodsof analysisfr eachconstituent Tbe analysis of bulk chemicals received fr wateTtreatment is
in a single section fr all types of waters. not included herein. A committee of the American WateTWorks
An effort has been made to present methods that apply gen Association prepares and issues standards fr wateT treatment
erally. Where alternative methods are necessaryfr sampIesof chemicals.
different composition, the basis fr selectingthe most appropriate Part 1000 contains information that is common to, or useful in,
method is presentedas clearly as possible.However, sampieswith labOratoriesdesiring to produce analytical results of known qual
extreme concentrations or otherwise unusual compositions or ity, that is, of known accuracyand with known uncertainty in that
characteristicsmay presentdifficulties that preclude the direct use accuracy.To accomplishthis, apply the quality assurancemethods
of these methods. Hence, some modification of a proceduremay described herein to the standardmethods described elsewherein
be necessaryin specific instances.Whenever a procedureis mod this publication. Other sections of Part 1000 addresslabOratory
ified, the analyst should state plainly the nature of modification equipment, labOratory safety, sampling procedures,and method
in the report of results. development and validation, all of which provide necessaryin
Certain proceduresare intended fr use with sludgesand sed formation.
iments. Here again, the effort has been to presentmethods of the
1010 B. Statistics
1. NormalDistribution analyst can only estimatethe standarddeviation becausethe num
her of observationsmade is finite; the estimate of 0" is denoted
If a measurementis repeated many times under essentially by s and is calculated as folio ws:
identical conditions, the results of each measurement,x, will be
distributed randomly abOut a mean value (arithmetic average) be s = [I(x::i)2/(nl)]'fz
causeof uncontrollable or experimental error. If an infinite num
her of such measurementswere to be accumulated,the individual Tb ta dard d . tI. fi th . ,or spread,oef th normaI
values would be dlstributed
... . . to those shown In
In a curve sIffillar . es n
distribution andeVIa
alsoon xes aefixed
includes WIdthfraction of the values mak
Figure 1010:1. Tbe left curve illustrates the Gaussianor normal ing up the ~urve. For example 68.27% of the measurementslie
distribution, ~h.ich is describedprecisely by the mean, ~,~d ~e between IJ. :t 10",95.45%betw'eenlJ.:t 20",and 99.70% between
~~dard deVIatIon,0". The mean,.~r average,of the dIstributIon IJ. :t 3<T.It is sufficiently accurateto state that 95% of the values
IS sImply the sum of all values divIded by the number of values are within :t 20"and 99% within :t 30".When values are assigned
so summed, i.e., IJ. = (k;x:;)/n. Because no measurementsare to the :t 0" multiples, they are confidence limits. For example, 10
repeatedan infinite number of times, an estimate of the mean is :t 4 indicates that the confidencelimits are 6 and 14, while values
made, using the same summation procedure hut with n equal to from 6 to 14 representthe confidence interval.
a finite number of repeatedmeasurements(10, or 20, or. . .). Tbis Another useful statistic is the standard error of the mean, <T,..,
estimateof IJ.is denoted by x. Tbe standarddeviation of the nor which is the standarddeviation divided by the squarefOOtof the
mal distribution is defined as 0" = [k(xlJ.)z/n]l/z. Again, the number of values, or <TIm.Tbis is an estimateof the accuracyof
12 INTRODUCTION (1000)
'Mode
,
I 'Median
I I
Mean
Median
'(;' Mode '(;'
c , c
Q) Q)
~ ~
co co
Q) Q)
' '
L1. : L1.
Class Class
the mean and implies that another sampie from the same popu 3. Rejection of Data
lation would have a mean within some multiple of this. Multiples
of this statistic include the same fraction of the values as stated Quite orten in aseries of measurements,Olle or more of the
above fr 0". In practice, a relatively small number of average results will differ greatly from the other values. Theoretically, no
values is available, so the confidence intervals of the mean are result should be rejected, becauseit may indicate either a faulty
expressedas x :t Islm where 1 hag the following values fr 95% technique that castsdoubt on all results or the presenceof a true
confidence intervals: variant in the distribution. In practice, reject the result of any
analysis in which a known error.has occurred. In environmental
studies, extremely high and low concentrationsof contaminants
n t n t may indicate the existenceof areaswith problems or areaswith
2 12.71 5 2.78 no contamination, so they should not be rejected arbitrarily.
3 4.30 10 2.26
4 3.18 ~ 1.96
TADLE1010:1.CRITICALVALUESFR5% AND 1% TESTSOF
DlSCORDANCYFR A SINGLE0UrL1ER IN ANORMAL SAMPLE
The use of 1 compensatesfr the tendency of a small number.of
values to underestimateuncertainty. For n > 15, it is common to Numberof Critical Va1ue
use 1 = 2 to estimate the 95% confidence interval. Measurements
S .ll th ... th I . d d d .. I n 5% 1%
ti ano er statistic IS e re atlve stan ar evlatlon, 0" IJ.,
with its estimate slx, also known as the coefficient of variation 3 1.15 1.15
(CV), which commonly is expressedas a percentage.This statistic 4 1.46 1.49
normalizes the standarddeviation and sometimesfacilitates mak 5 1.67 1.75
ing direct comparisonsamong analysesthat include a wide range 6 1.82 1.94
of concentrations.For example, if analysesat low concentrations 7 1.94 2.10
yield a result of 10 :t 1.5 mg/L and at high concentrations100 :t 8 2.03 2.22
8 mg/L, the standarddeviations do not appearcomparable.How 9 2.11 2.32
ever, the percent relative standarddeviations are 100 (1.5/10) = 10 ;.~~ ;.~~
1?:o and ~OO(8/1~) = ~%,whichindicate thesmallervaria~; 2:37 .2:66
bll1ty obtalned by usrng thlS parameter. 15 2.41 2.71
2. LogNormal Distribution
 16
18
2.44
2.50
2.75
2.82
20 2.56 2.88
In many casesthe results obtained from analysis of environ 30 2.74 3.10
mental sampieswill not be normally distributed, i.e., a graph of 40 2.87 3.24
the data will be obviously skewed, as shown at fight in Figure 50 2.96 3.34
1010:1, with the mode, median, and mean being distinctly differ 60 3.03 3.41
T b . I I di .b . th u1 100 3.21 3.60
et. 00 tarn
logarithms andathen
near y norma
calculate x andstrl
s.utlon, convert
Theantilogarithmse res ts to
of these 120 327
..
366
two values are estimates of the geometric mean and the geometric Source:BARNETf.V. & T. LEWIS.
19~. Outliersin StatisticalData.JohnWiley &
standard deviation, Xg and Sg. Sons.New York. N.Y.
(1010)/Glossary
INTRODUCTION 13
An objective test fr outliers has been described.\ If a set of ments, n, then the XH or XL is an outlier at thai level of signifi
data is ordered from low to high: XL, X2 . . . XH, and the average cance.
and standard deviation are calculated, then suspected high or low Further information on statistical techniques is available else
the statistic T:
4. References
T = (XH 
T = Ci' 
:i)ls fr a high value, or
xJIs for a low value. I. BARNE1T,V. & T. LEWJS.1984. Outliers in Statistical Data. lohn Wiley
& SOng, New York, N.Y.
, 2. NATREu..A,M.G. 1963. Experimental Statistics. National Bur. Stan
Second, compare the value of T with the value from Tlible dards Handbook 91, Washington, D.C.
1010:1 fr either a 5% or 1% level of significance. If thecalcu 3. SNEDECOR, G.W. & W.G. COCHRAN. 1980. Statistical Methods. Iowa
lated T is larger than the table value fr the number of measure State University Press,Ames.
1010 C. Glossary
1. Definition of Terms Level of quantitation (WQ)/minimum quantitation level
The purpose of this glossary is to defme concepts, not regula sufficiently greater than the blank thai it can be detected
tory terms; it is not intended to be allinclusive. within specified levels by good laboratories during routine
Accuracycombination of bias and precision of an analytical operating co~ditions. Typically it is the concentratio~ thai
procedure, which reflects the closeness of a measured value to produces a sIgnal lOs above the reagent wateT blank sIgnal.
Biasconsistent deviation of measured values from the true specifically herein refers to duplicate sampies, i.e., two sampies
value, caused by systematic errors in a procedure. laken at the same time from ODe location.
Calibration check standardstandard used to deterrnine the state Internal standarda pure compound added to a sampie extract
of calibration of an instrument between periodic recalibrations. jus.t be~ore instrumental analysis to permit correction fOT inef
confidence limits. outside agency, used to measure the bias in a procedure. For
Confidence intervalset of possible values within which the true certain constituents and matrices, use National Institute of Stan
value wililie with a specified level of probability. dards and Techno~ogy (NIST) Standard Reference Materials
Confidence limitone of the boundary values defining the con when they are avallabie.
Detection levelsVarious levels in increasing order are: analyses of a sampie, usually expressed as the standard devi
noise ratio of the instrument. This is similar, in many orat~ry measurements by use of data from internal and external
mean of blank analyses. This sets both Type I and Type 11 odology to assure that the process is in control.
errors at 5%. Other names fr this level are "detection Random errorthe deviation in any step in an analytical proce
level" and "level of detection" (LOD). dure thai can be treated by standard statistical techniques.
Method detection level (MDL)the constituent concentration Replicaterepeated operation occurring within an analytical pro
thai, when processed through the complete method, pro cedure. Two or more analyses fr the same constituent in an
duces a signal wirb a 99% probability that it is different extract of a single sampie constitute replicate extract analyses.
from the blank. For seven replicates of the sampie, the mean Surrogate standarda pure compound added to a sampie in the
must be 3.14s above the blank where s is the standard de laboratory just before processing so that the overall efficiency
viation of the seven replicates. Compute MDL from repli of a method can be deterrnined.
cate measurements Olle to five times the actual MDL. The Type I erroralso called alpha error, is the probability of decid
MDL will be targeT than the LLD because of the few rep ing a constituent is present when it actually is absent.
lications and the sampie processing steps and may vary with Type II erroralso called beta error, is the probability of not
1020 A. Introduction
This sectionapplies primarily to chemical analyses.SeeSection laboratory,4 as weil as an indication that managementhas made
9020 fr quality assuranceand control fr microbiologiqal anal a commitrnent to assurethat the quality systemsdefined in the
yses. QA manual are implemented and followed at all times.
Quality assurance(QA) is the definitive program fr laboratory In the QA manual,clearly specify and documentthe managerial
operation that specifies the measuresrequired to produce defen responsibility, authority, quality goals, objectives, and commit
sible data of known precision and accuracy.This program will be met to quality. Write the manual so that it is clearly understood
defmed in a documentedlaboratory quality system. and ensuresthat all laboratory personnel understandtheir roles
The laboratory quality system will consist of a QA manual, and responsibilities.
written procedures, work instructions, and records. The manual Implement and follow chainofcustody proceduresto ensure
should include a quality policy that defines the statisticallevel of that chain of custody is maintainedand documentedfr eachsam
confidence used to expressthe precision and bias of data, as weil pIe. Institute proceduresto permit tracing a sampie and its deriv
as the method detection limits. Quality systems, which include atives through all steps flom collection through analysis to re
QA policies and all quality control (QC) processes,rollSt be in porting final results to the laboratory's client and disposal of the
place to document and ensurethe quality of analytical data pro sampie.Routinely practice adequateand complete documentation,
duced by the laboratory and to demonstratethe competenceof which is critical to assuredata defensibility and to meet laboratory
the laboratory. Quality systems are essential fr any laboratory accreditation/certificationrequirements,and ensurefull traceabil
seekingaccreditationunder stateor federallaboratory certification ity fr all tests and sampies.
programs. Included in quality assurance are quality control Standard operating procedures(SOPs) describe the analytical
(Section 1020B) and quality assessment(Section 102OC). See methods to be used in the laboratory in sufficient detail that a
Section 1030 fr evaluation of data quality. competent analyst unfamiliar with the method can conduct a re
.. liable review and/or obtain acceptable results. Include in SOPs,
1. Quallty Assurance Planmng where applicable, the following items25: tide of referenced, con
. sensustest method; sampie matrix or matrices; method detection
Estab~ISha QA program and pr~pare a QA manual or pl.an. level (MDL); sco and a lication; summary of SOP; defini
Include m the QA manual and assocIateddocumentsthe followmg .. . ~
pp . .. .
. 14. h .th al . . qual.Ity po1ICY
. tions,mterferences, safetyconsIderations, wagtemanagement, ap
Items . cover .s eet
. WI . t d 1.
al approv taff sIgnatures, tu
.b.l . . al Para s, equIpmen ,an supp leg; reagents and sn; ta dards sample
statement;
organlzatlOn structure;
s responsl I Itles; an yst . preservation, .. shlpment, ..
tr. . d " . " ed b th collection, and storage requlfements, spe
ammg an pellormance requlfements; tests pellorm y e ... . .
.
1aboratory; procedures "lor handlmg d .. 1 cIfic quallty control practlces,frequency, acceptancecntena, and
an edrecelvmg samp es; . ed corrective . action. I.f acceptance . . are not met; c 
1 tr 1 d d tati. d requlf cntena al
samp e con 0 an ocumen on proc ures; proce ures .. ..'
. . . . lbratlon and standardlzation; detaIls on the actual test procedure,
fr achleVlng . traceabIlIty of measurements, major equIpment, m mc1uding samp1e
. . al 1 . al.fi . d
strumentation, and reference measurement standards used; stan .'.
Preparatlon. c cu ations. qu I lcations an
'.
P er
dard operating procedures (SOPs) fr each analytical method; formance requlfements fr analysts (mcluding number and type
proceduresfr generation,approval, and control of policies and of analyses);data assessment/da.ta
~anagement; references;and
procedures; procedures fr procurement of reference materials any tables, ~~wcharts, ~d validation or method performance
and supplies; proceduresfr procurementof subcontractors'serv data..~t a mImmum,valldatea n~w SO~~~foreuseby f~st de
ices; internat quality control activities; proceduresfr calibration, tenru~l~g th~ MDL and perfonrung ~ I~Itial demonstration of
! verification, and maintenanceof instrumentation and equipment; capabIlIty usmg relevant regu~atory~uldellnes. .
: dataverificationpracticesincluding interlaboratorycomparison Use an~documentp:eventlvemaInten~ceproced.ures fo~ m
: and proficiencytestingprograms;procedures to be followedfr strumentatlonand.equlpment:An effectlveprev~ntlvem~nt~
i feedback and corrective action whenevertesting discrepanciesare nance pro?ram wIll r~uce mstrument ~alfunctions, maIntam
! detected; procedures fr exceptions that permit departure from ~ore consIstentcallbratlon, be cost~~ffectlve,a.ndreduce.down
; documentedpolicies; proceduresfr system and performanceau time. Include measurementtraceabillty to National inStitute.of
dits and reviews; proceduresfr assessingdata precision and ac Standardsand Technology (NIST) StandardReferenceMatenals
!I, c., curacy and determining method detection limits; proceduresfr (SRMs) or commercially available reference materials certi~ed
i} data reduction, validation, and reporting; proceduresfr records ~acea~leto NIST SRMs in ~e qA manual or SOP to establlsh
i.: archiving; proceduresand systemsfr control of the testing en mtegnty of the laboratory callbration and measurementprogram.
il vironment; and proceduresfr dealing with complaints from users Formulate documentcontrol procedures, which are essential to
fi of the data. Also define and include the responsibility fr, and dat~defensibility, t~ co.ver.thecompleteprocessof ~~ument g~n
I t: frequency of, managementreview and updatesto the QA manual eratlon, approval, distribution, storage,recall, archIvmg, and dIS
t and associateddocuments. posal. Maintain logbooks fr each test or procedure performed
J,: On the title page,includeapprovalsignaturesand a statement with completedocumentation
on preparation
andanalysisof each
11; that the manual hag been reviewed and determined to be appro sampie,including sampie identification, associatedstandardsand
r priate fr the score, volume, and range of testing activities at the QC sampies,method reference,date/time of preparation/analysis,
QUALITY ASSURANCE (1020)/Qualily Control 15
analyst, weights and volumes used, results obtained, and any oratory Accreditation Conference,2nd Interim Meeting, Bethesda,Md.
problems encountered. Keep logbooks that document maintenance [available online]. U.S. Environmental Protection Agency, Washing
and calibration tor each instrument or piece of equipment. Cali ton, D.C.
bration procedures, corrective actions, internat quality control ac 4. lNTE~AnONAL OROANIZATION FORSTAND~RDlZAnoN. !99~. General
tivities, performance audits, and data assessments tor precision Req~Irementstor the. Competenceof Testing an.dCallbration Labo 
d (b . ) d. d . S ti. 1020Ban.d C ratones, ISO/IEC Gulde 25Draft Four. International Org. tor Stan
an accuracy las are lscusse m ec ons dard. ti. G S ',1
D ed . al .d . d . . lza on, eneva, Wltzer.an. d
ata r ucnon, v I anon, an reporting are the final steps m 5. U.S. ENVIRONMENTAL PROTECllON AOENCY.1995. Guidance tor the
the datageneration process. The data obtained from an analytical Preparationof StandardOperating Procedures(SOPs) tor QualityRe
instrumentmustfirst be subjectedto the datareductionprocesses latedDocuments.
EPAQA/G6,Washington,
D.C.
described in the applicable SOP before the final result cah be
obtained. Specify calculations and any correction factors, as weIl 3. Bibliography
as the steps to be followed in generating the sampie result, in the
QA manual or SOP. Also specify all of the data validation steps DELFINO,J.J. 1977. Quality assurancein water and wastewater analysis
to be followed before the final result is made available. Report laboratories. Water Sew. Works 124:79.
results in standard units of mass, volume, or concentration as lNHORN,S.L., ed. 1978. Quality AssurancePracticestor Health Labora
specified in the method or SOP. Report results below the MDL tories. American Public Health Assoc., Washington, D.C.
in accordance with the procedure prescribed in the SOP. Ideally, STA~Y, T.W. & s..S. VE~. 1983. Interim .Guidelinesand Specifica
include a statement of uncertainty with each result. See references tions tor Prep~ng QUallty Assw:anceProJectPlans: EPA600/483
and bibliography for other useful information and uidance on 004, U.S. Envlfonmental Protection Agency, Washmgton, D.C. .
. . ..g U.S. ENVIRONMENTAL PROTECllON AOENCY.1997. Manual tor the Certi
a QA programanddevelopmgan effective
estabhshmg QA man fi cati.on 0f Laborat ones
. Anal yzmg
. D nn. ki ng W ater. EPA  815  B  97 
ual. 001, U.S. Environmental Protection Agency, Washington, D.C.
lNTERNAnoNAL OROANIZATIONFR STANDARDIZAT1ON. 1990. General Re
2. References quirements tor the Competence of Testing and Calibration Labora
tories, ISO/IEC Guide 25. International Org. tor Standardization,
1. STANLEY, T.T. & S.S. VERNER. 1983. Interim Guidelines and Specifi Geneva, Switzerland. .
cations tor Preparing Quality Assurance Project Plans. EPA600/483 U.S. ENViRONMENTALPROTEcnONAOENCY. 1994. NationaiEnvironmental
004, U.S. Environmental Protection Agency, Washington, D.C. Laboratory Accreditation Conference (NELAC) Notice of Confer
2. QUALrrY SYSTEMSCOMMITfEE, NAnoNAL ENVIRONMENTALLABORA ence and Availability of Standards. Federal Register 59, No. 231.
TORYACCREDITATIONCONFERENCE. 1996. National Environmental Lab U.S. ENVlRONMENTALPROTEcnONAOENCY. 1995. Good Automated Lab
oratory Accreditation Conference, 2nd Annual Meeting, Washington, oratory Practices. U.S. Environmental Protection Agency, Research
D.C. [available online]. U.S. Environmental Protection Agency, Wash Triangle Park, N.C.
ington, D.C. AMERICAN AsSOClAnON FORLABORATORYACCREDITAnoN. 1996. General
3. QUALrrY SySTEMS COMMlTJ'EE, NAnONAL ENVIRONMENTALLABORA Requirements tor Accreditation. A2LA, American Ass~. Labora
TORYACCREDITATlONCONFERENCE. 1997. National Environmental Lab tory Accreditation, Gaithersburg, Md.
the method of choice, specify and reference the method or pro results with considerationto reagentblank results are as follows:
i cedure used fr demonstratingcapability.
. If the reagent blank is less than the MDL and sampie results
2. Ongoing Demonstration of Capability are greater than the MQL, then no qualification is required.
. If the reagent blank is greater than the MDL hut less than the
The ongoing demonstration of capability sometimesreferred MQL and sampieresults are greaterthan the MQL, then qualify
to as a "laboratory control sampie or lab~ratory control stan the results to indicate that analyte was detected in the reagent
dard," "quality control check sampie," or "laboratoryfortified blank.. .
blank," is used to ensure that the laboratory remains in control . If the reagentblank IS greaterthan the MQL, further corrective
during the period when sampiesare analyzed, and separ~teslab action and qualification is required.
oratory performancefrom method performanceon the sampiema
trix. See ~ 5 below fr further details on the laboratoryfortified 5. LaboratoryFortified Blank
blank. Preferably obtain this sampie from an externat source(not
the same stock as the cali~ration st~~ds). Analyze QC check A laboratoryfortified blank is areagent wateTsampieto which
sampieson a quarterly basIs,at amInImum. a known concentrationof the analytesof interest has been added.
A LFB is used to evaluate laboratory performance and analyte
3. Method Detection Level Determination and Application recovery in a blank matrix. As a minimum, include OlleLFB with
each sampie set (batch) or on a 5% basis, whichever is more
Determine the method detection level (MDL) fr each analyte frequent. The definition of a batch is typically methodspecific.
of interest and method to be used before data flom any sampies Processthe LFB through all of the sampie preparation and anal
are reported, using the proceduredescribedin Section 103OC.As ysis steps. Use an added concentration of at least 10 times the
a starti?g ~int fr dete~ning the con~entrationt~ use in MDL MDL, the midpoint of the calibration curve, or other level as
~ete~ation, use an estimate of. fiv~ times the.estI~ated detec specified in the method. Preparethe addition solution from a dif
tion lImIt. Perform MDL determmatIonsas an Iterative process. ferent referencesourcethan that usedfr calibration. Evaluate the
If calculate.d.MDL is not withi,n a ~actorof 10of th~ value fr the LFB fr percent recovery of the added analytes. If LFB results
kno.wnaddition, repeatdete~na~ons at a more sultable concen are out of control, take corrective action, including repreparation
tratI~n. Conduct MDL deterlnlnations at least annu~y (or other and reanalysis of associatedsampiesif required. Use the results
specified frequency) fr ~ach analyte and.m~thod muse .at the obtained fr the LFB to evaluate batch performance, calculate
laboratory. Perform or ven~ M?L deterlnlnatI~nfr eachmstru recovery limits, and plot control charts (see ~ 12 below). Refer
ment. Perform MDL deterlnlnatIonsover a penod of at least 3 d to the method of choice fr specific acceptancecriteria fr the
fr each part of the procedure. Calculate recoveries fr MDL LFB
sampies.Recoveriesshould be between 50 and 150% and %RSD .
values $ 20% or repeat the MDL determination.Maintain MDL
and IDC data and have them available fr inspection. 6. LaboratoryFortified Matrix
Apply the MDL to reporting sampie results as follows:
A laboratoryfortified matrix (LFM) is an additional portion of
. Report results below the MDL as "not detected." a sampie to which known amountsof the analytes of interest are
. Report results between the MDL and MQL with qualification added before sampie preparation.The LFM is used to evaluate
fr quantitation. analyte recovery in a sampiematrix. As aminimum, include Olle
. Report results above the MQL with a value and its associated LFM with each sampie set (batch) or on a 5% basis, whichever
error. is more frequent. Add a concentration of at least 10 times the
MRL, the midpoint of the calibration curve, or other level as
4. Reagent Blank specified in the method to the selectedsample(s).Preferably use
the same concentration as fr the LFB to allow the analyst to
Areagent blank or methodblank consistsof reagentwater (See separatethe effect of matrix from labo~atoryperformance.Prepare
Section 1080) and all reagentsthat normally are in contact with the LFM from a reference source dIfferent from that used fr
a samp1eduring the entire ana1ytica1 procedure.The reagentblank calibration. Make the addition suchthat sampiebac~groundlevels
is used to determine the contribution of the reagentsand the pre do not adversely affect the recovery (preferably adJustLFM con
parative analytical stepsto error in the measurement.As a mini centrations if the known sampie is above five times the back
mum, include one reagentblank with each sampie set (batch) or ground level). For example,if the sampiecontains the analyte of
on a 5% basis, whichever is more frequent. Analyze a blank after interest, make the LFM sampie at a concentration equivalent to
the daily calibration standardand after higWy contaminatedsam the concentrationfound in the known sampie.Evaluate the results
pIes if carryover is suspected.Evaluate reagent blank results fr obtained fr LFMs fr accuracyor percent recovery. If LFM re
the presenceof contamination. If unacceptablecontamination is sults are out of control, take corrective action to rectify the effect
! presentin the reagentblank, identify and eliminate sourceof con or use another method or the method of standardaddition. Refer
tamination. Typically, sampie results are suspectif analyte(s) in to the method of choice fr specific acceptancecriteria for LFMs
the reagent blank are greater than the MQL. Sampies analyzed unti1the laboratory developsstatistically valid, laboratoryspecific
with an associatedcontaminatedblank must be repreparedand performance criteria. Base sampiebatch acceptanceon results of
reanalyzed.Refer to the method of choice fr specific acceptance LFB analysesfalber than LFMs alone, becausethe matrix of the
criteria fr the reagent blank. Guidelines fr qualifying sampie LFM sampie may interfere with the method performance.
"Jcc
QUAlITY ASSURANCE (1020)/Quality Control 17
7. LaboratoryFortified Matrix Duplicate/Duplicate Sam pie reanalysis if required. Refer to the method of choice tor specific
is collected, this second portion of sampie is added and processed 10. Calibration
collected to analyze a LfM duplicate, use an additional portion a. Instrument calibration: Perform instrument calibration, as
of an alternate sampie to obtain results tor a duplicate sampie to weIl as maintenance, according to instrument manual instructions.
gather data on precision. As aminimum, include one LFM du Use instrument manufacturer's recommendations tor calibration.
plicate or one duplicate sampie with each sampie set (batch) or Perform instrument performance checks, such as those tor GC/
on a 5% basis, whichever is more frequent. Evaluate the results MS analyses, according to method or SOP instructions.
obtained tor LfM duplicates tor precision and accuracy (precision b. Initial calibration: Perform initial calibration with a mini
alone tor duplicate sampies). If LfM duplicate results are out of mum of three concentrations of standards tor linear curves, a min
control, take corrective action to rectify the effect or use another imum of five concentrations of standards tor nonlinear curves, or
method or the method of standard addition. If duplicate results as specified by the method of choice. Choose a lowest concen
are out of control, reprepare and reanalyze the sampie and take tration at the reporting limit, and highest concentration at the up
additional corrective action as needed (such as reanalysis of sam per end of the calibration range. Ensure that the calibration range
pIe batch). Refer to the method of choice tor specific acceptance encompasses the analytical concentration values expected in the
criteria tor LfM duplicates or duplicate sampies until the labo sampies or required dilutions. Choose calibration standard con
ratory develops statistically valid, laboratoryspecific performance centrations with no more than one order of magnitude between
criteria. If no limits are included in the method of choice, cal concentrations.
culate preliminary limits from initial demonstration of capability. Use the following calibration functions as appropriate: response
Base sampie batch acceptance on results of LFB analyses rather factor tor internal standard calibration, calibration factor tor ex
than LfM duplicates alone, because the matrix of the LfM sam ternal standard calibration, or calibration curve. Calibration curves
pIe may interfere with the method performance. may be linear through the origin, linear not through the origin, or
MS, some GC analyses, and some metals analyses by ICP/MS. If response factors or calibration factors are used, the.calculated
An internat standard is an analyte included in eachstandard and %RSD tor each analyte of interest must be less than the method
added to each samiJle or sampie extractidigestate just before sam specified value. When using response factors (e.g., tor GC/MS
pIe analysis. Internat standards should mimic the analytes of in analysis), evaluate the performance or sensitivity of the instru
terest hut not interfere with the analysis. Choose an internal stan ment tor the analyte of interest against minimum acceptance val
dard having retention time or mass spectrum separate from the ues tor the response factors. Refer to the method of choice tor
analytes of interest and eluting in a representative area of the the calibration procedure and acceptance criteria on the response
chromatogram. Internat standards are used to monitor retention factors or calibration factors tor each analyte.
time, calculate relative response, and quantify the analytes of in If linear regression is used, use the minimum correlation co
terest in each sampie or sampie extractidigestate. When quanti efficient specified in the method. If the minimum correlation co
fying by the internal standard method, measure all analyte re efficient is not specified, then a minimum value of 0.995 is rec
sponses relative to this internat standard, unless interference is ommended. Compare each calibration point to the curve and
suspected.
If internalstandardresultsareout of control,takecor recalculate.If anyrecalculated
valuesarenot within the method
rective action, including reanalysis if required. Refer to the acceptance criteria, identify the source of outlier(s) and correct
method of choice tor specific internat standards and their accep before sampie quantitation. Alternately, a method's calibration
radiochemistry analyses. Surrogates and tracers are used to eval titation of the analytes of interest in sampies. Use calibration ver
uate method performance in each sampie. A surrogate standard is ification, described in the next section, only tor checks on the
pound(s) unlikely to be found in environmental sampies, such as when the instrument is set up and whenever the calibration ver
tluorinated compounds or stahle, isotopically labeled analogs of ification criteria are not met.
the analytes of interest. Tracers are a different isotope of the an c. Calibration verification: Calibration verification is the peri
alyte or element of interest. Surrogates and tracers are introduced odic confirmation by analysis of a calibration standard that the
to sampies before extraction to monitor extraction efficiency and instrument performance hag not changed significantly from the
percent recovery in each sampie. If surrogate or tracer results are initial calibration. Base this verification on time (e.g., every 12 h)
out of control, take corrective action, including repreparation and or on the number of sampies analyzed (e.g., after every 10 sam
I
QUALITY ASSURANCE (1020)/Quality Control 19
UCL
40
UCL
110                  UWL
,
J  .                  UWL
'.
Q)
::0
0 100. . .. ..
~
E 35
. u...
& ... .. .. .'x
g
.
.. .
. X
~...

1
+"'.
. 90
.
C
Q)
g
0
()
30 .. .
 LWL
ro
~
~
Q)
g 80
LWL
LCL
0
()
LCL
25 70
Date Date
curacy chart includes upper and lower warning levels (WL) and range so that the analyst need only subtractthe two results to plot
upper and lower controllevels (CL). Common practice is to use the value on the precision chart. The mean range is computed as:
:t 2s and :t 3s limits tor the WL and CL, respectively, where s
represents standard deviation. These values are derived from R= DoS
a.
ca
I, .
U
aJ aJ
::3 aJ > > In
'"" 0 0
I
, L' > .c .c aJ
, 0 In
40 .c '" '"
~~
'0 :i?:o
I 0
Q)
~ .In
'" aJ
.In>
aJ
.>In
'"
C
'"
I:: !!l:. ~ ~ ~
ca
CI: !!R '"
c'" :.
()
() :.
()
() .
In
In
+' In ' In aJ
> aJlX
()
20 ~d :r:s; lO~ ()~
c. aJ 
OaJ

0(1)
:.
In .c0
aJ
a:5 (')5 :r.l 00'"
:!::!: :!:
f
tt
r
0
100 200 300 400 500 35.
1! ~
..
:..:
.... ..
!! I
...
I
.. I..
I CL
;r Concentration,
mgiL 25...: ...:: : : . :.. WL
t,
0'
>CL I IIII .I III 15
.
"".'
"'.".
. ..
. :.. :.:
~
...
.
.
.:.
. ..
:.
.'
,
<CL
<WL . ""
. . .. . . ... . . ..,.
X
Date
X
. .. ..
Figure 1020:3. Rangechart tor variable concentrations. Figure 1020:5. Meanscontrol chart with outofcontrol data (upper half).
QUALITYASSURANCE (1020)/Quality Control 111
The above considerations apply when the conditions are either TABLE1020:11.EXAMPLE
DATAQUALIFIERS
above or below the centralline, but not on both sides, e.g., tour
f fi 1 ed ' . Symbol Explanation
0 Ive va ues must exce elther + Is or Is. Mter correctmg
the problem, reanalyze the sampies analyzed between the last in B Analyte found in reagentblank. Indicates possible
control measurement and the outofcontrol Olle. reagelit or background contarnination.
Another important function of the control chart is assessment E Reported value exceeded calibration range.
of improvements in method precision. In the accuracy and pre J Reported value is .an e~ti~ate because conc.entration is
Cl.sl' on charts l.f meas t I eed th WL less than reportmg hffilt or because certam QC
, uremen s never or rare y exc e , ' t .
al 1 t th WL d CL . th 0 cn ena were no t met.
rec
points. cu Trends
a e e in precision
an usmgbe detected
can e I to sooner
20 mostif runnirig
recenr data
av N Organic constituents
is needed. tentatively identified . Confinnation
erages of 10 to 20 are kept. Trends indicate systematic error; PND Precision not detennined.
random error is revealed when measurements randomly exceed R Sampie results rejected because of grass deficiencies
warning or controllimits. in QC or method perfonnance. Resampling and/or re
analysis is necessary.
13. QC Evaluationtor SmallSampie Sizes RND Recoverynot deterrnined,
U Compound was analyzed rar, hut not detected.
Small sampie sizes, such as for fjeld blanks and duplicate sam * Based on V.S. Environmental Protection Agency guidelines.'
14. Corrective Action . If a second LFB fails, check an independent reference material.
If the second source is acceptable, reprepare and reanalyze af
Quality control
a trendareevidenceof unacceptable
data outside
erroi:in the analyticalproc
the acceptance limits or exhibiting
. If a LFM falls, checkLFB. If the LFB IS acceptable,quallfy
fected sampl~(s). . .
ess. Take corrective acti!l promptly to deterrnine and eliminate the data fr the LFM .s~ple or use another method or the
the source of the error. Do not report data until the cause of the method of standard add~tIon. .
problem is identified and either corrected or qualified. Example . If a LFM and the assoclated LFB fall, reprepare and reanalyze
U.S. En
. Check to Sf;e if sample(s) was prepared and analyzed according vironmental Protection Agency, Washington, D.C. . ,
to the approvedmethodandSOP.If it wasnot prepareand/or 2. U.S.ENVIRONMENTAL
PROTECfION
AGENCY.1997. 304h Streamlmmg
analyze again. ' Proposal Rule, Federal Register, March 28, 1997 (15034).
. . .. 3. U.S. ENVIRONMENTAL PRTECrION AGENCY.1994. National Func
. Check callbratIon . standards.. agamst an Inde pe ndent
, standard or. ti.ona I GUl.de Ifies
. "lor I norganlc. Data ReVlew.
. EPA  5401R  94013 ,
refe~ence matenal. If cahbrati~n stan?ards fall, reprepare Call U.S. Environmental Protection Agency, Contract Laboratory Pro
bratIon standards and/or recallbrate Instrument and reanalyze gram, Office of Emergency and Remedial Response, Washington,
affected sample(s). D.C.
112 INTRODUCTION (1000)
Quality assessment is the process used to ensure that quality TABLE 1020:ill. AUDrr OF A SOlL ANALYSIS PROCEDURE
control measures are being performed as required and to deter Proced C t R ks .
mine the quality of the data produced by the laboratory. It m
. ure ammen emar
cludes such items as proficiency sampIes, laboratory intercom I. Sampie entered into yes lab number assigned
parison sampIes, and performance audits. These are applied to logbook
test the precision,accuracy,anddetectionlimits of the methods 2. Sampieweighed yes dry weight
in use, and to assess adherence to standard operating procedure 3. Orying procedure no rnaintenance of oven
requirements. followed not clone
l the analysts. For routine procedures, semiaQnual analyses are 5, Management Review
I
i:
I :
Complianceauditsareconductedto evaluate
oratory meets the applicable requirements
whetherthe lab
of the SOP or consen
results,input from endusercomplaints,andcorrectiveactions.
j can be. used to d~ument the manne! m WhlCh a sampIe IS treated JARVIS, A.M. & L. Sru. 1981. Environmental Radioactivity Laboratory
t from ~me of re~el~t to final reportmg ~f ~e result. The goal of Intercomparison Studies Prograrn. EPA600/481004, U.S. Environ
j comphance audIts IS to detect any devlatlons from the SOP or mental Protection Agency, Las Vegas, Nev.
t consensus
methodso thatcorrectiveactioncanbe takenon those INTERNAnONAL
ORGANlZAnON FOR STANDARDlZAnoN. 1990. General Re
, : deviations. An
Table lO20:ill.
example format fr achecklist is shown in qu~ments fr the ~ompetence of ~esting and Calibration La~ra
tones, ISO/IEC Gulde 25. International arg. fr StandardlZatlOn,
I ' Geneva,
Switzerland.
.. 4. Laboratory Quality Systems Audits AMERICAN Soc1ETYFR TESTINGAND MATERIALS. 1996. Standard Practice
..1. for Detennination of Precision and Bias of Applicable Test Methods
A quality systemsauditprogramis designedandconductedto of Committee
019onWater.ASTM0277796,
Arnerican
Society
addressall programelementsandprovidea reviewof the quality Testing& Materials,
WestConshohocken,
Pa,
,~
DATA QUALITY (1030)/Measurement Uncertainty 113
1030 A. Introduction
The role of the analyticallaboratory is to produce measure its usable information decrease. Reporting tools, such as detection
mentbased information that is technically valid, legally d~fensi or quantitation limits, frequently are used to establish a lower limit
ble, and of known quality. Quality assurance is aimed at opti on usable information content.
mizing the reliability of the measurement process. All Laboratory data may be used fr such purposes as regulatory
measurements contain error, which may be systematic (with an monitoring, environmental decisionmaking, and process control.
unvarying magnitude) or random (with equal probabiJity of being The procedures used to extract information fr these different
positive or negative and varying in magnitude). Determination of purposes vary and may be diarnetrically opposed. For exarnple, a
the systematic and randOn1 error components of an analytical measurement fr regulatory monitoring may be appropriately
method uniquely defines the analytical performance of that qualified when below the detection level because the error bar is
method.1 Quality control (QC) procedures identify and control relatively large and may preclude a statistically sound decision.
these sources of error. Data collected over aperiod of time, however, may be treated by
statistical methods to provide a statistically sound decision even
1. Measures of Quality Control when many of the data are below detection levels}
Random error (precision) and systematic error (bias) are two 3. The Analyst's Responsibility
routine indicators of measurement quality used by analysts to as
sess validity of the analytical process. Precision is the closeness The analyst must understand the measures of quality control
of agreement between repeated measurements. A measurement and how to apply them to the'data quality objectives of process
has acceptable precision if the random errors are low. Accuracy control, regulatory monitoring, and environmental field studies. It
is the closeness of a measurement to the true value. A measure is important that the quality objectives fr the data be clearly
met is acceptably accurate when both thesystematic and random defined and detailed before sampie analysis so that the data will
errors are low. QC results outside the acceptancelimits, as set by be technically correct and legally defensible.
the data quality objectives, are evidence of an analytical process
that may be out of control due to determinant errors such as con 4. Reference
tarninated reagents or degraded standards.
1. YOUDEN, W J. 1975. Statistical Manual of the Association of Official
2. Measurement Error and Data Use Analytical Chemists. Assoc. Official Analytical Chemists, Arlington,
Va.
Measurement error, whether random or systematic,reducesthe 2. OSBORN,K.E. 1995.You Can't Computewith LessThans.Wafer En
usability of laboratory data. As a measuredvalue decreases,its vironmentLaboratorySolutions,WaterEnvironmentFederation,Al
relative error (e.g., relative standard deviation) mayincrease and exandria, Va.
a traceable value, or a consensusvalue. The purpose of the sub assumedto be zero becauseany nonzero component is part of
stitutiqn may be tor convenience or because the measurement bias, by definition. The population standarddeviation, O"E,canbe
processthat produced 1"* hag less bias or variation than the Olle usedto characterizethe random componentof measurementerror
that produced M. For example, based on the average of many because the critical values of the normal distribution are weIl
measurements,avessei might be thought to contain 1"* ==50 jl,gJ knownandwidely available.For example,about95%of thenor
L of galt in wateT.It then may be sampledand routinely measured, mal distribution lies within the interval jl, :t 20"E.Hence, if there
resulting in a reported concentration of M = 51 jl,g/L. The actual is no measurementbias, and measurementerrors are independent
concentration may be T = 49.9 jl,g/L, resulting in E = 51  49.9 and normally distributed, M :t 20"E (95% confidence, assumed
= 1.1 jl,g/L. normal) is a suitable war to report a measurementand its uncer
To generalizethe natureof uncertainty, measurementerror may tainty. More generally, normal probability tables and statistical
be negligible or large in absoluteterms (i.e., in the original units) software give the proportion of the normal distribution and thus
or relative terms (i.e., unitless, E + T or 1"*). The perceived the % confidence gained that is contained within :t kO"Efr any
acceptability of the magnitudeof an absoluteerror dependson its value of scalar k.
intended use. For example, an absoluteerror of 1.1 jl,g/L may be Usually, however, the population standarddeviation, O"E,is not
inconsequentialfr an application where any concentration over known and must be estimated by the sampie standard deviation,
30 jl,g/L will be sufficient. However, if it is to be used instead as SE.This estimate of the standarddeviation is based on multiple
a standardfr precision measurement(e.g., of pharmaceuticalin observationsand statistical estimation. In this case,the choice of
gredients), 1.1 jl,g/L too much could be unacceptable. the scalark must be basednot on the normal distribution function,
but on the Student's t distribution, taking into accountthe number
3. Uncertainty of degreesof freedom associatedwith SE.
Systematic error (B) is all error that is not random, and typi
Reported measurement uncertainty will contain the actual cally is equatedwith bias. Systematicerror also can contain out
measurementerror with astated level of confidence.For example, fight mistakes (blunders) and lack of control (drifts, fluctuations,
if M :t U is presentedas a 95% confidence interval, approxi etc.).3 In this manual, the terms "systematic error" and "bias"
mately 95% of the time, the measurementerror E will fall within are used interchangeably.
the range of :t U. Systematic uncertainty oftenis more difficult to estimate and
make useful than is random uncertainty. Knowledge about bias is
4. Bias likely to be hard to obtain, and once obtained it is appropriately
and likely to be exploited to make the measurementless biased.
Bias is the systematiccomponent of error. It is defmed as the If measurementbias is known exactly (or nearly so), the user can
signed deviation betweenthe limiting averagemeasuredvalue and subtract it troll M to reduce total measurementerror.
the true value being measuredas the number of measurementsin If measurementbias is entirely unknown, and could take on
the averagetends to infinity and the uncertainty about the average any value from a wide but unknown distribution of plausible val
tends to zero. For example, the reason the T = 49.9 jl,g/L galt ues, usersmayadopt a worstcaseapproachand report an extreme
solution is thought to be 1"* = 50 jl,g/L could be a bias, bound, or they may simply ignore the bias altogether. For ex
B = 0.1 jl,g/L.The "leftover" error, 1.1  0.1 = 1.0 jl,g/L, is ample, historical data may indicate that significant interlaboratory
the random component.This random component(also called sto biases are present, or that every time a measurementsystem is
chastic error) changeswith each measurement.The bias is fixed, cleaned, a shirt is observedin QC measurementsof standards.In
~d may be related to the laboratory method used to produce T.* the absenceof traceablestandards,it is hard fr laboratory man
Usually, a recognized method will be used to produce or certify agementor analyststo do anything other than ignore the potential
the traceable standard,a sampie with a certificate statingthe ac problem.
cepted true value T.* The method may be the best method avail The recommendedpractice is to conduct routine QAfQC meas
able or simply the most widely acceptedmethod. It is chosen to urementswith a suite of internat standards.Plot measurementson
have very low error, both bias and random. Such a traceablestan control charts, and when an outofcontrol condition is encoun
dard may be purchased from a standardsorganization such as tered, recalibratethe systemwith traceablestandards.This permits
NIST. the laboratory to publish a boundary on bias, assliming that the
underlying behavior of the measurementsystemis somewhatpre
5. Bias and Random Variation dictable and acceptably small in scale in between QAfQC sam
pling (e.g., slow drifts and small shirts).
Measurementerror, E, (and measurementuncertainty) can be
split into two components,random and systematic: 6. Repeatability, Reproducibility, and Sources of Bias and
Variation
E=Z+B
a. Sources and measurement:The sourcesof bias and varia
Random error, Z, is the component of the measurementerror bility in measurementsare many; they include sampling error,
thai changes from Olle measurementto the next, under certain sampie preparation,interferenceby matrix or other measurement
conditlons. Randorn rneasurernenterrors are assurnedto be in quantities/qualities, calibration error variation, software errors,
dependentand have a distribution, orten assumedto be Gaussian counting statistics,deviations from method by analyst, instrument
(i.e., they are normally distributed). The normal distribution of Z differences (e.g., chamber volume, voltage level), environmental
is characterizedby the distribution mean, jl" and standarddevia changes (temperature,humidity, ambient light, etc.), contamina
tion, O"E.In discussion of measurementerror distribution, jl, is tion of sampieor equipment(e.g., carryover and ambi~ntcontam
:
j
DATA QUALITY(1030)/Measurement
Uncertainty 115
ination), variations in purity of solvent, reagent,catalyst, etc., sta results if they are homogeneous.Treat factors varied in the study
bility and age of sampIe, analyte, or matrix, and warmnp or as random factors and assumethem to be independentnormal
cooldown effects, or a tendencyto drift over time. random variableswith zero mean.However, this assumptionoften
The simplest strategy fOTestimating typicaI measurementbias can be challenged, becausethe sampIe and possibly the target
is to measure a traceable (known) standard, then compute the populations may be small (they may even be identical), and there
difference between the measuredvalue M and the known value may be a question of "representativeness."For example, six lab"'
T, assumedto be the true value being measuied. oratories (or analysts,or instruments)may report usablemeasure
ments out of a total population of twenty capableof doing tandem
M  T=B+ Z ,mass spectrometryfOTa particular analyte and matrix. It is hard
, to know how representative
the six are of the twenty,especially
The uncertainty in the measurementof the traceable standard after a ranking and exclusion processthat can follow a study, and
is assumedto be small, although in practice there may be situa whether th~ bias~sof the t~enty are normally dis~buted ~prob
tions where this is not an appropriate assumption. If random ably not dlscernlble from SIXmeasurements,even lf the SIXare
measurementuncertainty is negligible (i.e., 2 ~ 0), the difference, representative).
M  T, will provide an estimate of bias (B). If random uncer It may be more appropriateto treat eachfactor with few, known
tainty is not negligible, it can be observedand quantified by mak factor values (i.e., choices such as laboratories) as fixed factors,
ing a measurementrepeatedly on the same test specimen (if the to use the statistical term. Fixed factorshave fIXed effects. That
measurementprocess is not destructive). This may be part of a is, each laboratory has a different bias, as might each analyst,
QA/QC procedure. each instrument, and each day, but these biasesare not assumed
b. Repeatability: As quantified by the repeatability standard to have a known (or knowable) distribution. Therefore, a small
deviation (URPT),repeatability is the minimal variability of a sampie cannot be usedto estimatedistribution parameters,partic
measurementsystem obtained by repeatedlymeasuringthe same ularly a standarddeviation. For example, assumin.gthat v~abl~s
specimenwhile allowing no controllable sourcesof variability to are random, normal, and have zero mean may be mappropnatem
affect the measurement.Repeatability also can be obtained by an interlaboratory roundro~in stud~. .lt ~ust be assumed~at
pooling sampIestandarddeviationsof measurementsof J different every.laboratory has some blas, but It IS.difficult to charactenze
specimens,as folIows: the blasesbecauseof laboratory anonymlty, the small number of
laboratories contributing usabledata, and other factors.
0" =
.g~J~:;
.L
.!. 0" 2
Becauseof theseconcernsabout assumptionsand the potential
ambiguity of itswith
definition,do not and
report reproducibility ulilessofit
Rn J /= I RPT./ is accompanied study design a list of known sources
To perform a measurementcapability study to assessmeasure erated (each set has ODerandom deviate fr each variable), and
met tlncertainty through systematicerror budgeting, proceed as the value of M is computed and archived. The archived distri
folIows; bution is an empirical characterizationof the uncertainty in M.
ldentify sourcesof bias and variation that affect measurement e. Sensitivity study (designedexperiment): If the identities and
error. This can be dne with a causeandeffectdiagram, perhaps distributions of sourcesof bias and variation are known and these
with source categoriesof: equipment, analyst, method (i.e., pro sources are continuous factors, but the functional form of the
cedure and algorithm), material (i.e., aspectsof the test speci relationship betweenthem and M is not known, an empirical sen
mens), and environment: sitivity study (i.e., MCS) can be conductedto estimate the low
Select sourcesto study, either empirically or theoretically,.Typ order coefficients (SM/SG) fr any factor G. This will produce a
ically, study sourcesthat are influential, that can be varied during Taylor series approximation to the SM, which can be used to
the MCS, and that cannot be eliminated during routine measure estimate the distribution of SM, as in 'J c above.
met. Select models fr the sources.Treat sourcesof bias as fixed f Random effectsstudy: This is the nestedMCS and variance
factors, and sourcesof variation as random factors. componentsanalysis describedin 'J 7 above.
Design and conduct the study, allowing (or requiring) the se g. Passive empirical (QA/QCtype data): An even more em
lected sources to contribute to measurementerror. Analyze the pirical and passiveapproachis to rely solelyon QA/QC or similar
data graphically and statistically (e.g., by regressionanalysis,AN data. The estimated standarddeviation of sampIe measurements
OVA, or variance components analysis). ldentify and possibly taten on many different days, by different analysts,using differ
eliminate outliers (observationswith responsesthat are rar out of et equipment, perhaps in different laboratories can provide a
line with the general pattern of the data), and leverage points useful indication of uncertainty.
(observationsthat exert high, perhapsundue, influence).
Refine the models, if necessary(e.g., basedon residual analy 9. Statements of Uncertainty
sis), and draw inferences fr future measurements.For random
effects, this probably would be a confidence interval; fr fixed Always report measurementswith a statement of uncertainty
effects, a table of estimatedbiases. and the basis fr the statement.
Develop uncertaintystatementsas follows:46
8. Other Assessments of Measurement Uncertainty lnvolve experts in the measurementprinciples and use of the
measurementsystem,individuals familiar with sampling contexts,
In addition to the strictly empirical MCS approachto assessing and potential measurementusers to generatea causeandeffect
measurementuncertainty, there are alternative procedures, dis diagram fr measurementerror, with sourcesof bias and variation
cussedbelow in order of increasing empiricism. (' 'factors' ') identified and prioritized. Consult literature quanti
a. Exact theoretical: Some measurement methods are tied fying bias and variation. If needed,conduct ODeor more meas
closely to exact firstprinciples models of physics or chemistry. urement capability studies incorporating those sourcesthought to
For example, measurementsystemsthat count or track the posi be most important. In some cases, Gage R&R studies may be
tion and velocity of atomic particles can have exact formulas fr sufficient. These studies will provide "snapshot" estimates of
measurementuncertainty basedon the known theoretical behavior bias and variation.
of the particles. Institute a QA/QC program in which traceableor internal stan
b. Delta method (law of propagation of uncertainty): If the dards are measured routinely and the results are plotted on X and
measurementresult can be expressedas a function of input var R control charts (or equivalent charts). React to outofcontrol
iables with known error distributions, the distribution of the meas signalson the control charts.In particular, recalibrate using trace
urement result sometimescan be computed exactly. able standardswhen the mean control chart shows a statistically
c. Linearized: The mathematicsof the delta method may be significant change.Use the control charts, relevant literature, and
difficult, so a linearized form of M = T + E may be usedinstead, the MCSs to develop uncertainty statementsthat involve both bias
involving a firstorder Taylor seriesexpansionaboutkey variables and variation.
that influence E:
10. References
(M + 8M) = T + 8M/8G, + 8M/8G2+ 8M/8G3+ ...
1. YOUDEN,W.J. 1972. Enduring values. Technometrics 14:1.
fr sources GI, G2, G3, etc. of bias and variation that are contin 2. HENRION, M. & B. FISCHHOFF. 1986. Assessinguncertainty in physical
uous variables (or can be represented by continuous variables). constant.Amer. J. Phys.54:791.
The distribution of this expression may be simpler to determine, 3. CURRIE,L. 1995. Nornenclature in evaluation of analytical rnethods
as it involves the linear combination of scalar multiples of the ' including detection and quantification capabilities. Pure Appl. Chem.
!
DATA QUALITY (1030)/Method Detection Level 117
1. Introduction MDL. From a table of the onesidedt distribution select the value
of t for 7  1 = 6 degreesof freedomand at the 99% level;
Detection levels are controversial, principally becauseof in this value is 3.14. The product 3.14 times s is the desired MDL.
adequatedefinition and confusion of terms. Frequently, the in Although the LOQ is useful within a laboratory, the practical
strumental detection level is used for the method detection level quantitation limit (PQL) has been proposed as the lowest level
and vice versa. Whatever term is used, most analysts agree that achievableamong laboratorieswithin specified limits during rou
the smallest amount that can be detected above the noise: in a tine laboratory operations.3The PQL is significant becausedif
procedure and within astated confidence level is'the detection ferent laboratories will produce different MDLs even though us
level. The confidence levels are set so that the probabilities of ing the same analytical procedures, instruments, and sampie
both Type I and Type n errors are acceptablysmall. matrices. The PQL is about five times the MDL and representsa
Current practice identifies severaldetectionlevels (see 101OC), practical and routinely achievabledetectionlevel with a relatively
each of which has a defined purpose. These are the instrument good certainty that any reported value is reliable.
detection level (illL), the lower level of detection (LLD), the
method detection level (MDL), and the level of quantitation 3. Description of Levels
(LOQ). Occasionally the instrument detection level is used as a
guide ~or dete~ining the MDL. The relationship among these Figure 1030:1 illustrates the detection levels discussedabove.
levels IS approxlmately illL:LLD:MDL:LOQ = 1:2:4:10. For this figure it is assumed that the signals from an analytical
2. GLASER,J.A., D.L. FOERST, J.D. McKEE,S.A. QUAVE& W.L. BUDDE. Colorado, December25, 1984).American Wafer Works Assoc., Den
1981. Trace analysesforwastewaters. Environ. Sci. Technol. 15:1426. ver, Col0.
I .