Professional Documents
Culture Documents
Estimation of Measurement Uncertainty Printout 2016-2
Estimation of Measurement Uncertainty Printout 2016-2
chemicalanalysis(analyticalchemistry)
course
Thisisanintroductorycourseonmeasurementuncertaintyestimation,specificallyrelatedto
chemicalanalysis.
IvoLeito,LauriJalukse,IrjaHelm
https://sisu.ut.ee/measurement 1
Tableofcontents
Courseintroduction
1. Theconceptofmeasurementuncertainty(MU)
2. Theoriginofmeasurementuncertainty
3. Thebasicconceptsandtools
3.1.TheNormaldistribution
3.2.Mean,standarddeviationandstandarduncertainty
3.3.AandBtypeuncertaintyestimates
3.4.Standarddeviationofthemean
3.5.Rectangularandtriangulardistribution
3.6.TheStudentdistribution
4. Thefirstuncertaintyquantification
4.1.Quantifyinguncertaintycomponents
4.2.Calculatingthecombinedstandarduncertainty
4.3.Lookingattheobtaineduncertainty
Selftest4.3
4.4.Expandeduncertainty
4.5.Presentingmeasurementresults
4.6.Practicalexample
5. Principlesofmeasurementuncertaintyestimation
5.1.Measuranddefinition
5.2.Measurementprocedure
5.3.Sourcesofmeasurementuncertainty
5.4.Treatmentofrandomandsystematiceffects
6. Randomandsystematiceffectsrevisited
7. Precision,trueness,accuracy
8. Overviewofmeasurementuncertaintyestimationapproaches
9. TheISOGUMModelingapproach
9.1.Step1Measuranddefinition
9.2.Step2Modelequation
Selftest9.2A
Selftest9.2B
https://sisu.ut.ee/measurement 2
9.3.Step3Uncertaintysources
9.4.Step4Valuesoftheinputquantities
9.5.Step5Standarduncertaintiesoftheinputquantities
9.6.Step6Valueoftheoutputquantity
9.7.Step7Combinedstandarduncertainty
9.8.Step8Expandeduncertainty
9.9.Step9Lookingattheobtaineduncertainty
Selftest9B
10. Thesinglelabvalidationapproach
10.1.Principles
10.2.Uncertaintycomponentaccountingforrandomeffects
10.3.Uncertaintycomponentaccountingforsystematiceffects
10.4.Roadmap
10.5.DeterminationofacrylamideinsnacksbyLCMS
Selftest10.5A
11. Comparisonoftheapproaches
12. Comparingmeasurementresults
13. Additionalmaterialsandcasestudies
13.1.Differentanalyticaltechniques
13.2.DissolvedoxygenbyWinklermethod
13.3.CoulometricKFtitration
14. TestsandExercises
Frequentlyaskedquestions
Whatourparticipantssay?
Pleasecomment!
https://sisu.ut.ee/measurement 3
Courseintroduction
ThiscoursewillbeofferedagainasaMOOCinspring2016.Thecoursewill
takeplaceduringMarch28May8,2016.Registrationisopen.
Ifyouwanttoknowhowthecoursewentinspring2014and2015pleaseseethepresentation at
theEuroanalysis2015conference.
Courseintroduction
http://www.uttv.ee/naita?id=17710
https://www.youtube.com/watch?v=r34Ygzf62Y
The course was offered as an online course (MOOC) in Moodle environment during March 03
April 13, 2014 and March 02 April 12, 2015. Altogether more than 700 participants from more
than 70 countries have participated. Altogether more than 40% of them successfully completed
thecourseandwereawardedacertificatefromUniversityofTartu.
ThecoursewillbeofferedagaininMarch28May08,2016.
This course has been described in the paper: I. Leito, I. Helm, L. Jalukse. Using MOOCs for
teaching analytical chemistry: experience at University of Tartu. Anal. Bioanal. Chem. 2015,
DOI:10.1007/s002160148399y.
YoucanhaveapreviewoftheMOOCinMoodleenvironmentasaguest.Guestaccessallowsyou
to view the course contents and read the discussions in forums, posted during the Spring 2014
editionoftheMOOC.Asaguestyoucannotposttoforumsandtakequizzes.
Shortdescriptionofthecourse
This is an introductory course on estimation of measurement uncertainty, specifically related to
chemical analysis (analytical chemistry). The course gives the main concepts and mathematical
apparatus of measurement uncertainty estimation and introduces two principal approaches to
measurement uncertainty estimation the ISO GUM modeling approach (the bottomup or
modeling approach) and the singlelab validation approach as implemented by Nordtest (the
topdown or Nordtest approach). The course contains lectures, practical exercises and
numeroustestsforselftesting.
In spite of being introductory, the course intends to offer sufficient knowledge and skills for
carrying out uncertainty estimation for most of the common chemical analyses in routine
laboratoryenvironment.Thetechniquesforwhichthereareexamplesorexercisesincludeacid
base titration, Kjeldahl nitrogen determination, UVVis spectrophotometry, atomic absorption
spectroscopy and liquid chromatography mass spectrometry (LCMS). It is important to stress,
however, that for successful measurement uncertainty estimation experience (both in analytical
chemistry as such and also in uncertainty estimation) is crucial and this can be acquired only
throughpractice.
The materials of this course can also be useful for people who do not intend to follow the full
coursebutonlywanttofindanswerstosomespecificquestions.
Requiredpreliminaryknowledge
Introductory level knowledge of analytical chemistry is required. More advanced knowledge of
analytical chemistry and introductory knowledge of mathematical statistics is an
advantage. Fluency with and access to a spreadsheet software package (MS Excel, OpeOffice,
etc)ishighlyrecommended.
https://sisu.ut.ee/measurement 4
Whyismeasurementuncertaintyimportant
http://www.uttv.ee/naita?id=17711
https://www.youtube.com/watch?v=tn2DLYA72Dk
Studyoutcomes
Thestudentwhohassuccessfullypassedthecourseknows:
themainconceptsrelatedtomeasurementresultsandmeasurementuncertainty,including
theirapplicationtochemicalanalysis
themainmathematicalconceptsandtoolsinuncertaintyestimation
themainmeasurementuncertaintysourcesinchemicalanalysis
themainapproachesformeasurementuncertaintyestimation.
Thestudentwhohassuccessfullypassedthecourseisableto:
decide what data are needed for uncertainty estimation, understand the meaning of the
availabledataanddecidewhethertheavailabledataaresufficient
selecttheuncertaintyestimationapproachsuitablefortheavailabledata
quantify the uncertainty contributions of the relevant uncertainty sources using the
availabledata
carryoutestimationofuncertaintyusingthemainapproachesofuncertaintyestimation.
Organizationofthecoursematerial
The course (overall volume 1 ECTS) is organized in 12 sections, of which some are in turn split
intosmallersubsections.Thefollowingpartsarefoundinthesections:
1. The sections (and also many subsection) start with a briefintroduction stating the main
topic(s)andstudyoutcomesofthesection.
2. Themaintopicoftherespectivesectionisexplainedinashortvideolecture.
3. Thelectureisfollowedbyatextualpart.Thistextisinmostcasesmeanttocomplement,
not substitute the lecture (although in some cases the contents of the lecture are also
repeated in some extent). It rather gives additional explanations and addresses some
additionaltopicsthatwerenotcoveredbythelecture.
4. Mostsectionsendwithaselftest,whichenablestotesttheacquiredknowledgeandskills.
Thetestscontainquestions,aswellascalculationproblems.Theselftestsareononehand
meantforthestudentstomonitorhis/herprogress.Ontheotherhand,however,theyalso
promote thinking and provide (by the feedback of the questions) additional knowledge
about measurement uncertainty estimation in different practical situations. So, the self
tests are an intrinsic component of the course and it is strongly recommended to take all
ofthem.
The printout of the current version of the course materials (including lecture slides) can be
downloadedfromhere. The text surrounded with grey dashed line is footnote text and appears also
at the end of the page.
Ifyouconsistentlygetamessage"Servernotfound"whenattemptingtowatchvideosthenwith
highprobabilitythereasonisthefirewallofyourlocalnetwork.Thelocalnetworkadministrators
should enable outcoming connections from your network via port 1935. More specifically it is
necessarytoaccesstheserverrtmp://flash.ut.ee:1935.
Selftesting
Throughout the course there are numerous selftests for enabling the student to test his/her
knowledgeandskillsinspecifictopics.Eachtestisgradedasapercentage(100%corresponding
tocorrectlyansweringallquestionsandcorrectlysolvingallproblems).
https://sisu.ut.ee/measurement 5
Feedbackisgivenas:
Correctanswer,correctlyrecognisedandmarkedbythestudent.
Correctanswer,notrecognisedandnotmarkedbythestudent.
Incorrectanswer,however,consideredcorrectbythestudent.
Explanatory feedback is displayed when wrong answer is selected. All tests can be taken as
many times as needed and the success of taking these tests will not influence the final grade.
Werecommendthatyoutakeallthetestsandworkwiththemuntilyouachievescore100%and
onlythenmovetonexttopic.
Terminologyanddefinitions
Wherever possible, the used terminology adheres to the 3rd edition of the International
vocabulary of metrology Basic and general concepts and associated terms (VIM), JCGM
200:2008, International vocabulary of metrology Basic and general concepts and associated
terms(VIM),3rdedition.BIPM,IEC,IFCC,ILAC,ISO,IUPAC,IUPAPandOIML,2008.Availablefor
downloading free of charge at http://www.bipm.org/en/publications/guides/vim.html. referred
toasVIMthroughoutthecourse.However,intheinterestofbetterunderstandingandinorder
to stress the most important aspects of concepts, in many cases concepts are introduced by
definitions that are somewhat simplified compared to the VIM. More deeply interested students
areencouragedtoconsulttheVIM.
Mainliteraturesources
Thislistofliteraturereferencesisselecive,notexhauxtive.Thereferenceswereselectedbased
onthefollowingcriteria:(1)Widelyusedandcited(2)usefulunderpracticallabconditions(i.e.
not too deeply scientific) (3) a fairly recent version is available and (4) the document is
preferably available free of charge on the Internet. These references are referred to in the
courseviasuperscriptnumbersinroundbrackets,e.g.:VIM(1).
(1) JCGM 200:2008, International vocabulary of metrology Basic and general concepts and
associatedterms(VIM),3rdedition.BIPM,IEC,IFCC,ILAC,ISO,IUPAC,IUPAPandOIML, 2008.
Availableonlinefromhttp://www.bipm.org/en/publications/guides/vim.html
(2)JCGM100:2008EvaluationofmeasurementdataGuidetotheexpressionofuncertaintyin
measurement. JCGM, 2008. Available online from
http://www.bipm.org/en/publications/guides/gum.html
(3) Quantifying Uncertainty in Analytical Measurement, 2nd ed. Ellison, S. L. R. Williams, A.,
Eds. EURACHEM/CITAC, 2012. Available online from
http://eurachem.org/index.php/publications/guides
(4) Measurement Uncertainty Revisited. Eurolab Technical Report No 1/2007. Eurolab, 2007.
Availableonlinefromhttp://www.eurolab.org/documents/12007.pdf
(5) Handbook for Calculation of Measurement Uncertainty in Environmental Laboratories. B.
Magnusson, T. Nykki, H. Hovind, M. Krysell. Nordtest technical report 537, ed. 3. Nordtest,
2011. Available online from http://www.nordtest.info/index.php/technicalreports/
item/handbookforcalculationofmeasurementuncertaintyinenvironmentallaboratoriesnttr
537edition3.html
(6) Analytical Measurement: Measurement Uncertainty and Statistics. Eds: N. Majcen, V.
Gegevicius. ECJRC IRMM, 2012. Available online from
http://publications.jrc.ec.europa.eu/repository/ bitstream/111111111/29537/1/lana2207enn
web.pdf
Courseteam
https://sisu.ut.ee/measurement 6
IvoLeito,professorofanalyticalchemistryatUniversityofTartu.
Ivoteachesanalyticalchemistryandmetrologyinchemistryatallstudylevels
and organizes short training courses for practitioners on different topics of
analytical chemistry and metrology in chemistry. His research work embraces
awideareaoftopicsrangingfromstudiesofsuperacidsandsuperbasestoLC
MS analysis. He is the initiator of the masters programme Applied
MeasurementScienceatUniversityofTartu.
LauriJalukse,researchfellowinanalyticalchemistryatUniversityofTartu.
Lauri teaches analytical chemistry and metrology in chemistry at all study
levels. He is continuously introducing innovative and active learning
approachesintoteaching.Hisresearchworkisfocusedonmetrologicalstudies
of electrochemical and optical sensors, measurements of dissolved oxygen
concentration and moisture content, as well as organization of interlaboratory
comparisons.
IrjaHelm,researchfellowinanalyticalchemistryatUniversityofTartu.
Irja teaches practical classes of analytical chemistry. She takes care that
metrological concepts and approaches are introduced to students at as early
stageofanalyticalchemistrystudiesaspossible.
Technicaldesign:EducationalTechnologyCentre,UniversityofTartu.
Contactandfeedback
IvoLeito
UniversityofTartu
InstituteofChemistry
Ravila14a,Tartu50411
ESTONIA
Tel:+3725184176
email:[email protected]
Skype:leitoivo
Copyright
ThislearningobjectisprotectedbycopyrightunderthecopyrightlawoftheRepublicofEstonia.
Proprietary rights belong to the University of Tartu. It is allowed to reproduce, distribute and
perform the learning object in public only for learning and teaching. The learning object can be
used only as a whole, except in the case of citation. The learning object cannot be used for
commercialpurposesandallusagehastobeaccompaniedbyasufficientacknowledgement.Itis
notallowedtomodifyormakeanadaptationofthelearningobject.
***
[1] JCGM 200:2008, International vocabulary of metrology Basic and general concepts and
associatedterms(VIM),3rdedition.BIPM,IEC,IFCC,ILAC,ISO,IUPAC,IUPAPandOIML, 2008.
Available for downloading free of charge at
http://www.bipm.org/en/publications/guides/vim.html.
[2] This list of literature references is selecive, not exhauxtive. The references were selected
based on the following criteria: (1) Widely used and cited (2) useful under practical lab
conditions (i.e. not too deeply scientific) (3) a fairly recent version is available and (4) the
documentispreferablyavailablefreeofchargeontheInternet.Thesereferencesarereferredto
inthecourseviasuperscriptnumbersinroundbrackets,e.g.:VIM(1).
u_mooc_pdf_printout_2014d.pdf 1.62MB
https://sisu.ut.ee/measurement 7
1. Theconceptofmeasurementuncertainty
(MU)
Brief summary: This section introduces the concepts of measurand, true value, measured
value,error,measurementuncertaintyandprobability.
Theconceptofmeasurementuncertainty
http://www.uttv.ee/naita?id=17583
https://www.youtube.com/watch?v=BogGbA0hC3k
Measurementisaprocessofexperimentallyobtainingthevalueofaquantity.Thequantitythat
we intend to measure is called measurand. In chemistry the measurand is usually the content
(concentration) of some chemical entity (molecule, element, ion, etc) in some object. The
chemical entity that is intended be determined is called analyte. Measurands in chemistry can
be, for example, lead concentration in a water sample, content of pesticide thiabendazole in an
orangeorfatcontentinabottleofmilk.Intheprecedingexamplelead(element),ascorbicacid
(molecule) and fat (group of different molecules) are the analytes. Water, orange and milk are
analysisobjects(orsamplestakenfromanalysisobjects).
In principle, the aim of a measurement is to obtain the true value of the measurand. Every
effort is made to optimize the measurement procedure (in chemistry chemical analysis
procedureoranalyticalprocedure Analyticalchemistsmostlyusethetermanalyticalmethod.
In this course we use the term procedure instead of method, as this usage is supported by
the VIM. ) in such a way that the measured value is as close as possible to the true
value.However,ourmeasurementresultwillbejustanestimateofthetruevalueandtheactual
true value will (almost) always remain unknown to us. Therefore, we cannot know exactly how
near our measured value is to the true value our estimate always has some uncertainty
associatedwithit.
The difference between the measured value and the true value is called error. Error can have
either positive or negative sign. Error can be regarded as being composed of two parts
random error and systematic error which will be dealt with in more detail in coming
lectures. Like the true value, also the error is not known to us. Therefore it cannot be used in
practice for characterizing the quality of our measurement result its agreement with the true
value.
https://sisu.ut.ee/measurement 8
Scheme1.1.Interrelationsbetweentheconceptstruevalue,measuredvalue,errorand
uncertainty.
Measurementuncertaintyisalwaysassociatedwithsomeprobabilityaswillbeseeninthenext
lectures, it is usually not possible to define the uncertainty interval in such a way that the true
valuelieswithinitwith100%probability.
Measurement uncertainty, as expressed here, is in some context also called the absolute
measurementuncertainty. This means that the measurement uncertainty is expressed in the
same units as the measurand. As will be seen in subsequent lectures, it is sometimes more
usefultoexpressmeasurementuncertaintyasrelativemeasurementuncertainty,whichisthe
ratiooftheabsoluteuncertaintyU absandthemeasuredvaluey:
(1.1)
Relativeuncertaintyisaunitlessquantity,whichsometimesisalsoexpressedaspercent.
Measurementuncertaintyisdifferentfromerrorinthatitdoesnotexpressadifferencebetween
two values and it does not have a sign. Therefore it cannot be used for correcting the
measurementresultandcannotberegardedasanestimateoftheerrorbecausetheerrorhasa
sign. Instead measurement uncertainty can be regarded as our estimate, what is the highest
probable absolute difference between the measured value and the true value. With high
probability the difference between the measured value and the true value is in fact lower than
the measurement uncertainty. However, there is a low probability that this difference can be
higherthanthemeasurementuncertainty.
Boththetruevalueanderror(randomandsystematic)areabstractconcepts.Theirexactvalues
cannot be determined. However, these concepts are nevertheless useful, because their
estimatescanbedeterminedandarehighlyuseful.Infact,assaidabove,ourmeasuredvalue
isanestimateofthetruevalue.
***
[1]Analyticalchemistsmostlyusethetermanalyticalmethod.Inthiscourseweusetheterm
procedureinsteadofmethod,asthisusageissupportedbytheVIM.
[2]HereandinthelecturethecapitalUisusedtodenoteagenericuncertaintyestimate.The
symbolUispickedonpurpose,becauseexpandeduncertainty(generallydenotedbycapital
https://sisu.ut.ee/measurement 9
U)fitsverywellwiththeusageofuncertaintyinthissection.However,itisnotexplicitlycalled
expandeduncertaintyhere,asthistermwillbeintroducedinlaterlectures.
https://sisu.ut.ee/measurement 10
2. Theoriginofmeasurementuncertainty
The first video demonstrates how pipetting with a classical volumetric pipette is done and
explainswheretheuncertaintyofthepipettedvolumecomesfrom.
Whymeasurementresultshaveuncertainty?Theconceptofuncertaintysourceexplainedontheexample
ofpipetting
http://www.uttv.ee/naita?id=17577
https://www.youtube.com/watch?v=ufWJB9orWdU
Thesecondvideodemonstratespipettingwithamodernautomaticpipetteandexplainsthe
uncertaintysourcesinpipettingwithanautomaticpipette Thissecondvideoexplainshowto
pipetwithanautomaticpipetteifaccuratevolumeisdesired.Inmanyroutine,highvolume
applications(e.g.inbiochemistry),however,speedismoreimpoirtantthanaccuracyandinsuch
casessomestepsshownhere,mostimportantly,rinsing,canbeomitted.Also,insomecases
thereisaverylimitedvolumeavailableofthesolutionthatispipettedinsuchcasealso
rinsingisnotpossible.Ontheotherhand,ifstillhigheraccuracyisdesiredthenthesocalled
reversepipettingtechniquecanbeused.Reversepipettingismoreaccuratethanthecommonly
usedforwardpipetting,whichisthetechniquedemonstratedandexplainedinthisvideo..
Measurementuncertaintysourcesofpipettingwithanautomaticpipette
http://www.uttv.ee/naita?id=18164
https://www.youtube.com/watch?v=hicLweJcJWY
Measurement results have uncertainty because there are uncertainty sources (effects that
causeuncertainty).Theseareeffectsthatcausedeviationsofthemeasuredvaluefromthetrue
value. Thesesourcesalsocausetheexistenceoferrorandcouldthereforealsobecallederror
sources. If the used measurement procedure is well known then the most important
uncertaintysourcesareusuallyalsoknown.Effortsshouldbemadetominimizeand,ifpossible,
eliminate uncertainty sources by optimizing the measurement procedure (analysis procedure).
The uncertainty sources that cannot be eliminated (and it is never possible to eliminate all
uncertaintysources)havetobetakenintoaccountinuncertaintyestimation.
The magnitudes of the deviations caused by uncertainty sources are usually unknown and in
many cases cannot be known. Thus, they can only be estimated. If we can estimate the
magnitudes of all important uncertainty sources then we can combine them and obtain the
estimateofmeasurementuncertainty,whichinthiscasewillbecalledcombinedmeasurement
uncertainty. How this combining is mathematically done, will be demonstrated in the coming
lectures.
If we make a number of repeated measurements of the same measurand then ideally all these
repeated measurements should give exactly the same value and this value should be equal to
the true value of the measurand. In reality the results of the repeated measurements almost
alwaysdiffertosomeextentandtheirmeanvaluealsousuallydiffersfromthetruevalue.The
uncertaintysourcescausethis.Inasomewhatsimplifiedwaytheuncertaintysources(oreffects)
canbedividedintorandomeffectsandsystematiceffects. Itisinprinciplenotwrongtocall
themrandomandsystematicsourcesofuncertainty,butthisisnotusuallydone.Thisislargely
https://sisu.ut.ee/measurement 11
because, as we will see in a coming lecture, the concept of measurement uncertainty stresses
because, as we will see in a coming lecture, the concept of measurement uncertainty stresses
that random and systematic effects should be treated the same way. The following scheme
illustratesthis(greencirclesdenotetruevalues,yellowcirclesdenotemeasuredvalues):
Scheme2.1.Theinfluenceofrandomandsystematiceffectsonmeasurement
uncertainty.
Random effects cause the difference between the repeated measurement results (and thus,
obviously,alsofromthetruevalue).However,ifalargenumberofrepeatedmeasurementsare
made then the mean value will have little influence from the random effects (situation 2 on the
scheme). So, the influence of random effects can be decreased by increasing the number of
repetitions. Systematic effects cause deviation of all measurements in the series into the same
direction by the same magnitude. It is more correct to say by a predictable magnitude. This
means that the magnitude is not necessarily always the same it can vary, e.g. as the
magnitude of the measurand value varies but it can be predicted, i.e. it is not random.
Increasingthenumberofrepetitionsdoesnotenabledecreasingtheirinfluence(situation3on
thescheme).
Inprincipleitisdesirabletodeterminethemagnitudeanddirectionofthesystematiceffectsand
correctthemeasurementresultsforthesystematiceffects.However,itcanoftenbesodifficult
andworkintensive,thatitbecomesimpractical.Thereforeinmanycases,ratherthanaccurately
determining the systematic effects and correcting for them their possible magnitudes is
estimated and are taken into account as uncertainty sources. In lectures 5.4 and 6 random and
systematiceffectsaretreatedmorecomprehensively.
There are in general four main sources of uncertainty in volumetric measurements, i.e.
measurementsbypipettes,burettes,measuringcylindersandvolumetricflasks
There are some other uncertainty sources that usually turn out to be less important, because
they can be minimized or eliminated by correct working practices (however, they can be
important if these correctly practices are not applied). The remaining effects will usually
influencetherepeatabilityofpipettingoritscalibrationuncertaintyandcanbetakenintoaccount
withinthoseuncertaintysources.
Ifthepipetteisnotkeptvertically(bothglassandautomaticpipettes),waitingisnotlong
enough after the end of drainage of solution (glass pipette) then the pipetted volume will
belowerthantheoneobtainedwithcorrectpipetting.Nowaitingisneededinthecaseof
automatic pipettes because no liquid film remains (and must not remain) on pipette to
innerwalls.
Whenusingaglasspipettethenthereisalwayssomepossibilitythatsmallresiduesofthe
previous solution are still in the pipette. It is therefore a good idea to rinse the pipette
beforepipetting(e.g.twotimes)withthesolutionthatwillbepipetted(anddiscardingthe
rinse solution it into waste, not into the vessel from where the solution is taken). In the
case of automatic pipettes it is a good idea to use a new tip every time e new solution is
pipetted. In that case such contamination is usually negligible. Also, when pipetting the
samesolutionnumeroustimeswiththesamepipetteitisagoodideatomonitoritforthe
absenceofdropletsontheinnerwallsandreplacethetipwhenthedropletsappear.
Ifthewallsofaglasspipettearenotcleanthendropletsmayremainonthewallsafterthe
pipettedsolutionhasbeendrained.Thisleadstoadifferentvolumefromthecasewhenno
droplets are left on pipette walls after draining the solution. The obvious thing to do is to
cleanthepipette.
Ifthepipettedliquidisverydifferentfromwater(e.g.somehighlyviscousliquid,suchas
vegetable oil) then the pipetted volume may be systematically different from the nominal
volume of the pipette. This effect exists both with glass pipettes and with automatic
https://sisu.ut.ee/measurement 13
pipettes. In such a case the pipette should either be recalibrated using the liquid under
question or weighing should be used instead of volumetry In the case of automatic
pipettesalsoreveresepipettinginsteadofthemorecommonforwardpipettingcanbeused
todecreasetheuncertaintywhenpipettingviscousliquids..
In section 4 the uncertainty sources of pipetting (the same pipetting experiment that was
performed in the video) will be quantified and combined into the measurement uncertainty
estimate of pipetted volume. Sections 4.1 to 4.5 present the uncertainty calculation using a
factorycalibrated pipette. Section 4.6 presents an example of measurement uncertainty
calculation of pipetted volume using a selfcalibrated pipette. In section 5 an overview of the
majorityofuncertaintysourcesthatareencounteredinchemicalanalysiswillbegiven.
***
[1] This second video explains how to pipet with an automatic pipette if accurate volume is
desired. In many routine, highvolume applications (e.g. in biochemistry), however, speed is
more impoirtant than accuracy and in such cases some steps shown here, most importantly,
rinsing, can be omitted. Also, in some cases there is a very limited volume available of the
solution that is pipetted in such case also rinsing is not possible. On the other hand, if still
higher accuracy is desired then the socalled reverse pipetting technique can be used. Reverse
pipetting is more accurate than the commonly used forward pipetting, which is the technique
demonstratedandexplainedinthisvideo.
[2] These sources also cause the existence of error and could therefore also be called error
sources.
[3] It is in principle not wrong to call them random and systematic sources of uncertainty, but
thisisnotusuallydone.Thisislargelybecause,aswewillseeinacominglecture,theconcept
of measurement uncertainty stresses that random and systematic effects should be treated the
sameway.
[4]Itismorecorrecttosaybyapredictablemagnitude.Thismeansthatthemagnitudeisnot
necessarilyalwaysthesameitcanvary,e.g.asthemagnitudeofthemeasurandvaluevaries
butitcanbepredicted,i.e.itisnotrandom.
[5] In fact, even if the endpoint is determined potentiometrically, it can still have some
systematic deviation from the equivalence point. However, this effect is usually so small that it
canbeneglected.
[6] In the case of automatic pipettes also reverese pipetting instead of the more common
forwardpipettingcanbeusedtodecreasetheuncertaintywhenpipettingviscousliquids.
https://sisu.ut.ee/measurement 14
Briefsummary:Thissectionpresentsthemostbasicconceptsandtoolsforpracticalestimation
of measurement uncertainty. First, the concepts of random quantities and distribution functions
are introduced. Then the Normal distribution the most important distribution function in
measurementscienceisexplainedanditstwomainparametersthemeanvalueandstandard
deviationareintroduced(3.1).Basedonstandarddeviationtheconceptofstandarduncertainty
is introduced (3.1, 3.2). Thereafter the A type and B type uncertainty estimation is introduced
(3.3). The mean value of random quantities is also a random quantity and its reliability can be
described by the standard deviation of the mean (3.4). Besides the normal distribution three
more distribution functions are introduced: rectangular and triangular distribution (3.5) as well
astheStudentdistribution(3.6).
3.1.TheNormaldistribution
3.2.Mean,standarddeviationandstandarduncertainty
3.3.AandBtypeuncertaintyestimates
3.4.Standarddeviationofthemean
3.5.Rectangularandtriangulardistribution
3.6.TheStudentdistribution
3. Thebasicconceptsandtools
3.1.TheNormaldistribution
Brief summary: This lecture starts by generalizing that all measured values are random
quantities from the point of view of mathematical statistics. The most important distribution in
measurement science the Normal distribution is then explained: its importance, the
parametersoftheNormaldistribution(meanandstandarddeviation).Theinitialdefinitionsof
standarduncertainty(u),expandeduncertainty(U ) and coveragefactor (k ) are given. A
linkbetweentheseconceptsandtheNormaldistributioniscreated.
TheNormaldistribution
http://www.uttv.ee/naita?id=17589
https://www.youtube.com/watch?v=NF6leWyNZk
All measured quantities (measurands) are from the point of view of mathematical statistics
random quantities. Random quantities can have different values. This was demonstrated in the
lecture on the example of pipetting. If pipetting with the same pipette with nominal volume 10
ml is repeated multiple times then all the pipetted volumes are around 10 ml, but are still
slightlydifferent.Ifasufficientlylargenumberofrepeatedmeasurementsarecarriedoutandif
the pipetted volumes It is fair to ask, how do we know the individual pipetted volumes if the
pipettealwaystellsusjustthatthevolumeis10ml?Infact,ifwehaveonlythepipetteandno
other (more accurate) measurement possibility of volume then we cannot know how much the
volumes differ form each other or from the nominal volume. However, if a more accurate
method is available then this is possible. In the case of pipettig a very suitable and often used
moreaccuratemethodisweighing.Itispossibletofindthevolumeofthepipettedwater,which
is more accurate than that obtained by pipetting, by weighing the pipetted solution (most often
water) and divided the obtained mass by the density of water at the temperature of water.
Water is used in such experiments because the densities of water at different temperatures are
very accurately known (see e.g.
http://en.wikipedia.org/wiki/Properties_of_water#Density_of_water_and_ice). are plotted
accordingtohowfrequentlytheyareencounteredthenitbecomesevidentthatalthoughrandom,
the values are still governed by some underlying relationship between the volume and
frequency: the maximum probability of a volume is somewhere in the range of 10.005 and
https://sisu.ut.ee/measurement 15
10.007 ml and the probability gradually decreases towards smaller and larger volumes. This
relationshipiscalleddistributionfunction.
There are numerous distribution functions known to mathematicians and many of them are
encountered in the nature, i.e. they describe certain processes in the nature. In measurement
science the most important distribution function is the normaldistribution (also known as the
Gaussian distribution). Its importance stems from the socalled Central limit theorem. In a
simplified way it can be worded for measurements as follows: if a measurement result is
simultaneously influenced by many uncertainty sources then if the number of the uncertainty
sources approaches infinity then the distribution function of the measurement result approaches
the normal distribution, irrespective of what are the distribution functions of the
factors/parameters describing the uncertainty sources. In reality the distribution function of the
resultbecomesindistinguishablefromthenormaldistributionalreadyifthereare35(depending
on situation) significantly contributing Significantly cointributing uncertainty sources are the
important uncertainty sources. We have already qualitatively seen in section 2 that different
uncertaintysourceshavedifferentimportance.Inthecominglectureswewillalsoseehowthe
importance of an uncertainty source (its uncertainty contribution) can be
quantitatively expressed. uncertainty sources. This explains, why in so many cases measured
quantities have normal distribution and why most of the mathematical basis of measurement
scienceandmeasurementuncertaintyestimationisbasedonthenormaldistribution.
Scheme3.1.ThenormaldistributioncurveofquantityYwithmeanvalueymandstandard
deviations.
Thenormaldistributioncurvehasthebellshapedappearance(Scheme3.1),andisexpressedby
equation3.1:
(3.1)
In this equation f ( y ) is the probability that the measurand Y has value y. ym is the mean
valueofthepopulationandsisthestandarddeviationofthepopulation.ym characterizesthe
position of the normal distribution on the Y axis, s characterizes the width (spread) of the
https://sisu.ut.ee/measurement 16
distribution function, which is determined by the scatter of the data points. The mean and
standard deviation are the two parameters that fully determine the shape of the normal
distribution curve of a particular random quantity. The constants 2 and p are normalization
factors,whicharepresentinordertomaketheoverallareaunderthecurveequalto1.
Thewordpopulationheremeansthatwewouldneedtodoaninfinitenumberofmeasurements
inordertoobtainthetrueym ands values. In reality we always operate with a limited number
of measurements, so that the mean value and standard deviation that we have from our
experiments are in fact estimates of the true mean and true standard deviation. The larger is
the number of repeated measurements the more reliable are the estimates. The number of
parallel measurements is therefore very important and we will return to it in different other
partsofthiscourse.
The normal distribution and the standard deviation are the basis for definition of standard
uncertainty. Standard uncertainty, denoted by u, is the uncertainty expressed at standard
deviation level, i.e., uncertainty with roughly 68.3% coverage probability (i.e. the probability of
thetruevaluefallingwithintheuncertaintyrangeisroughly68.3%).Theprobabilityof68.3%is
oftentoolowforpracticalapplications.Thereforeuncertaintyofmeasurementresultsisinmost
casesnotreportedasstandarduncertaintybutasexpandeduncertainty.Expandeduncertainty,
denoted by U, is obtained by multiplying standard uncertainty with a coverage factor, This
definitionofexpandeduncertaintyissimplified.Amorerigorousdefinitiongoesviathecombined
standarduncertaintyandisintroducedinsection4.4.denotedbyk,whichisapositivenumber,
larger than 1. If the coverage factor is e.g. 2 (which is the most commonly used value for
coverage factor) then in the case of normally distributed measurement result the coverage
probability is roughly 95.5%. These probabilities can be regarded as fractions of areas of the
respectivesegmentsfromthetotalareaunderthecurveasillustratedbythefollowingscheme:
Scheme3.2.ThesamenormaldistributioncurveasinScheme3.1with2sand3s
segmentsindicated.
Since the exponent function can never return a value of zero, the value of f ( y ) (eq 3.1) is
higherthanzerowithanyvalueofy. This is the reason why uncertainty with 100% coverage is
(almost)neverpossible.
It is important to stress that these percentages hold only if the measurement result is normally
distributed. As said above, very often it is. There are, however, important cases when
https://sisu.ut.ee/measurement 17
measurement result is not normally distributed. In most of those cases the distribution function
has heavier tails, meaning, that the expanded uncertainty at e.g. k = 2 level will not
correspondtocoverageprobabilityof95.5%,butless(e.g.92%).Theissueofdistributionofthe
measurementresultwillbeaddressedlaterinthiscourse.
***
[1]Itisfairtoask,howdoweknowtheindividualpipettedvolumesifthepipettealwaystells
us just that the volume is 10 ml? In fact, if we have only the pipette and no other (more
accurate)measurementpossibilityofvolumethenwecannotknowhowmuchthevolumesdiffer
form each other or from the nominal volume. However, if a more accurate method is available
thenthisispossible.Inthecaseofpipettigaverysuitableandoftenusedmoreaccuratemethod
isweighing.Itispossibletofindthevolumeofthepipettedwater,whichismoreaccuratethan
that obtained by pipetting, by weighing the pipetted solution (most often water) and divided the
obtained mass by the density of water at the temperature of water. Water is used in such
experimentsbecausethedensitiesofwateratdifferenttemperaturesareveryaccuratelyknown
(seee.g.http://en.wikipedia.org/wiki/Properties_of_water#Density_of_water_and_ice).
[2] Significantly cointributing uncertainty sources are the important uncertainty sources. We
have already qualitatively seen in section 2 that different uncertainty sources have different
importance. In the coming lectures we will also see how the importance of an uncertainty
source(itsuncertaintycontribution)canbequantitativelyexpressed.
[3] This definition of expanded uncertainty is simplified. A more rigorous definition goes via
thecombinedstandarduncertaintyandisintroducedinsection4.4.
https://sisu.ut.ee/measurement 18
3.2.Mean,standarddeviationandstandard
uncertainty
Brief summary: the lecture explains calculation of mean (Vm ) and standard deviation (s).
Illustrates again the 68% probability of s. Explains how the standard uncertainty of
repeatabilityu(V,REP)canbeestimatedasstandarddeviationofparallelmeasurementresults.
Stressestheimportanceofstandarduncertaintyasthekeyparameterincarryingoutuncertainty
calculations: uncertainties corresponding to different sources (not only to repeatability) and to
different distribution functions are converted to standard uncertainties when uncertainty
calculationsareperformed.
Mean,standarddeviationandstandarduncertainty
http://www.uttv.ee/naita?id=17554
https://www.youtube.com/watch?v=ND3iryaVQ68
One of the most common approaches for improving the reliability of measurements is making
replicatemeasurementsofthesamequantity.Insuchacaseveryoftenthemeasurementresult
ispresentedasthemeanvalueofthereplicatemeasurements.Inthecaseofpipettingn times
withthesamepipettevolumesV1, V2, , Vn are obtained and the mean value Vm is calculated
asfollows:
(3.2)
Asexplainedinsection3.1,themeanvaluecalculatedthiswayisanestimateofthetruemean
value(whichcouldbeobtainedifitwerepossibletomakeaninfinitenumberofmeasurements).
(3.3)
Twoimportantinterpretationsofthestandarddeviation:
1.Standard deviation can be calculated also for quantities that are not normally distributed.
Thisenablestoobtainforthemstandarduncertaintyestimates.
2.Furthermore, also uncertainty sources that are systematic by their nature and cannot be
evaluated by repeating measurements can still be expressed numerically as standard
uncertaintyestimates.
3.Converting different types of uncertainty estimates to standard uncertainty is very
important, because as we will see in section 4, most of the calculations in uncertainty
evaluation, especially combining the uncertainties corresponding to different uncertainty
sources,arecarriedoutusingstandarduncertainties.
Standarduncertaintyofaquantity(inourcasevolumeV)expressedintheunitsofthatquantity
is sometimes also called absolute standard uncertainty. Standard uncertainty of a quantity
divided by the value of that quantity is called relativestandarduncertainty, urel (similarly to
eq1.1).InthecaseofvolumeV:
(3.4)
***
[1] We will see later that standard deviation of measurements repeated under conditions that
changer in predefined way (i.e. it is not repeatability) is also extremely useful in uncertainty
calculation,asitenablestakinganumberofuncertaintysourcesintoaccountsimultaneously.
https://sisu.ut.ee/measurement 20
3.3.AandBtypeuncertaintyestimates
Carrying out the same measurement operation many times and calculating the standard
deviation of the obtained values is one of the most common practices in measurement
uncertaintyestimation.Eitherthefullmeasurementoronlysomepartsofitcanberepeated.In
bothcasesusefulinformationcanbeobtained.Theobtainedstandarddeviation(orthestandard
deviation of the mean, explained in section 3.4) is then the standard uncertainty estimate.
Uncertainty estimates obtained as standard deviations of repeated measurement results are
called A type uncertainty estimates. If uncertainty is estimated using some means other than
statistical treatment of repeated measurement results then the obtained estimates are called B
type uncertainty estimates. The other means can be e.g. certificates of reference materials,
specificationsormanualsofinstruments,estimatesbasedonlongtermexperience,etc.
UncertaintyestimatesofAandBtype
http://www.uttv.ee/naita?id=18165
https://www.youtube.com/watch?v=jdbx5UMQD9k
https://sisu.ut.ee/measurement 21
3.4.Standarddeviationofthemean
Brief summary: Like the individual values, the mean value calculated from them is also a
randomquantityandforitalsoastandarddeviationcanbecalculated.Itispossibletocalculate
it from the standard deviation of the individual value. It is explained when to use the standard
deviationoftheindividualvalueandwhentousethestandarddeviationofthemean:whenever
the individual result is used in further calculation the standard deviation of the individual result
hastobeusedwheneverthemeanvalueisusedinfurthercalculations,thestandarddeviation
ofthemeanhastobeused.
Standarddeviationofthemean
http://www.uttv.ee/naita?id=17580
https://www.youtube.com/watch?v=GLsHHIW1yjo
Thestandarddeviations(V)calculatedusingtheformula3.3isthestandarddeviationofan
individualpipettingresult(value).Whenthemeanvalueiscalculatedfromasetofindividual
valueswhicharerandomlydistributedthenthemeanvaluewillalsobearandomquantity.As
foranyrandomquantity,itisalsopossibletocalculatestandarddeviationforthemeans(Vm).
Onepossiblewaytodothatwouldbecarryingoutnumerousmeasurementseries,findthemean
foreveryseriesandthencalculatethestandarddeviationofalltheobtainedmeanvalues.This
is,however,tooworkintensive.However,thereisaverymuchsimplerapproachforcalculating
s(Vm),simplydividethes(V)bysquarerootofthenumberofrepeatedmeasurementsmade:
(3.5)
So, for a set of repeated pipetting values we have in fact two standard deviations: standard
deviationofthesinglevalues(V)andstandarddeviationofthemeans(Vm).Itisimportantto
ask:whenweuseoneandwhenanotherofthem?
The general rule of thumb is the following: when the measured value reported or used in
subsequent calculations is a single value then we use standard deviation of the single value
whenitisthemeanvaluethenweusethestandarddeviationofthemean.
Letusillustratethisbytwoexamples:
In the case of single pipetting or single weighing the repeatability uncertainty of course cannot
be estimated from this single operation. In these cases repeatability is determined separately
andthenusedfortheconcretemeasurements.
https://sisu.ut.ee/measurement 22
***
[1]Aswewillseelater,modernbalancesarehighlyaccurateinstrumentsanduncertaintydueto
weighingisseldomamongtheimportantuncertaintysources.So,unlesssomedisturbingeffects
interferewithweighing,itisusuallynotnecessarytoweighmaterialswithmanyrepetitions.
https://sisu.ut.ee/measurement 23
Briefsummary:Rectangulardistributionandtriangulardistributionareexplained,aswellas
howtheuncertaintiescorrespondingtorectangularortriangulardistributioncanbeconvertedto
standarduncertainties.Oftentheinformationondistributionfunctionismissingandthenusually
somedistributionfunctionisassumedorpostulated.Rectangularandtriangulardistributionsare
amongthemostcommonpostulateddistributionfunctions.Recommendationsaregiven,whichof
thesedistributionstoassume.
Otherdistributionfunctions:rectangularandtriangulardistribution
http://www.uttv.ee/naita?id=17584
https://www.youtube.com/watch?v=g_PefybO2Ao
Scheme3.3.Rectangularandtriangulardistributions.Bothofthemcorrespondtothe
situation(10.0000.063)ml.
Inmeasurementuncertaintyestimationsituationsoftenoccurwhereitisnecessarytomake
choicebetweentwoalternativesofwhichonemaypossiblyleadtosomewhatoverestimated
uncertaintyandtheotheronetosomewhatunderestimateduncertainty.Insuchsituationitis
usuallyreasonabletorathersomewhatoverestimatethanunderestimatetheuncertainty.
https://sisu.ut.ee/measurement 24
3.5.Rectangularandtriangulardistribution
Brief summary: Like the individual values, the mean value calculated from them is also a
randomquantity.IftheindividualvaluesaredistributedaccordingtotheNormaldistributionthen
the mean value calculated from them is distributed according to theStudentdistribution (also
calledastdistribution).ThepropertiesofthetdistributioncomparedtotheNormaldistribution
are explained. Importantly, the shape of the tdistribution curve depends on the number of
degreesoffreedom. If the number of individual values approaches infinity then the shape of
thetdistributioncurveapproachestheNormaldistributioncurve.
Otherdistributionfunctions:theStudentdistribution
http://www.uttv.ee/naita?id=17708
https://www.youtube.com/watch?v=CWU8KM2z59I
Ifameasurementisrepeatedandthemeaniscalculatedfromtheresultsofthesingleindividual
measurementsthen,justastheindividualresults,theirmeanwillalsobearandomquantity.If
theindividualresultsarenormallydistributedthentheirmeanisdistributedaccordingtothe
Studentdistribution(alsoknownasthetdistribution).Studentdistributionispresentedin
Scheme3.5.
Scheme3.5.theStudentdistribution.
SimilarlytothenormaldistributiontheStudentdistributionalsohasmeanvalueym andstandard
deviation s. ym is the mean value itself, It can look strange at first sight that while the mean
value ym is the only mean value we have, we immediately take it as the mean value of the
https://sisu.ut.ee/measurement 25
distributionofmeanvalues.However,ifwehadmoremeanvalues,wewouldanywaypoolthem
distributionofmeanvalues.However,ifwehadmoremeanvalues,wewouldanywaypoolthem
intoasinglemeanvalue(withamuchhigherdf!)andusethisvalue.andstandarddeviationis
the standard deviation of the mean, calculated as explained in section 3.4. But differently from
the normal distribution there is in addition a third characteristic the number of degrees of
freedomdf.Thisnumberisequaltothenumberofrepeatedmeasurementsminusone.So,the
four Student distribution graphs in Scheme 3.5 correspond to 101, 11, 6 and 3 repeated
measurements,respectively.
Ifdfapproachesinfinitythenthetdistributionapproachesthenormaldistribution.Inreality30
50degreesoffreedomissufficientforhandlingthetdistributionasthenormaldistribution.So,
thecurvewithdf=100inScheme3.5canberegardedasthenormaldistributioncurve.
The lower is the number of degrees of freedom the heavier are the tails of the Student
distribution curve and the more different is the distribution from the normal distribution. This
means that more probability resides in the tails of the distribution curve and less in the middle
part. Importantly, the probabilities pictured in Scheme 3.2 for the 1s, 2s and 3s ranges
aroundthemeandonotholdanymore,butarealllower.
***
[1] It can look strange at first sight that while the mean value ym is the only mean value we
have,weimmediatelytakeitasthemeanvalueofthedistributionofmeanvalues.However,if
we had more mean values, we would antway pool them into a single mean value (with a much
highedf!)andusethisvalue.
[2] Student coefficients (i.e. tdistribution values) for a given set of coverage probability and
numberofdegreesoffreedomcanbeeasilyobtainedfromspecialtablesinstatisticalhandbooks
(use twosided values!), from calculation or data treatment software, such as MS Excel or
Openoffice Calc or from the Internet, e.g. from the address
https://en.wikipedia.org/wiki/Student_distribution
[3] The contributions of different uncertainty sources can be expressed numerically. This is
explained in section 9.9 and the respective calculations are shown in 9.7. In this context the
phrase heavily dominating means that the contribution (uncertainty index) of the respective
inputquantityisabove75%.
https://sisu.ut.ee/measurement 26
3.6.TheStudentdistribution
4.Thefirstuncertaintyquantification
Briefsummary:Inthissectionthebasicconceptsandtoolsoftheprevioussectionsareputinto
practice on the example of a simple analytical chemistry operation pipetting. The uncertainty
sources identified in section 2 are now quantified (4.1), the obtained individual uncertainty
estimatesare(whenneeded)convertedtostandarduncertaintiesandarethencombinedintothe
combined standard uncertainty (4.2). The uncertainty components making up the combined
standard uncertainty are compared and some conclusions are made (4.3). The combined
standard uncertainty is converted into expanded uncertainty (4.4) and the result is presented
(4.5).Thisapproachisthenpracticedonecalculationexample(calibrationofpipette,4.6).
4.1.Quantifyinguncertaintycomponents
4.2.Calculatingthecombinedstandarduncertainty
4.3.Lookingattheobtaineduncertainty
4.4.Expandeduncertainty
4.5.Presentingmeasurementresults
4.6.Practicalexample
https://sisu.ut.ee/measurement 27
4.1.Quantifyinguncertaintycomponents
Briefsummary: The same pipetting as in lecture 2 is now examined from the point of view of
quantifying the uncertainty sources. All the important sources of uncertainty are quantitatively
expressed as uncertainty components uncertainty estimates quantitatively describing the
respective uncertainty source. The uncertainty components are quantified. An example is given
on converting an uncertainty estimate with (assumedly) rectangular distribution into a standard
uncertaintyestimate.
Introductiontoquantifyingmeasurementuncertainty
http://www.uttv.ee/naita?id=17555
https://www.youtube.com/watch?v=CaYJGxtQBzo
The main uncertainty sources are the same as explained in lecture 2 and here the uncertainty
componentscorrespondingtothemarequantified.
(4.1)
(4.2)
Quantifyingtheuncertaintyduetotemperatureeffectinvolumetricmeasurement
http://www.uttv.ee/naita?id=17825
https://www.youtube.com/watch?v=CDjX8K_Vsds
u(V, TEMP) is dependent on the volume V of liquid delivered, [5] In the case of a volumetric
pipettethenominalvolumeisthesameasthedeliveredvolumebutinthecaseofaburetteitis
usually not. So, if from a 25 ml burette 12.63 ml of solution is delivered then the volume that
hastobeusedfortemperatureeffectcalculationis12.63ml,not25ml.themaximumpossible
temperature difference from 20C (t) and the thermal expansion coefficient of water . It is
calculatedasfollows:
(4.3)
Dividing by square root of 3 is for transforming the uncertainty estimate into standard
uncertainty(assumingrectangulardistributionoft).ItisimportanttonotethattheV in eq 4.3
alwaysreferstotheactualmeasuredvolume,notthefullcapacityofthevolumetricdevice.For
example, if 21.2 ml of solution was measured with a 50 ml burette then the volume to use is
21.2ml,not50ml.
***
[1] Since pipetting for delivering a certain liquid volume is done only once and cannot be
averaged (i.e. it is not possible to pipet several times and then average the volumes) the
suitableestimateofrepeatabilityuncertaintyisthestandarddeviationofasinglemeasurement,
notstandarddeviationofthemean.
[2] The accurate volumes can be measured by weighing the water delivered by the pipette and
convertingitintovolumebyusingaccuratedensitydata.
[3] This uncertainty can be significantly reduced if the pipette is recalibrated in laboratory by
weighingthedeliveredwater.
[4]Ifvolumetricglasswareiscalibratedinthesamelaboratorythenadifferenttemperaturecan
beused.
[5]Inthecaseofavolumetricpipettethenominalvolumeisthesameasthedeliveredvolume
butinthecaseofaburetteitisusuallynot.So,iffroma25mlburette12.63mlofsolutionis
deliveredthenthevolumethathastobeusedfortemperatureeffectcalculationis12.63ml,not
25ml.
https://sisu.ut.ee/measurement 29
4.2.Calculatingthecombinedstandard
uncertainty
The uncertainty components that were quantified in the previous lecture are now combined into
the combined standard uncertainty (uc) standard uncertainty that takes into account
contributions from all important uncertainty sources by combining the respective uncertainty
components.Theconceptofindirectmeasurementwherebythevalueoftheoutputquantity
(measurement result) is found by some function (model) from several input quantities is
introduced and explained. The majority of chemical measurements are indirect measurements.
The general case of combining the uncertainty components into combined standard uncertainty
aswellasseveralspecificcasesarepresentedandexplained.
Thefirstvideolectureexplainsinasimplewayhowtheuncertaintycomponentsarecombinedin
the particular example of pipetting. The second video lecture presents the general overview of
combiningtheuncertaintycomponents.
Combiningtheuncertaintycomponentsintothecombinedstandarduncertaintyinthecaseofpipetting
http://www.uttv.ee/naita?id=17556
https://www.youtube.com/watch?v=S5v58VQ4zSg
In all cases where combined standard uncertainty is calculated from uncertainty components all
theuncertaintycomponentshavetobeconvertedtostandarduncertainties.
Intheexampleofpipettingthecombinedstandarduncertaintyiscalculatedfromtheuncertainty
componentsfoundintheprevioussectionasfollows:
(4.4)
This is the typical way of calculating combined standard uncertainty if all the uncertainty
components refer to the same quantity and are expressed in the same units. It is often used in
the case of direct measurements measurements whereby the measurement instrument
(pipette in this case) gives immediately the value of the result, without further calculations
needed.
Combiningtheuncertaintycomponentsintothecombinedstandarduncertainty:simplecasesandthe
generalcase
http://www.uttv.ee/naita?id=17826
https://www.youtube.com/watch?v=FJ4hn9LgGmw
An indirect measurement is one where the output quantity (result) is found ba calculation
(usingamodelequation)fromseveralinputquantities.Atypicalexampleistitration.Incaseof
titration with 1:1 mole ratio the analyte concentration in the sample solution CS (the output
quantity) is expressed by the input quantities volume of sample solution taken for titration
(VS),titrantconcentration(CT)andtitrantvolumeconsumedfortitration(VT)asfollows:
(4.5)
https://sisu.ut.ee/measurement 30
InthegeneralcaseiftheoutputquantityYisfoundfrominputquantitiesX1,X2,Xn according
tosomefunctionFasfollows
(4.6)
thenthecombinedstandarduncertaintyoftheoutputquantityuc(y)canbeexpressedviathe
standarduncertaintiesoftheinputquantitiesu(xi)asfollows:
(4.7)
derivatives.Atfirstsighttheeq4.7mayseemverycomplexbutitisinfactnottoodifficultto
usetheuncertaintycomponentscanbecalculatednumericallyusingtheKragtensspreadsheet
method(asisdemonstratedinsection9.7).
In specific cases simpler equations hold. If the output quantity is expressed via the input
quantitiesasfollows
(4.8)
then (4.9)
Importantly, irrespective of whether the input quantities are added or subtracted, the squared
standarduncertaintiesunderthesquarerootarealwaysadded.
This way of combining uncertainty components is in principle the same as used above for the
caseofpipetting.
Ifthemeasurementmodelis
(4.10)
then (4.11)
Ascanbeseen,hereitistherelativestandarduncertaintiesthatarecombinedandthesquared
summinggivesustherelativecombinedstandarduncertaintyoftheoutputquantity.The
absolutecombinedstandarduncertaintyoftheoutputquantityisfoundasfollows:
(4.12)
Thefileusedinsecondvideocanbedownloadedfromhere.
https://sisu.ut.ee/measurement 31
Combined Standard Uncertainty
1 2
Titrant concenration V C
CS T T
Volume of sample solution taken for titration VS
Volume of titrant consumed for titartion
Analyte concentration
in sample solution Volume of sample solution
taken for titration
3 4
Finding uc Finding uc
If the model is: If the model is:
X1 X 2
Y
Y X 1 X 2 ... X n X3 X4
then: Then: 2 2 2 2
u ( x1 ) u ( x2 ) u ( x3 ) u ( x4 )
uc ( y ) y
uc ( y ) u ( x1 ) u ( x2 ) ... u ( xn )
2 2 2
x1 x2 x3 x4
2 2 2 2
uc ( y ) u ( x1 ) u ( x2 ) u ( x3 ) u ( x4 )
All uncertainties have to be converted to standard
uncertainties before calculations!
y x1 x2 x3 x4
5 6
Important:
Y F ( X 1 , X 2 , ... , X n )
Uncertainty component
7 8
Comparingtheuncertaintycomponents
http://www.uttv.ee/naita?id=17578
https://www.youtube.com/watch?v=Xnq27nq_bg
We can see the uncertainty component with the largest magnitude is the calibration
uncertainty u(V, CAL) = 0.017 ml. The combined standard uncertainty is in fact quite similar to
it: uc(V) = 0.019 ml. If the summing were made not by the squaring and square root approach
but by simple arithmetic sum then the value would be 0.028. This is a good illustration of the
propertyofthesquaredsumming:thesmalleruncertaintycomponentsaresuppressedbylarger
uncertaintycomponents.
The idea of the squared summing of the components is that the different effects causing
uncertainty influence the result in different directions (thus partially canceling) and their
magnitudesarenotnecessarilyequaltothevaluesoftheuncertaintyestimatesbutcanalsobe
smaller(seesection1).
Lookingattheuncertaintycontributionsisveryusefulifonewantstoreducetheuncertainty.In
order to reduce the uncertainty of a particular measurement it is always necessary to focus on
decreasing the uncertainties caused by the largest components. So, in this case it is not very
useful to buy a more expensive air conditioner for the room because the resulting uncertainty
improvement will be small. It will also not be possible to improve the uncertainty markedly by
reducing the repeatability component. Clearly, whatever is done with these two components the
combinedstandarduncertaintyuc(V)(eq4.4)cannotdecreasefrom0.019mltolowerthan0.017
ml,whichisnotasignificantdecrease.Thus,ifmoreaccuratepipettingisneeded,thentheway
to go is to calibrate the pipette in the laboratory. This way it is realistic to achieve threefold
lower calibration uncertainty, which leads to two times lower combined uncertainty of the
pipettedvolume.Seesection4.6foranexample.
https://sisu.ut.ee/measurement 33
4.4.Expandeduncertainty
Briefsummary: The probability of roughly 68% that is provided by the standard uncertainty is
often too low for the users of measurement uncertainty. Therefore measurement uncertainty is
presentedtocustomersmostlyasexpandeduncertainty,U.Expandeduncertaintyiscalculated
fromthestandarduncertaintybymultiplyingitwithacoveragefactor,k.
Inthecaseofthepipettingexamplethek=2expandeduncertaintyisfoundasfollows:
U(V)=uc(V)k=0.019ml2=0.038ml (4.13)
Expandeduncertaintyatk=2levelisthemostcommonwayofexpressinguncertaintyof
measurementandanalysisresults.
Calculatingtheexpandeduncertainty
http://www.uttv.ee/naita?id=17579
https://www.youtube.com/watch?v=tBwkuL0ap14
https://sisu.ut.ee/measurement 35
4.5.Presentingmeasurementresults
Briefsummary: The pipetting result the value and expanded uncertainty is presented. It is
stressed that it is important to clearly say, what was measured. The correct presentation of
measurement result includes value, uncertainty and information about the probability of the
uncertainty. It is explained that in simplified terms we can assume that k = 2 corresponds to
roughly 95% of coverage probability. It is explained how to decide how many decimals to give
whenpresentingameasurementresultandtheuncertainty.
Thecorrectpresentationofthemeasurementresultinthiscasewouldlookasfollows:
Whencanweassumethattheoutputquantityisnormallydistributed?Thatis,whencanwewrite
norm.besidesthecoveragefactor?Rigorousanswertothisquestionisnotstraightforward,but
a simple rule of thumb is that when there are at least three main uncertainty sources of
comparableinfluence(i.e.thesmallestandlargestoftheuncertaintycomponentsdifferbyca3
times or less) then we can assume that the distribution function of the output is sufficiently
similar to the normal distribution. In this particular case there is a dominating quantity with
assumedly rectangular distribution, which leads to a distribution function with very weak tails
(meaning: this in fact not exactly a normal distribution). So, 95% coverage probability is
achieved already by expanded uncertainty of 0.0034 ml (as evidenced by Monte Carlo
simulations). Thus, the presented uncertainty of 0.0038 ml is a conservative estimate (which is
notbadasexplainedinsection3.5).
Section 9.8 presents a more sophisticated approach of calculating expanded uncertainty that
correspondstoaconcretecoverageprobability.
Presentingofmeasurementresult
http://www.uttv.ee/naita?id=17576
https://www.youtube.com/watch?v=x9l1jIVqT7k
***
[1]Inthisparticularcasethereisadominatingquantitywithassumedlyrectangulardistribution,
which leads to a distribution function with very weak tails (meaning: this in fact not exactly a
normal distribution). So, 95% coverage probability is achieved already by expanded uncertainty
of 0.0034 ml (as evidenced by Monte Carlo simulations). Thus, the presented uncertainty of
0.0038mlisaconservativeestimate(whichisnotbadasexplainedinsection3.5).
https://sisu.ut.ee/measurement 36
4.6.Practicalexample
Thisisanexampleofcalculatingthevolumeanditsuncertaintyofliquid
deliveredfromaselfcalibratedvolumetricpipette
Theuncertaintyofthepipettedvolumeu(V)hasthreemainuncertaintycomponents:uncertainty
duetorepeatability,u(V,rep)uncertaintyduetopipettecalibration,u(V,cal)anduncertaintydue
tothetemperaturedifferencefrom20C,u(V,temp).
The estimate of the probable maximum difference of the pipette volume from the nominal
volume, expressed as x is often used as the estimate of calibration uncertainty of the pipette
(as was done in section 4.1). It is usually given by the manufacturer without any additional
informationaboutitscoverageprobabilityordistributionfunction.Insuchacaseitisthesafest
toassumethatrectangulardistributionholdsandtoconverttheuncertaintyestimatetostandard
uncertaintybydividingitwithsquarerootofthree.
Usually in the case of highaccuracy work the pipette is calibrated in the laboratory in order to
obtain lower calibration uncertainty. As was seen in sections 4.2 and 4.3, if the uncertainty due
to factory calibration is used, then this calibration uncertainty component is the most influential
one. So, reducing it would also reduce the overall uncertainty. It is very important, that
calibration and pipetting are performed under the same conditions and preferably by the same
person. From the calibration data we can obtain two important pieces of information: (1) the
correctiontermforthepipettevolumeVcorrectionwithuncertaintyu(V,cal) and (2) repeatability
ofpipettingu(V,rep).Theexamplepresentedhereexplainsthis.
For calibration of a pipette water is repeatedly pipetted (at controlled temperature in order to
know its density), the masses of the pipetted amounts of water are measured and the pipetted
volumes of water are calculated (using density of water at the calibration temperature). Here
arethecalibrationdataofa10mlpipette:
Thestandarddeviationiscalculatedaccordingtotheequation:
Uncertaintyduetorepeatabilityofpipettingu(V,REP)isequaltothisstandarddeviation0.0057
ml.Pipettingisoftenusedintitrationanalysis.Ifthesolutionthatistitratedispipettedthenthis
repeatabilitycontributionisalreadyaccountedforintherepeatabilityofthetitrationresultsand
isnotseparatelytakenintoaccountintheuncertaintyofpipettevolume.
Uncertainty of the calibration (in fact the uncertainty of the correction) has to be always taken
into account and this is expressed (when calibration is done) as the standard deviation of the
mean:
Inthisexamplethecorrectionis0.0080mlanditsstandarduncertaintyis0.0018ml.
When there is a possibility that pipetting is performed at a different temperature from the
https://sisu.ut.ee/measurement 37
calibration (and this possibility exists almost
always), then an additional uncertainty source due
to temperature change is introduced and it has to
betakenintoaccount.
Now,whenwewillperformasinglepipetting,thevolumeis9.992mlanditscombinedstandard
uncertaintyis
Thek=2expandeduncertainty Laterinthiscourse(section9.8)wewillsee,howtorigorously
find,whetherwecansaythatthek=2expandeduncertaintyinaparticularcasecorrespondsto
95% (this depends on the socalled effective number of degrees of freedom). And if not then
what k should be used to achieve approximately 95% coverage probability. In the case of this
example the effective number of degrees of freedom is 26 and the respective coverage factor
(actually the Student coefficient) with the probability of 95% is 2.06, which is only very slightly
different from 2 (the expanded uncertainty would increase from 0.015 ml to 0.016 ml). of the
pipettedvolumecanbefoundasfollows:
U(V)=uc(V)k=0.00772=0.0154ml
Asexplainedinsection4.5ifthefirstsignificantdigitoftheuncertaintyis1...4thenuncertainty
shouldbepresentedwithtwosignificantdigits.Thuswecanwritetheresult:
Thevolumeofthepipettedliquidis:
V=(9.9920.015)ml,k=2,norm.(4.15)
https://sisu.ut.ee/measurement 38
It is interesting to compare now this expanded uncertainty with the expanded uncertainty
obtained in section 4.5 (eq 4.14). We see that when the pipette is calibrated in our laboratory
then the uncertainty of the volume is more than two times lower. We also see that the
uncertainty component due to pipette calibration, which back then was the largest uncertainty
component,isnowthesmallest.
***
[1]Laterinthiscourse(section9.8)wewillsee,howtorigorouslyfind,whetherwecansaythat
thek=2expandeduncertaintyinaparticularcasecorrespondsto95%(thisdependsontheso
calledeffectivenumberofdegreesoffreedom).Andifnotthenwhatkshouldbeusedtoachieve
approximately95%coverageprobability.
Inthecaseofthisexampletheeffectivenumberofdegreesoffreedomis26andtherespective
coverage factor (actually the Student coefficient) with the probability of 95% is 2.06, which is
only very slightly different from 2 (the expanded uncertainty would increase from 0.015 ml to
0.016ml).
https://sisu.ut.ee/measurement 39
5.Principlesofmeasurementuncertainty
estimation
Briefsummary:ThemainprinciplesofmeasurementuncertaintyestimationthesocalledGUM
principles are presented on the example of determination of pesticides in oranges. These
principles have been laid down in the ISO GUM ISO GUM originally refers to the Guide To The
ExpressionofUncertaintyinMeasurement,ISO,Geneva,Switzerland,1993(Reprinted1995).In
2008thisdocumentwasrevisedandreissuedasISOJCGM100:2008Evaluationofmeasurement
data Guide to the expression of uncertainty in measurement. The latter document is available
online from http://www.bipm.org/en/publications/guides/gum.html and they are now
universally accepted as being the common foundation of all the different uncertainty estimation
approaches.Theseprinciplesarethefollowing:
1.The basis of any measurement (thus obviously also measurement uncertainty evaluation)
isthedefinitionofthemeasurand
2.Theusedmeasurementprocedurehastocorrespondtothemeasuranddefinition
3.All relevant sources of uncertainty have to be carefully considered and those that are
importanthavetobetakenintoaccount
4.The random and systematic effects are treated the same way when estimating
measurementuncertaintybothareevaluatedasstandarduncertainties,whichthereafter
arecombinedintothecombinedstandarduncertainty.
5.1.Measuranddefinition
5.2.Measurementprocedure
5.3.Sourcesofmeasurementuncertainty
5.4.Treatmentofrandomandsystematiceffects
***
[1] ISO GUM originally refers to the Guide To The Expression of Uncertainty in Measurement,
ISO, Geneva, Switzerland, 1993 (Reprinted 1995). In 2008 this document was revised and
reissued as ISO JCGM 100:2008 Evaluation of measurement data Guide to the expression of
uncertainty in measurement. The latter document is available online from
http://www.bipm.org/en/publications/guides/gum.html
https://sisu.ut.ee/measurement 40
5.1.Measuranddefinition
Brief summary: The first principle of measurement uncertainty is: the measurand must be
correctlyandunambiguouslydefined.Theimportanceofmeasuranddefinitionisexplainedonthe
exampleofpesticidedeterminationinoranges.
Definingthemeasurand
http://www.uttv.ee/naita?id=17585
https://www.youtube.com/watch?v=Bf_0EssHQ60
Defining the measurand in the case of pesticide determination in oranges is not trivial. On one
handitisimportanttodefinewhethertheresultisappliedtoasingleorangeorfeworangesthat
were taken as the sample or whether it is applied to the whole lot of oranges (the whole
analysis object, also called sampling target). On the other hand, oranges are not homogenous.
Pesticidesareappliedonorangesurface,notinside.Atthesametimepesticidecandiffusefrom
theorangepeeltotheinside.So,anumberofdifferentpossibilitiesexist:wholeorange,whole
peel,outsidepartofthepeel,onlyorangeflesh.
Combiningtogetherweget8possibilities,inwhatexactlywecandeterminepesticides.
Measuringpesticidecontentaccordingtoanyofthesewillleadtodifferentandmutuallynon
comparableresults.
In addition (not explained in the video) instead of defining the measurand via the total analyte
content in the sample (or part of the sample) it is often more practical to look at some part of
theanalyteonly.
Agoodexampleisphosphorusdeterminationinsoil.Althoughitispossibletodeterminethe
totalphosphoruscontentinsoilitisinfactmoreinterestingtodetermineonlythepartthatis
availabletoplantsthebioavailablephosphorusbecauseitisthispartofthetotal
phosphoruscontentthatcontributestothefertilityofthesoilandisthereforeofinterestin
agriculture.
Totalphosphoruscontentandbioavailablephosphoruscontentaredifferentmeasurandsandtheir
values for the same soil differ strongly. This has important implications for the measurement
procedure.Thisisexplainedinsection5.2.
https://sisu.ut.ee/measurement 41
5.2.Measurementprocedure
Brief summary: The main steps of a measurement/analysis procedure are presented on the
example of pesticide measurement are presented: Sample preparation (in this case:
homogenization, extraction(s), extract purification), instrument calibration, actual analysis. It is
stressedthatthemeasurementproceduremustcorrespondtomeasuranddefinition.
Measurementprocedure
http://www.uttv.ee/naita?id=17586
https://www.youtube.com/watch?v=BKlB_iB4wp4
This scheme of a chemical analysis procedure is very general. In specific cases there can be
deviationsfromthisscheme(morestepsorlesssteps).Inparticular,samplingisnotintroduced
here as a step of chemical analysis procedure. This holds if samples are brought to the
laboratory for analysis and the laboratory itself does not do sampling (see section 5.3 for more
details).
Samplepreparationisoftenthemostworkintensivepartofchemicalanalysisandinmostcases
it is also the part, which has the largest uncertainty contribution. Sample preparation usually
involveseitherofthetwoapproaches:
1.Essentially destroying the sample matrix so that a solution containing the analyte(s) and
few matrix components is obtained. This is often done by digestion with acids or fusing
withalkaliesorsalts.Thisapproachissuitablefordeterminingelements.
2.Separating the analyte(s) from the sample matrix so that a solution containing the
analyte(s) is obtained where the amount of matrix components is as small as possible.
Thisisusuallydonebyasetofextractions.Thisapproachissuitablefororganicanalytes.
Obviously the choice of sample preparation procedure depends on whether the measurand
corresponds to the total analyte content in the sample or some part of it, e.g. the bioavailable
analyte content (see the text in section 5.1). In the case of e.g. total phosphorus content
determination in soil the analyst can make the choice of the sample preparation procedure. All
procedures that lead to determination of the total phosphorus content (often involving complete
destruction of the matrix) are suitable. In the case of determining of e.g. bioavailable
phosphoruscontentinsoilthesamplepreparationproceduremustmimicthewaytheplantsget
phosphorus from soil. So, sample preparation involves leaching at predefined conditions. Such
samplepreparationproceduresareoftenstandardizedandwhenevertheresultsaremeanttobe
mutually comparable they must be obtained with the same procedure. Thus, in this latter case
thesamplepreparationprocedurebecomespartofthemeasuranddefinition.
https://sisu.ut.ee/measurement 42
5.3.Sourcesofmeasurementuncertainty
Brief summary: The overview of possible uncertainty sources, on the example pesticide
analysis, is presented. Although the uncertainty sources are presented on the example of
pesticide analysis, the same uncertainty sources hold for the majority of other analytical
methods.Mostoftheuncertaintysourcesarelinkedtospecificstepsintheanalysisprocedure.It
is stressed that sample preparation is usually the biggest contributor to measurement
uncertainty. When performing chemical analysis then every care should be taken to minimize
(preferably eliminate) the influence of the uncertainty sources, as far as possible. And what
cannot be eliminated, has to be taken into account. It is not necessary to quantify every
uncertaintysourceindividually.Instead,itisoftenmorepracticaltoquantifyseveraluncertainty
sourcesjointly.
Measurementuncertaintysources
http://www.uttv.ee/naita?id=17587
https://www.youtube.com/watch?v=4y2cjJ8Jpsg
https://sisu.ut.ee/measurement 43
5.4.Treatmentofrandomandsystematic
effects
Briefsummary:Althoughwithinameasurementseriesrandomandsystematiceffectsinfluence
measurementresultsdifferently,theyaremathematicallytakenintoaccountthesamewayas
uncertaintycomponentspresentedasstandarduncertainties.
Treatmentofrandomandsystematiceffects
http://www.uttv.ee/naita?id=17712
https://www.youtube.com/watch?v=hdh5xVVZTbg
Inthecaseofpipetting(demonstratedandexplainedinsections2and4.1)therearethreemain
sources of uncertainty: repeatability, calibration uncertainty of the pipette and the temperature
effect.Theseeffectsinfluencepipettingindifferentways.
1.Repeatabilityisatypicalrandomeffect.Everypipettingoperationisinfluencedbyrandom
effects that altogether cause the differences between the volumes that are pipetted under
identicalconditions
2.Theuncertaintyduetocalibrationofthepipetteisatypicalsystematiceffect:Ifinsteadof
10.00 ml the mark on the pipette is, say, at 10.01 ml then the pipetted volume will be
systematicallytoohigh.Thismeansthatalthoughindividualpipettingresultscanbelower
than 10.01 ml (and in fact even below 10.00 ml), the average volume will be higher than
10.00ml:approximately10.01ml.
3.The temperature effect can be, depending on the situation, either systematic or random
effector(verycommonly)mixtureofthetwo.Whichwayitisdependsonthestabilityof
the temperature during repetitions (which is influenced by the overall duration of the
experiment).
Although the three uncertainty sources influence pipetting results in different ways they are all
taken into account the same way via uncertainty contributions expressed as standard
uncertainties.
In principle, it is possible to investigate the systematic effects, determine their magnitudes and
take them into account by correcting the results. When this is practical, this should be done. If
this is not done then the results will be biased, i.e. will be systematically shifted from the true
value.
An example where systematic effect can be determined and correction introduced with
reasonableeffortiscalibrationofpipette,explainedintheexampleinsection4.Twocaseswere
examined:withoutcorrectingandwithcorrecting:
1.In subsection 4.1 the calibration uncertainty of 0.03 ml as specified by the producer is
used. This corresponds to the situation that there is possibly a systematic effect the
possible The word possible means here that in fact there may be no systematic effect
theactualpipettevolumecanbe10.00ml.Wesimplydonotknow.differenceofthetrue
pipettevolumefromitsnominalvolume,butitisnotcloselyinvestigatedorcorrectedand
the uncertainty 0.03 ml is assigned to it, which with very high probability covers this
effect. As a result the standard uncertainty of calibration was u (V, CAL) = 0.017 ml
(rectangulardistributionisassumed).
2.In subsection 4.6 it was explained how to determine the actual volume of pipette by
calibration. It was found that the pipette volume was 10.006 ml. The calibration that was
carried out in the laboratory still has uncertainty, but this uncertainty now is due to the
repeatability during calibration (i.e. random effects) and is by almost 10 times smaller:
u(V,CAL)=0.0018ml.
https://sisu.ut.ee/measurement 44
Thus,whencorrectingforsystematiceffectscanbedonewithreasonableeffortthenitcanlead
to significant decrease of measurement uncertainty. However, in many cases accurate
determination of a systematic effect (accurate determination of bias) can involve a very large
effort and because of this can be impractical. It can also happen that the uncertainty of
correctionisnotmuchsmallerthantheuncertaintyduetopossiblebias.Infact,withreasonable
(i.e. not very large) effort the outcome of bias determination often is that there may be a
systematic effect and may not be. In such cases correction cannot be done and the uncertainty
duetotheeffecthastocoverthepossiblesystematiceffect(possiblebias).
***
[1]Thewordpossiblemeansherethatinfacttheremaybenosystematiceffecttheactual
pipettevolumecanbe10.00ml.Wesimplydonotknow.
https://sisu.ut.ee/measurement 45
6.Randomandsystematiceffectsrevisited
Brief summary: This section explains that whether an effect will influence the measurement
resultasarandomorasasystematiceffectdependsontheconditions.Effectsthatarerandom
inshorttermcanbecomesystematicinlongterm.Thisisthereasonwhyrepeatabilityisbyits
value smaller than withinlab reproducibility and the latter is in turn smaller than the combined
standard uncertainty. This section also explains that the A and B type uncertainty estimates do
notcorrespondonetoonetotherandomandsystematiceffects.
Howawithindaysystematiceffectcanbecomealongtermrandomeffect?
http://www.uttv.ee/naita?id=17713
https://www.youtube.com/watch?v=qObLSS7mfDo
Aneffectthatwithinashorttimeperiod(e.g.withinaday)issystematiccanoveralongertime
periodberandom.Examples:
1.Ifanumberofpipettingoperationsaredonewithinadayusingthesamepipettethenthe
difference of the actual volume of the pipette from its nominal volume (i.e. calibration
uncertainty)willbeasystematiceffect.Ifpipettingisdoneondifferentdaysandthesame
pipetteisusedthenitisalsoasystematiceffect.However,ifpipettingisdoneondifferent
daysanddifferentpipettesareusedthenthiseffectwillchangeintoarandomeffect.
2.An instrument is calibrated daily with calibration solutions made from the same stock
solution, which is remade every month. In this case the difference of the actual stock
solutionconcentrationanditsnominalconcentrationisasystematiceffectwithinadayand
also within few weeks. But over a longer time period, say, half a year, It cannot be
strictly defined, how long is longterm. An approximate guidance could be: one year is
good, several months (at least 45) is minimum. Of course it also depends on the
procedure. this effect becomes random, since a number of different sock solutions will
havebeeninuseduringthattimeperiod.
Conclusions:
1.Aneffect,whichissystematicinshorttermcanberandominlongterm
2.Thelongeristhetimeframethemoreeffectscanchangefromsystematicintorandom.
As explained in a past lecture if the measurement of the same or identical sample is repeated
under identical conditions (usually within the same day) using the same procedure then the
standard deviation of the obtained results is called repeatability standard deviation and
denoted as sr. If the measurement of the same or identical sample is repeated using the same
procedure but under changed conditions whereby the changes are those that take place in the
laboratory under normal work practices then the standard deviation of the results is called
withinlab reproducibility or intermediate precision and it is denoted as sRW. The terms
withinlab reproducibility and intermediate precision are synonyms. The VIM(1) prefers
intermediate precision. The Nordtest handbook(5) uses withinlab reproducibility (or
reproducibility within laboratory). In order to stress the importance of the longterm, in this
courseweoftenrefertosRWasthewithinlablongtermreproducibility
The conclusions expressed above are the reason why sr is smaller than sRW. Simply, some
effects that within day are systematic and are not accounted for by sr become random over a
longer time and sRW takes them into account. The combined standard uncertainty uc is in turn
larger that the intermediate precision, because it has to take into account all significant effects
that influence the result, including those that remain systematic also in the long term. This
importantrelationbetweenthesethreequantitiesisvisualizedinScheme6.1.
https://sisu.ut.ee/measurement 46
Scheme6.1.Relationsbetweenrepeatability,withinlabreproducibilityandcombined
uncertainty.
TherandomandsystematiceffectscannotbeconsideredtobeinonetoonerelationwithtypeA
andBuncertaintyestimation.Thesearecategoricallydifferentthings.Theeffectsrefertothe
intrinsiccausalrelationships,whiletypeAandBuncertaintyestimationrefersratherto
approachesusedforquantifyinguncertainty.Table6illustratesthisfurther.
Table6.1.InterrelationsbetweenrandomandsystematiceffectsandAandBtypesof
uncertaintyestimates.
Whenestimatingtheuncertaintycontributionsduetorandomeffects,thenitisimportantthata
number of repeated measurements are carried out. On the other hand, if, e.g. repeatability of
some analytical procedure is estimated then each repetition has to cover all steps in the
procedure, including sample preparation. For this reason making extensive repetitions is very
workintensive.Inthissituationtheconceptofpooledstandarddeviationbecomesveryuseful.
Itsessenceispoolingstandarddeviationsobtainedfromalimitednumberofmeasurements.The
followingvideoexplainsthis:
Pooledstandarddeviation
http://www.uttv.ee/naita?id=18228
https://www.youtube.com/watch?v=xsltS41PZW0
Dependingonhowtheexperimentsareplanned,thepooledstandarddeviationcanbeusedfor
calculatingofeitherrepeatabilitysrorwithinlabreproducibilitysRW.Theexperimentalplanand
calculationswhenfindingrepeatabilitysrareexplainedinthefollowingvideo:
Pooledstandarddeviationinpractice:estimatingrepeatability
http://www.uttv.ee/naita?id=18232
https://www.youtube.com/watch?v=DM_zf85PYic
Pooledstandarddeviationinpractice:estimatingwithinlablongtermreproducibility
http://www.uttv.ee/naita?id=18234
https://www.youtube.com/watch?v=nPJY8HfPxNs
https://sisu.ut.ee/measurement 47
***
[1]Itcannotbestrictlydefined,howlongislongterm.Anapproximateguidancecouldbe:one
year is good, several months (at least 45) is minimum. Of course it also depends on the
procedure.
[2]Thetermswithinlabreproducibilityandintermediateprecisionaresynonyms.TheVIM(1)
prefers intermediate precision. The Nordtest handbook (5) uses withinlab reproducibility (or
reproducibility within laboratory). In order to stress the importance of the longterm, in this
courseweoftenrefertosRWasthewithinlablongtermreproducibility.
***
The slides of the presentation and the calculation files with initial data only, as well, as with
calculationsperformedareavailablefromhere:
pooled_standard_deviation.pdf 18KB
pooled_standard_deviation_repeatability_initial.xlsx 16KB
pooled_standard_deviation_repeatability_solved.xlsx 16KB
pooled_standard_deviation_reproducibility_initial.xlsx 15KB
pooled_standard_deviation_reproducibility_solved.xlsx 16KB
https://sisu.ut.ee/measurement 48
Pooled Standard Deviation
If it is impossible to make many repeated
measurements with the same sample
Pooled Standard Deviation Then precision can be estimated during longer
time in the form of pooled standard deviation
Ivo Leito
Pooled standard deviation can be used to
University of Tartu
calculate:
Institute of Chemistry Repeatability
2013 Within-lab reproducibility
1 2
l d
spooled spooled
n1 n2 ... nk k k
Symbols: Symbols:
k number of samples k number of samples
s1, s2, etc are within sample standard deviations s1, s2, etc are within sample standard deviations
n1, n2, etc are numbers of measurements made for n1, n2, etc are numbers of repeated measurements with
different samples every sample
3 4
1
7.Precision,trueness,accuracy
Briefsummary: Interrelations between the different error types (random, systematic, total),
their corresponding performance characteristics (precision, trueness, accuracy) and the
parameters for quantitatively expressing these performance parameters (standard deviation,
bias,measurementuncertainty)areexplainedinthissection.
Interrelationbetweentheconceptsofprecision,trueness,accuracyandmeasurementuncertainty
http://www.uttv.ee/naita?id=17824
https://www.youtube.com/watch?v=NdDK03f0wew
Scheme7.1.Interrelationsbetweenthedifferenterrortypes,theperformance
characteristicsusedtoestimatethemandthewaysofexpressingtheestimates
quantitatively.ThistypeofschemewasoriginallypublishedinarticleA.Menditto,etal
Accred.Qual.Assur.2006,12,45.
Thedifferencebetweenthemeasuredvalueandthetruevalueiscallederrorortotalerror(see
section1).Thiserrorcanbedividedintotwopartsrandomerror (having different magnitude
and sign in the case of repeated measurements) and systematic error (having the same or
systematically changing magnitude and sign in the case of repeated measurements). As seen in
section 1 errors cannot be known exactly. Therefore instead of errors themselves we operate
withestimatesoferrorstheperformancecharacteristics.
Thus,truenessistheestimateofthesystematicerror.Fordeterminingtruenesswedonotneed
to know the true value but we need to know a reference value. Reference value (differently
from the true value) has uncertainty, but usually a small one. Different types of precision are
estimates of the random error. For obtaining the true precision we would need to make an
infinitenumberofrepeatedmeasurements.Therearedifferenttypesofprecision,dependingon
https://sisu.ut.ee/measurement 49
the conditions under which precision is determined, e.g. repeatability (section 1) and
intermediate precision (section 6). Accuracy embraces both trueness and precision and be
consideredasdescribingthetotalerror.
https://sisu.ut.ee/measurement 50
8.Overviewofmeasurementuncertainty
estimationapproaches
Briefsummary: In this section an overview is given about the main types of approaches that
canbeusedforestimationofmeasurementuncertainty.
Overviewoftheapproachesforestimatingmeasurementuncertainty
http://www.uttv.ee/naita?id=17704
https://www.youtube.com/watch?v=Ed8S8KN1GlU
https://sisu.ut.ee/measurement 51
This section presents step by step the modeling approach to measurement uncertainty
estimation. This approach is described in detail in the ISO GUM(2) and has been interpreted for
chemistryintheEurachemmeasurementuncertaintyguide(3).Itisoftencalledalsothebottom
up approach. This means that the uncertainties of the input quantities are found and thereafter
combined into the combined standard uncertainty. The uncertainty estimation carried out in
section 4 in principle also used this approach. The presentation in this section is based on a
practicalexampledeterminationofammoniumnitrogeninwater.
9.1.Step1Measuranddefinition
9.2.Step2Modelequation
9.3.Step3Uncertaintysources
9.4.Step4Valuesoftheinputquantities
9.5.Step5Standarduncertaintiesoftheinputquantities
9.6.Step6Valueoftheoutputquantity
9.7.Step7Combinedstandarduncertainty
9.8.Step8Expandeduncertainty
9.9.Step9Lookingattheobtaineduncertainty
Theslidesusedinthissectioncanbedownloadedfromhere:
uncertainty_of_photometric_nh4_determination_iso_gum_modeling.pdf 82KB
https://sisu.ut.ee/measurement 52
Measurement uncertainty by the
modeling approach:
Estimation of measurement
uncertainty in chemical analysis Determination of NH4+ in water
(analytical chemistry)
Ivo Leito A dye (photometric complex) is formed
University of Tartu
Institute of Chemistry
quantitatively from NH4+ and its absorbance is
measured at 640-660 nm by a photometer
2013
Compiling calibration
Photometric
graph, b1, b0
measurement, Asample Our measurand:
Concentration of NH4+ expressed as ammonium
Ammonium nitrogen concentration CN_sample mg/l in the water sample
content in sample from
calibration graph,
CN_sample
3 4
The result is
Sampling expressed for
All possible uncertainty sources need to be Sample non-representativeness sample, sampling is
considered not included
Step 4 Finding values of input quantities Step 4 Finding values of input quantities
A = b0 + b1C
Calibration graph A = 0.017 + 0.981C Quantity Value Unit
1.2
Asample 0.1860 AU*
1
b0 0.0171 AU*
0.8
b1 0.9808 AUl/mg
Absorbance
0.6
0.4
fd 1.2500
0.2
Cdc 0.0000 mg/l
0
0 0.2 0.4 0.6 0.8 1 1.2
NH4+ (mg/l)
* Absorbance in fact does not have a unit, AU is used for clarity
11 12
Absorbance of the sample solution Asample: Standard deviations of b0 and b1 as found from
regression statistics are used as standard
u (A sample, rep) = 0.0010 AU uncertainty estimates
u (A sample, drift) = 0.0012 AU
u (A sample, chem) = 0.0030 AU
u(b0) = 0.0025 AU
u ( Asample , rep) u ( Asample , drift )
2 2 u(b1) = 0.0046 AUl/mg
u ( Asample ) 0.0034 AU
u ( Asample , chem) 2
This is an approximation!
13 14
Step 5 Standard uncertainties of the input Step 5 Standard uncertainties of the input
quantities: fd quantities: Cdc
15 16
19 20
( Asample b0 )
CN_sample fd Cdc
b1 23 24
9.1.Step1Measuranddefinition
Definitionofthemeasurand
http://www.uttv.ee/naita?id=17636
https://www.youtube.com/watch?v=P_i268pDgvM
The measurand definition is the most basic step of any measurement. In this step it is defined
what is actually measured and this definition is also the basis for the measurement procedure
andmodelequation.
https://sisu.ut.ee/measurement 53
9.2.Step2Modelequation
Modelequation
http://www.uttv.ee/naita?id=17637
https://www.youtube.com/watch?v=N45YNwToyac
The model equation (equation 9.1) enables calculating the outputquantity value (result value)
from the input quantity values. Input quantities are the directly measured quantities (or are
calculated from directly measured quantities). In addition, the model equation has to enable
accounting for all important measurement uncertainty sources. In the case of this analysis the
modelequationisthefollowing:
(9.1)
TheoutputquantityisCN_sampleammoniumionconcentrationinthewatersample.
Theinputquantitiesare:
Asampleabsorbanceofthedyesolutionobtainedfromthesample
b1andb0slopeandinterceptofthecalibrationgraph
f ddilutionfactor
Cdccomponenttakingintoaccountuncertaintyoriginatingfrompossibledecompositionor
contamination.
The parameters Asample, b1, b0 and f d in the equation account for the directly measured input
quantities.Incontrast,Cdcisintroducedonlyfortakingintoaccountuncertaintyduetopossible
decomposition of the photometric complex and due to possible contamination. Its value is zero
as will be seen in section 9.4, so, it does not contribute to the value of CN_sample. However, its
uncertaintyisdifferentfromzeroandthereforewillcontributetotheuncertaintyofCN_sample.
https://sisu.ut.ee/measurement 54
9.3.Step3Uncertaintysources
Allpossibleuncertaintysourceshavetobeconsideredandthosethatarelikelytobeinfluential
havetobetakenintoaccount.
Uncertaintysources:ingeneral
http://www.uttv.ee/naita?id=17638
https://www.youtube.com/watch?v=EWYNWtRgvmI
Uncertaintysources:onebyone
http://www.uttv.ee/naita?id=17639
https://www.youtube.com/watch?v=kEfhKCj44N0
The following schemes list the main uncertainty sources in chemical analysis and comment on
the presence or absence of the respective uncertainty sources in our case. Comment on the
memory effect as uncertainty source: Memory effect is problematic first of all in the case of
trace analysis and secondly, if the analyte is specifically prone to adsorption on glass or plastic
surfaces. In this case, although the concentration of ammonium in the sample is quite low, it is
still not yet a true trace analysis. Ammonium ion is by no means a strongly adsorbing species.
Thereforewecanleavethememoryeffectasuncertaintysourceoutofconsideration.
https://sisu.ut.ee/measurement 57
***
[1]Commentonthememoryeffectasuncertaintysource:Memoryeffectisproblematicfirstof
allinthecaseoftraceanalysisandsecondly,iftheanalyteisspecificallypronetoadsorptionon
glassorplasticsurfaces.Inthiscase,althoughtheconcentrationofammoniuminthesampleis
quitelow,itisstillnotyetatruetraceanalysis.Ammoniumionisbynomeansastrongly
adsorbingspecies.Thereforewecanleavethememoryeffectasuncertaintysourceoutof
consideration.
https://sisu.ut.ee/measurement 58
9.4.Step4Valuesoftheinputquantities
Findingthevaluesoftheinputquantities
http://www.uttv.ee/naita?id=17640
https://www.youtube.com/watch?v=TP8AguK1jSo
ThevaluesofAsample,b1,b0andfdarefoundfromthemeasureddata(b1andb0arefound
fromregressionanalysis).ThevalueofCdciszero.
https://sisu.ut.ee/measurement 59
9.5.Step5Standarduncertaintiesofthe
inputquantities
Standarduncertaintiesoftheinputquantities
http://www.uttv.ee/naita?id=17641
https://www.youtube.com/watch?v=ZIEhfCoNyxU
ThestandarduncertaintyofAsampleisfoundfromthefollowingthreeuncertaintycomponents:
(1)Theuncertaintyduetotherepeatabilityofphotometricmeasurement:
u(Asample,rep)=0.0010AU(9.2)
This uncertainty includes the repeatability of the instrument, repeatability of positioning the cell
in the instrument and possible disturbances, such as a random dust particle on the optical
windowsofthecell.
(2)Theuncertaintyduetothepossibledriftofthespectrophotometerparameters:
u(Asample,drift)=0.0012AU(9.3)
(3)Theuncertaintyduetothepossibleinterferingeffects:
u(Asample,chem)=0.0030AU(9.4)
Thesecanbeduetosomecompoundabsorbing(orscattering)lightatthesamewavelengththat
is used for measurement (leading to increase of the absorbance value) or due to some
disturbanceinformingthephotometriccomplex(leadingtodecreaseoftheabsorbancevalue).
Asaresult:
(9.5)
The standard uncertainties of the slope b1 and intercept b0 are found as standard deviations of
the respective regression coefficients (see the XLS files in section 9.7). This is an approximate
way of taking into account the uncertainty due to linear regression analysis, because it (1)
neglectsthesystematiceffectsaffectingallthepointsontheregressionlineand(2)neglectsthe
negativecorrelationbetweenb1andb0(whichalwaysexists).Thefirstoftheseeffectsleadsto
underestimation of uncertainty and the second one to overestimation of uncertainty. So, this
approach should only be used if it is not expected that linear regression analysis will be among
themaincontributorstouncertainty. Thesameexamplesolvedwithfullrigorisavailablefrom
http://www.ut.ee/katsekoda/GUM_examples/. Please look at the example Ammonium by
Photometry with elaboration level High (uncertainty estimated at full rigor, suitable for
experts). Comparison of the obtained combined uncertainties: 0.00686 mg/l obtained here
(section 9.7) and 0.0065 mg/l obtained with full rigor shows that this approach is acceptable,
especiallysinceitleadsrathertouncertaintyoverestimationthanunderestimation.
The u(f d) is found on an assumption that the relative combined standard uncertainty of all
involvedvolumetricoperationsisnothigherthan0.5%.Ifvolumetricoperationsarecarriedout
carefully then this is a safe assumption under usual laboratory conditions. Considering that the
valueoff dis1.25(unitless)wegetthefollowing:
u(fd)=1.250.5%/100%=0.0065(unitless)(9.6)
https://sisu.ut.ee/measurement 60
The uncertainty of Cdc accounts for possible decomposition of the photometric complex and
possible contamination of the sample. The word possible is stressed here: it is well possible
thatactuallythereisneitherdecompositionofthephotometriccomplexnorcontaminationofthe
sample. However, in order to rigorously establish this, extensive research would be needed.
Thereforeinthisexampleweuseanestimatebasedonexperiencefromourlaboratory:
u(Cdc)=0.004mg/l
Thestandarduncertaintiesoftheinputquantitiesaresummarizedintable:
***
[1] The same example solved with full rigor is available from
http://www.ut.ee/katsekoda/GUM_examples/. Please look at the example Ammonium by
Photometry with elaboration level High (uncertainty estimated at full rigor, suitable for
experts). Comparison of the obtained combined uncertainties: 0.00686 mg/l obtained here
(section 9.7) and 0.0065 mg/l obtained with full rigor shows that this approach is acceptable,
especiallysinceitleadsrathertouncertaintyoverestimationthanunderestimation.
https://sisu.ut.ee/measurement 61
9.6.Step6Valueoftheoutputquantity
Calculatingthevalueoftheoutputquantity
http://www.uttv.ee/naita?id=17642
https://www.youtube.com/watch?v=OKzpnlMsjJM
The output quantity value (the measurand value) is calculated from the input quantity values
(section 9.4) using the mathematical model (section 9.2). The measurand value in this example
is:CN_sample=0.215mg/l.
https://sisu.ut.ee/measurement 62
9.7.Step7Combinedstandarduncertainty
Calculatingthecombinedstandarduncertainty
http://www.uttv.ee/naita?id=17643
https://www.youtube.com/watch?v=v6JZT2PkEM
Numericalcalculationoftheuncertaintycomponents:theKragtenmethod
http://www.uttv.ee/naita?id=17721
https://www.youtube.com/watch?v=qRx8cFVitgk
Theinfluenceofmeasurementuncertaintysources,groupedaccordingtoinputquantities,onthe
resultcanbeschematicallypresentedintheformofthesocalledfishbonediagram:
https://sisu.ut.ee/measurement 63
Scheme9.1.Causeeffectdiagram:Photometricammoniadetermination
TheinitialXLSfile(i.e.containingonlythedatabutnotthecalculations)usedinthisexample
andtheXLSfilecontainingthealsothecombinedstandarduncertainty(andexpanded
uncertainty)calculationaccordingtotheKragtensapproachcanbedownloadedfromhere:
uncertainty_of_photometric_nh4_determination_kragten_initial.xls 49KB
uncertainty_of_photometric_nh4_determination_kragten_solved.xls 52KB
https://sisu.ut.ee/measurement 64
9.8.Step8Expandeduncertainty
Theexpandeduncertaintycanbefoundattwodifferentlevelsofsophistication.Thesimpler
approachusessimplyapresetkvalue(mostoften2)andtheactualcoverageprobabilityisnot
discussed.Thisapproachispresentedinthefirstvideolecture.
Findingtheexpandeduncertainty(simplerapproach)
http://www.uttv.ee/naita?id=17644
https://www.youtube.com/watch?v=KomDnLRArDs
Thesecondapproachismoresophisticated.Itisanapproximationapproachbasedonthe
assumptionthatthedistributionfunctionoftheoutputquantitycanbeapproximatedbyaStudent
distributionwiththeeffectivenumberofdegreesoffreedomfoundbythesocalledWelch
Satterthwaitemethod.ThisenablesthentousetheStudentcoefficientcorrespondingtoa
desiredlevelofconfidence(coverageprobability)asthecoveragefactor.Thisapproachis
explainedinthesecondvideolecture.
Findingtheexpandeduncertainty(theWelchSatterthwaitemethod)
http://www.uttv.ee/naita?id=17916
https://www.youtube.com/watch?v=CylWJjG_8ck
TheXLSfilecontainingthecombinedstandarduncertaintyandexpandeduncertaintycalculation
andtheXLSfilecontainingtheexpandeduncertaintycalculationusingcoveragefactorfound
usingtheeffectivenumberofdegreesoffreedomformtheWelchSatterthwaiteapproachcanbe
downloadedfromhere:
uncertainty_of_photometric_nh4_determination_kragten_solved.xls 52KB
uncertainty_of_photometric_nh4_determination_kragten_solved_df.xls 48KB
https://sisu.ut.ee/measurement 65
9.9.Step9Lookingattheobtained
uncertainty
Calculationoftheuncertaintycomponents(uncertaintyindexes)isexplainedanddemonstrated
insection9.7.
Analysisoftheuncertaintysources
http://www.uttv.ee/naita?id=17645
https://www.youtube.com/watch?v=zE5yi6NwAvI
Conclusions1
http://www.uttv.ee/naita?id=17647
https://www.youtube.com/watch?v=7VqqLGLQHgg
Conclusions2
http://www.uttv.ee/naita?id=18096
https://www.youtube.com/watch?v=ENg2aTyBwKg
https://sisu.ut.ee/measurement 66
Briefsummary:This section explains the socalled singlelab validation approach. We will look
at the formalization of this approach published by Nordtest. Handbook for Calculation of
MeasurementUncertaintyinEnvironmentalLaboratories.B.Magnusson,T.Nykki,H.Hovind,M.
Krysell. Nordtest technical report 537, ed. 3. Nordtest, 2011. Available online from
http://www.nordtest.info/index.php/technicalreports/item/handbookforcalculationof
measurementuncertaintyinenvironmentallaboratoriesnttr537edition3.html Therefore in
this course this approach is often called the Nordtest approach. The singlelab validation
approach, contrary to the ISO GUM modeling approach, does not go deeply into the
measurement procedure and does not attempt to quantify all uncertainty sources individually.
Insteaduncertaintysourcesarequantifiedinlargebatchesviacomponentsthattakeanumber
of uncertainty sources into account. Most of the data that are used come from validation of the
analytical procedure. This is the reason for the word validation in the name of the approach.
Thistypeofapproachisalsosometimescalledthetopdownapproach.
Sections 10.1 to 10.3 present the Nordtest approach step by step and explain the way of
obtainingnecessarydata.Section10.4givesaroadmapoftheNordtestapproach.Section10.5
presents a practical example of applying the Nordtest approach in the case of determination of
acrylamideinsnacksbyliquidchromatographymassspectrometry(LCMS).
10.1.Principles
10.2.Uncertaintycomponentaccountingforrandomeffects
10.3.Uncertaintycomponentaccountingforsystematiceffects
10.4.Roadmap
10.5.DeterminationofacrylamideinsnacksbyLCMS
***
***
Theslidespresentedinthissectionareavailablefromhere:
singlelab_validation_nordtest_uncertainty.pdf 33KB
https://sisu.ut.ee/measurement 68
Single-laboratory validation approach
Uncertainty estmation approach Effects contributing to
based on validation and Quality uncertainty
This and subsequent equations work with absolute In general: whichever is more constant
and relative values
Appendix E.4 from Quantifying Uncertainty in Analytical Measurement,
Nordtest Technical Report 537, 3rd ed (2011) EURACHEM/CITAC Guide, Second Edition (2000)
http://www.nordtest.info/ 3 Available from : http://www.eurachem.org/ 4
Single lab validation approach: in practice u(Rw) is the uncertainty component that u(Rw)
takes into account long-term variation of
results within lab, that means: within-lab
reproducibility (sRw)
Steps of the process: Include sample
Ideally: preparation!
The same sample
1. Specify measurand Sample similar to test samples matrix,
2. Quantify Rw component u(Rw) concentration, homogeneity
The same lab
3. Quantify bias component u(bias) The same procedure
4. Convert components to standard uncertainties u(x) Different days (preferably over 1 year)
Different persons
5. Calculate combined standard uncertainty uc Different reagent batches
6. Calculate expanded uncertainty U
5 6
1
u(Rw) u(bias)
The possible bias of labs results from the
best estimate of true value is taken into
u(Rw) = sRw
account
Include sample
preparation!
u(bias) can be found:
From repeated analysis of the same samples with a
Ideally: separately for The control sample reference procedure
different matrices and analysis has to cover From repeated analysis of certified reference materials
different concentration the whole analytical (CRMs)
levels! process From repeated interlaboratory comparison measurements
From repeated spiking experiments
u(bias) u(bias)
The averaging is done using the root mean
square:
u (bias ) RMS bias u (Cref ) 2 (biasi ) 2
2
11 12
2
Roadmap:
Possible bias u (Cref i ) from certificates
u (Cref i )
si
u (Cref )
u(Cref ) i
2
ni n
biasi Clabi Cref i
RMSbias
(biasi ) 2
u (bias ) RMS bias u (Cref ) 2
2
n
3
10.Thesinglelabvalidationapproach
10.1.Principles
IntheNordtestapproachtheuncertaintyisregardedasbeingduetotwocomponents:
Introductiontouncertaintyestmationbasedonvalidationandqualitycontroldata(theNordtestapproach)
http://www.uttv.ee/naita?id=17909
https://www.youtube.com/watch?v=9oOX4CUsWjI
ThemainequationoftheNordtestapproachishere:
(10.1)
Here u (Rw) stands for the withinlab reproducibility component of uncertainty and u (bias)
stands for the uncertainty component taking into account possible bias. The resulting
measurementuncertaintyucisnotdirectlyrelatedtoanyspecificresult,becauseitiscalculated
using data from the past measurements. Therefore it can be said that the uncertainty obtained
withtheNordtestapproachcharacterizestheanalysisprocedureratherthanaconcreteresult.If
theuncertaintyofaconcreteresultisneededthenitisassignedtotheresult.
Because of this it is necessary to decide whether to express the uncertainty in absolute terms
(i.e.intheunitsofthemeasuredquantity)orintherelativeterms(i.e.asaratioofuncertainty
to the value of the measured quantity or a percentage of the value of the measured quantity).
Therulesofthumb:
Atlowconcentrations(neardetectionlimit,tracelevel)useabsoluteuncertainties
Uncertaintyisnotmuchdependentonanalytelevel
Atmediumandhigherconcentrationsuserelativeuncertainties
Uncertaintyisroughlyproportionaltoanalytelevel
Ingeneral:usewhicheverismoreconstantwithchangingconcentration.
ThemainequationoftheNordtestapproach.Absoluteandrelativequantities
https://sisu.ut.ee/measurement 69
http://www.uttv.ee/naita?id=17911
https://www.youtube.com/watch?v=MH8CixjySjI
OverviewofthepracticalimplementationoftheNordtestapproach
http://www.uttv.ee/naita?id=17912
https://www.youtube.com/watch?v=hPrncfXr7Ok
The main steps of the process of measurement uncertainty evaluation with the Nordtest
approach:
1.Specifymeasurand
2.QuantifyRwcomponentu(Rw )
3.Quantifybiascomponentu(bias)
4.Convertcomponentstostandarduncertaintiesu(x)
5.Calculatecombinedstandarduncertaintyuc
6.CalculateexpandeduncertaintyU
***
[1] A typical example is titration if new titrant is prepared weekly. Within a given week the
titrant concentration is a systematic effect, but in the long term it becomes random, because
many batches of titrant will be involved. A similarly typical example is calibration graph if it is
prepared daily: every day the possible bias in calibration is a systematic effect, but in the long
termitbecomesrandom.
https://sisu.ut.ee/measurement 70
10.2.Uncertaintycomponentaccountingfor
randomeffects
Estimatingthewithinlabreproducibilitycomponentofuncertainty
http://www.uttv.ee/naita?id=17913
https://www.youtube.com/watch?v=9803VViKo
The withinlab reproducibility (intermediate precision) component takes into account all
uncertaintysourcesthatarerandominthelongterm(i.e.severalmonths,preferablyoneyear).
So, quite some uncertainty sources that are systematic within a day will become random in the
long term. The simplest way to find u (Rw) is from a number of repeated measurements of a
control sample, organized, e.g. as a control chart. Alternatively, the pooled standard deviation
approach as explained in section 6 can be used. If this is done then the u (Rw) can be found
basedonseveraldifferentcontrolsamples,sothatitwillbeanaveragevalueofallofthem.
Thenumberofvaluesusedforevaluationofu(Rw)mustbesufficientlylarge.Aninitialestimate
ofu(Rw)canbeobtainedwith1015valuesbutthereaftermoredatashouldbecollected.Even
moreimportantly,thetimeperiodduringwhichthedataarecollected,mustbesufficientlylong
(at least several months, preferably around a year) so that all the sources of variability in the
procedure are taken into account. So, 10 values collected over a fivemonth time period is a
betteroptionthan20valuescollectedduring1.5months.
Dependingonsituationu(Rw)canbeusedasabsoluteorrelativevalue.
It is important to stress that u (Rw) should be estimated separately for different matrixes and
differentconcentrationlevels.
https://sisu.ut.ee/measurement 71
10.3.Uncertaintycomponentaccountingfor
systematiceffects
Estimatingtheuncertaintycomponentduetopossiblebias
http://www.uttv.ee/naita?id=17910
https://www.youtube.com/watch?v=hLGZsW_o81o
Thecomponentu(bias)takesintoaccountpossiblebiasofthemeasurementprocedure.
Reliablydeterminingthebiasoftheprocedureisnoteasyforthefollowingreasons:
1.If the random effects are strong then this needs a very large number of measurements.
When a limited number of measurements are made then the bias estimate will always
contain a contribution from random effects, which will make the bias estimate artificially
higher.Evenmoresotheprocedurecanactuallyhavenobiasatall.
2.Even with formally similar matrixes the bias can differ by magnitude and even by sign
(e.g. when determining pesticides in different varieties of apples, determining drug
residuesinbloodplasmafromdifferentpatients,etc).Thismeansthathavingdetermined
the bias in one variety of apples this bias is not automatically applicable for another
varietyandwecanspeakaboutuncerteintyinapplyingthebiastoanothervariety.
3.Biasisalwaysdeterminedagainstareferencevalue,whichalsohasanuncertainty.
Bias refers to difference between our measured value and a reference value. Therefore, for
findingu(bias)weneedasampleoramaterialwithareferencevalue.Inbroadtermsthereare
fourdifferentpossibilitieshowtheu(bias)canbedetermined:
In the case of all four possibilities it is critical to include also sample preparation in bias
determination.
In the case of all four possibilities it is necessary to make an as large as possible number of
replicate measurements in order to separate the bias from random effects as efficiently as
possible. Bias generally changes from matrix to matrix and usually is different at different
concentration levels. So, it is important to use several CRMs, several interlaboratory
comparisons,etc.
Theloweristhereliabilityofthereferencevaluethehigheristheu(bias)estimate.
Theu(bias)componentisfoundaccordingtothefollowingequation:
(10.2)
RMSbiasistheaverage(rootmeansquare)biasandisfoundasfollows:
(10.3)
Wherenisthenumberofbiasdeterminationscarriedoutandeachbiasiisaresultofan
individualbiasdeterminationandisfoundasfollows:
(10.4)
Where Clabi is a mean of the results of analyte determination in the reference sample (e.g. in
theCRM)obtainedbythelaboratoryandCref iisthereferencevalueofthereferencesample.It
isimportantthatClabicorrespondstoanumberofreplicates.
u (Cref) is the average standard uncertainty of the reference values of the reference samples
andisfoundasfollows:
(10.5)
Hereu(Cref i)isthestandarduncertaintyoftheithreferencevalue.InthecaseofCRM
analysis,spikingoranalysiswithareferenceproceduretheu(Cref i)canusuallybereasonably
found.However,inthecaseofinterlaboratorycomparisonswheretheconsensusvalueofthe
participantsisusedasthereferencevalueareliableuncertaintyofthereferencevaluecannot
befound.Thebestestimateinthatcasewouldbethestandarddeviationoftheaveragevalue
aftereliminationofoutliers:
https://sisu.ut.ee/measurement 73
(10.6)
Heresi is the standard deviation of the participants in the ith intercomparison after elimination
ofoutlayersandniisthenumberofparticipants(againaftereliminatingtheoutliers)inthe ith
intercomparison.
InthespecialcaseifanumberofbiasdeterminationswerecarriedoutusingonesingleCRMthe
equation10.2changesintothefollowingform:
(10.7)
Wheresbiasisthestandarddeviationofthebiasestimatesobtainedandnisthenumberofbias
estimatesobtained.
Dependingonsituationtheu(bias)canbeusedasabsoluteorasrelativevalue.
Finally, it is important to stress that the bias uncertainty component should be estimated
separatelyfordifferentmatrixesanddifferentconcentrationlevels.
***
[1]Insimplifiedtermscertifiedreferencematerialisamaterial,inwhichthecontentofthe
analyte(oranalytes)isreliablyknown(thematerialhasacertificate).Ifthecertifiedreference
materialsmatrixissimilartorealsamples(i.e.itisnotapurecompoundorasolutionbutis
e.g.milkpowder,soilorbloodplasma)thenwecallitcertifiedmatrixfererencematerial.
https://sisu.ut.ee/measurement 74
10.4.Roadmap
RoadmapofuncertaintyestimationusingtheNordtestapproach
http://www.uttv.ee/naita?id=17914
https://www.youtube.com/watch?v=KNjUAiq5mEQ
Scheme10.1.Roadmapofmeasurementuncertaintyestimationusingthesinglelab
validationapproach(theNordtestapproach).
https://sisu.ut.ee/measurement 75
10.5.Determinationofacrylamideinsnacks
byLCMS
Nordtestapproachinpractice:DeterminationofacrylamideinsnacksbyLCMS Pleasenotethatinthe
tablesoftheslide"MeasurementswiththeCRMs"themeasurementunitsmustbeg/kg,notmg/l.
http://www.uttv.ee/naita?id=18163
https://www.youtube.com/watch?v=P94xWjC6Og0
Somecommentsonthisexample:
1. We use relative uncertainties in this example. The reason is that the concentration in these
samplesisquitehighandinLCMSanalysisvariabilityoftheresultsisoftenroughlyproportional
tothevaluesoftheresults.
2.InthisexampleweassumethatthematrixesoftheusedCRMspotatochipsandcrispbread
are sufficiently similar. This means that both reproducibility and bias obtained with these
matrixes are similar. In such case the calculated uncertainty is applicable to both of these
matrixes.Ifafterwardswewillanalysesomematrixthatisdifferentfrompotatochipsandcrisp
bread then we cannot apply to that result the uncertainty estimate that we obtained here.
Assessing whether a new matrix is sufficiently similar to the one used for bias evaluation is
usuallybasedonexperience.
(10.8)
Inprincipleitcanbeassignedtoaresultofanymagnitude,however,itwouldnotbecorrectto
apply this uncertainty to analysis results that are very different from the one presented here.
Thedifferenceshouldnotbehigherthan34times.
WhenweusetheNordtestapproachforuncertaintyestimationtheningeneralwecanassume
thatthenumberofdegreesoffreedomissufficientlylargesothatthek=2uncertaintycanbe
assumedtohaveroughly95%coverageprobability.
***
[1]Pleasenotethatinthetablesoftheslide"MeasurementswiththeCRMs"themeasurement
unitsmustbeg/kg,notmg/l.
The slides of this example and the calculation files both the initial and the solved file are
availablefromhere:
nordtest_uncertainty_example_acrylamide_lcms.pdf 153KB
nordtest_uncertainty_example_acrylamide_lcms_initial.xls 32KB
nordtest_uncertainty_example_acrylamide_lcms_solved.xls 33KB
https://sisu.ut.ee/measurement 76
Single-lab validation approach in practice:
Single-lab validation approach Determination of acrylamide in crisp bread
in practice: by LC-MS
Determination of acrylamide in crisp bread Concentration level 998 g/kg
by LC-MS
Laboratory has analysed two certified reference
Ivo Leito materials (CRMs) with similar matrixes
University of Tartu
Institute of Chemistry Potato chips and crisp bread
2013 The crisp bread CRM is also used as a control sample
1 2
ni n
biasi Clabi Cref i
u(Rw) = sRW = 31 g/kg
RMSbias
(biasi ) 2
u(bias ) RMS bias u(Cref ) 2
2
n u(Rw)_rel = sRW_rel = 31/1150100 = 2.70 %
1
Finding u(bias) Result:
u c_rel = 4.8 %
Ref value U (k =2) uc Lab result u c_rel bias i bias_rel
uc = 48 g/kg
g/kg g/kg g/kg g/kg % g/kg %
1179 68 34 1150 2.88 -29 -2.45 U _rel (k =2) = 9.5 %
860 42 21 832 2.44 -28 -3.24 U (k =2) = 95 g/kg
u (C ref)_rel= 2.67 %
RMS bias_rel=
u (bias)_rel =
2.87 %
3.92 %
Acrylamide content in the sample
7 8
2
11.Comparisonoftheapproaches
Brief summary: This section summarized the main properties of the uncertainty estimation
approaches,theiradvantagesanddrawbacks.
Whenever possible, one of the socalled singlelab approaches should be used. The
interlaboratory approaches are only suitable for getting very crude uncertainty estimates. We
recommend using them only in case when the laboratory actually does not have the
measurementinplaceyetandwantstoknow,approximatelywhatuncertaintycanbeobtained.
If the laboratory has competence and time to carry out investigations of the analytical
procedures the the modeling approach is often suitable. If the laboratory has limited time, but
has validation and quality control data then the singlelab validation approach is the most
suitable.
Comparisonofmeasurementuncertaintyestimationapproaches
http://www.uttv.ee/naita?id=17917
https://www.youtube.com/watch?v=TlpJ1c9Rx8
Thefollowingtablesummarizestheprosandconsoftheapproaches:
https://sisu.ut.ee/measurement 78
Briefsummary:Thissectionexplainsthatmeasurementuncertaintyestimatesareindispensable
ifwewanttocomparetwomeasurementresults.
Comparingmeasurementresultsusingmeasurementuncertaintyestimates
http://www.uttv.ee/naita?id=18095
https://www.youtube.com/watch?v=l6nYn6Pe7f0
https://sisu.ut.ee/measurement 79
12.Comparingmeasurementresults
13.Additionalmaterialsandcasestudies
Briefsummary: This section collects some additional materials, examples and case studies on
specificchemicalanalysistechniques.
13.1.Differentanalyticaltechniques
13.2.DissolvedoxygenbyWinklermethod
13.3.CoulometricKFtitration
https://sisu.ut.ee/measurement 80
13.1.Differentanalyticaltechniques
Alargenumberofmeasurementuncertaintyestimationexamples(exampleuncertaintybudgets)
isavailablefromthefollowingaddress:http://www.ut.ee/katsekoda/GUM_examples/
https://sisu.ut.ee/measurement 81
13.2.DissolvedoxygenbyWinklermethod
Dissolved oxygen (DO) is one of the most important dissolved gases in water. Sufficient
concentrationofDOiscriticalforthesurvivalofmostaquaticplantsandanimals[2]aswellas
in waste water treatment. DO concentration is a key parameter characterizing natural and
wastewaters and for assessing the state of environment in general. Besides dissolved CO2, DO
concentration is an important parameter shaping our climate. It is increasingly evident that the
concentrationofDOinoceansisdecreasing[36].
Accurate measurements of DO concentration are very important for studying these processes,
understandingtheirroleandpredictingclimatechanges.
The Winkler titration method is considered the most accurate method for DO concentration
measurement. Careful analysis of uncertainty sources relevant to the Winkler method was
carried out and the results are presented as a Report on improved highaccuracy Winkler
methodfordeterminationofdissolvedoxygenconcentration.
In this report it is described how the Winkler method was optimized for minimizing all
uncertainty sources as far as practical. The most important improvements were: gravimetric
measurement of all solutions, pretitration to minimize the effect of iodine volatilization,
accurate amperometric end point detection and careful accounting for dissolved oxygen in the
reagents. As a result, the developed method is possibly the most accurate method of
determination of dissolved oxygen available. Depending on measurement conditions and on the
dissolved oxygen concentration the combined standard uncertainties of the method are in the
rangeof0.0120.018mgdm 3correspondingtothek=2expandeduncertaintyintherangeof
0.023 0.035 mg dm 3 (0.27 0.38%, relative). This development enables more accurate
calibrationofelectrochemicalandopticaldissolvedoxygensensorsforroutineanalysisthanhas
been possible before. Most of this report is based on the article I. Helm, L. Jalukse, I. Leito,
Anal.Chim.Acta.741(2012)2131(ref[1]).
Preparation of this report was supported by the European Metrology Research Programme
(EMRP),projectENV05"Metrologyforoceansalinityandacidity".
1.I.Helm,L.Jalukse,I.Leito,Anal.Chim.Acta.741(2012)2131.
2.R.F.Keeling,H.E.Garcia,PNAS,99(12)(2002)78487853.
3.G.Shaffer,S.M.Olsen,J.O.P.Pedersen,Nat.Geosci.2(2009)105109.
4.R.F.Keeling,A.Krtzinger,N.Gruber,Annu.Rev.Mar.Sci.2(2010)199229.
5.D.Gilbert,N.N.Rabalais,R.J.Diaz,J.Zhang,Biogeosci.Discuss.6(2009)91279160.
6.J.P.A.Hobbs,C.A.McDonald,JournalFishBiol.77(2010)12191229.
g_winkler_report_280613.pdf 1.95MB
https://sisu.ut.ee/measurement 82
13.3.CoulometricKFtitration
Coulometric Karl Fischer method is currently the most accurate method of moisture
measurement. A Survey of the factors determining the uncertainty of coulometric Karl
Fischertitrationmethodhasbeencarriedout.
This survey gives an overview of the factors that determine the uncertainty of coulometric Karl
Fischer (cKF) method for water determination. Distinction is made between uncertainty sources
originating from the cKF method itself and uncertainty sources due to sample handling. The
compounduncertaintysourcesrepeatability,reproducibilityandbiasthatactuallyincorporate
thecontributionsfromthesetwoclassesofuncertaintysourcesarealsobrieflydiscussed.
Based on the literature data the most influential uncertainty sources of coulometric KF titration
method are possible chemical interferences, instrument instability and the accuracy of the end
point determination. The uncertainty sources due to sample handling are more problematic with
solid samples than with liquid samples. The most important sample preparation related
uncertainty sources are change of water content in the sample before the measurement and
incompletetransferofwaterfromthesampletothereactionvessel.
ThegeneralconclusionisthatalthoughtheuncertaintysourcesofthecKFmethodareingeneral
rather well known the discussion is almost always either qualitative or limited to the compound
uncertainty sources and there is very limited quantitative information available on the
contributionsoftheactualuncertaintysources.
Preparation of this report was supported by the European Metrology Research Programme
(EMRP),projectSIB64METefnet"MetrologyforMoistureinMaterials".
coulometric_kf_titration_measurement_uncertainty_sources_survey.pdf 227KB
https://sisu.ut.ee/measurement 83
14.TestsandExercises
Thissectioncontainsacompilationofallthetestsandexercisesofthiscourse.
Theconceptofmeasurementuncertainty(MU)Selftest1
TheoriginofmeasurementuncertaintySelftest2
TheNormaldistributionSelftest3.1
Mean,standarddeviationandstandarduncertaintySelftest3.2
StandarddeviationofthemeanSelftest3.4
RectangularandtriangulardistributionSelftest3.5
TheStudentdistributionSelftest3.6
QuantifyinguncertaintycomponentsSelftest4.1
CalculatingthecombinedstandarduncertaintySelftest4.2
LookingattheobtaineduncertaintySelftest4.3
ExpandeduncertaintySelftest4.4
PresentingmeasurementresultsSelftest4.5
SourcesofuncertaintySelftest5.3
TreatmentofrandomandsystematiceffectsSelftest5.4
RandomandsystematiceffectsrevisitedSelftest6
Precision,trueness,accuracySelftest7
OverviewofmeasurementuncertaintyestimationapproachesSelftest8
ModelequationSelftest9.2
StandarduncertaintiesoftheinputquantitiesSelftest9.5
TheISOGUMModelingapproachSelftest9AandSelftest9B
UncertaintycomponentaccountingforrandomeffectsSelftest10.2
UncertaintycomponentaccountingforsystematiceffectsSelftest10.3
DeterminationofacrylamideinsnacksbyLCMSSelftest10.5AandSelftest10.5B
ComparisonoftheapproachesSelftest11
https://sisu.ut.ee/measurement 84
Frequentlyaskedquestions
Howmanydecimalplacesshouldweleaveaftercommawhen
presentingresults?
Thenumberofdecimalsafterthecommadependsontheorderofmagnitudeoftheresultand
canbeverydifferent.Itismoreappropriatetoask,howmanysignificantdigitsshouldbeinthe
uncertaintyestimate.Thisisexplainedinthevideoinsection4.5.Thenumberofdecimals
accordingtothatvideoisOKfortheresults,unlesstherearespecificinstructionsgivenhow
manydecimalsafterthepointshouldbepresented.Whenpresentingresulttogetherwithits
uncertaintythenthenumberofdecimalsintheresultandinuncertaintymustbethesame.
Ifweneedtofindstandarddeviationofthosewihtinlab
reproducibilitymeasurments,thenweneedcertainlyusethepooledone?
Wecannottakethesimplieststandarddeviation,whichiscalculatedby
standarddeviationformula?
ThewithinlabreproducibilitystandarddeviationsRWcharacteriseshowwellcanthe
measurementprocedurereproducethesameresultsondifferentdayswiththesamesample.If
thesampleisnotthesame(asinthisselftest)thenifyoujustcalculatethestandarddeviation
oftheresultsthentheobtianedstandarddeviationincludesboththereproducibilityofthe
procedureandalsothedifferencebetweenthesamples.Thedifferencebetweenthesamplesis
inthecaseofthisselftestmuchlargerthanthewithinlabreproducibility.So,ifyousimply
calculatethestandarddeviationoveralltheresultsthenyouwillnotobtainwithinlab
reproducibilitybutratherthevariabilityofanalyteconcentrationsinsamples,whitha(small)
withinlabreproducibilitycomponentadded.
Inestimationofuncertaintyviathemodellingapproach:Whenwhenwe
canusetheKragtenapproachandwhenwejustusethecombination
ofuncertainties?
Inprinciple,youcanalwaysusetheKragtenapproach.However,iftherelativeuncertaintiesof
theinputquantitiesarelarge,andespeciallyifsuchaquantityhappenstobeinthedenominator,
thentheuncertaintyfoundwiththeKragtenapproachcandifferfromthatfoundusingequation
4.11.ThisisbecausetheKragtenapproachisanapproximativeapproach.
Exactlywhatishumanfactor?Ithoughtthatitmaybeforexamplepersons
psychologicalconditionsandpersonalexperienceandsoon?Thiswill
definitelyinfluencemeasurement,butisthistakenintoaccountthen?
The "human factor" is not a strict term. It collectively refers to different sources of uncertainty
that are due to the person performing the analysis. These uncertainty sources can either cause
randomvariationoftheresultsorsystematicshift(bias).Inthetablebelowaresomeexamples,
what uncertainty sources can be caused by the "human factor". In correct measurement
uncertainty estimation the human factor will be automatically taken into account if the
respectiveuncertaintysourcesaretakenintoaccount.
CanwereportasV=(10.0060.016)mLat95%CLatcoveragefactorof
2?
Weuseinthiscoursetheconventionalroundingrulesforuncertainty.Thereforeuncertainty
0.0154mlisroundedto0.015ml.Sometimesitisrecommendedtorounduncertaintiesonly
upwards(leadinginthiscaseto0.016ml).However,inthegradedtestquizespleaseusethe
conventionalroundingrules.
https://sisu.ut.ee/measurement 86
Whatourparticipantssay?
"I have five years experience in accreditation and Q.A. in my lab, and
though participated in many seminars, it is the first time I fully
understood uncertainty. The course was over my expectations and the
Tutorsoverdidthemselvesforourbenefit.IamsopleasedthatIalready
startedspreadingthewordtomycolleagues!"
VasileiosRagias,
Greece
April2015
"Iwouldliketothanktheorganisersforavery
well organised and executed programme for
Measurement Uncertainty in Chemical
Analysis. I will definitely use this in trying to
elaborate the concept to laboratories here in
Trinidad and Tobago. As the Manager of the
Trinidad and Tobago Laboratory Accreditation
Service,thishastrulyenhancedmyknowledge
and application of the concept to real KarleneLewis
situations." TrinidadandTobago
April2015
Valya
Ruseva
Bulgaria
May
2015
"Iamnotsurprisedatallitwaselectedthe"ecourseofthe
year"inEstonia.Thedivisionofthecourseintosections,
videolectures,readingmaterialsandselftestswere
excellent.Thetutorsdidnotanswerquizzesforstudents.
Theyhelpedstudentsresolveproblemsontheirownand
issuesofviolationofrulesguidingstudentscontributionto
learningforumdiscussionswereneveroverlookedbut
immediatelydealtwithtactfully.Thismadebothtutorsand
studentsattachseriousnesstothecourse.Ihavelearntmany
newthingsinthiscourseandtheorganisersdeservelotsof
commendations.Pleaseacceptmyheartfeltgratitudefora EmmanuelOfoeOfosu
coursewellplannedandcongratulations!" Ghana
April2015
https://sisu.ut.ee/measurement 87
"Thiscoursewasaverypleasantsurprise!Honestly,whenIenrolledinthe
coursedidnothavehighexpectationsbecauseIhavecomplicatedschedules,but
sincetheyallowstudentstomanagetheirowntimeinthewaythatsuitsyou
helps.Ithinkmuchofthesuccessofthiscourseisduetothisandit'sareallya
big"Pro"ofthecourse.WiththiscourseIdeepenedmyknowledgeinanarea
whereIhaveexpertisebutwhereIlackedcertainimportantpoints:thecourse
wasaresponsetothesefailures.
Mafalda
Flores, Nodoubtthatallthisinformationwillbeveryusefultomeintheshortandlong
Portugal termandIthankyouforyourworkandcommitmentandthewholeteam.Itis
April2014 truethatknowledgeissharing,andthatfeelingwasquitenotoriousevenamong
studentsenrolledwhowerehelpingeachother.
Inshort,IthinktheUniversityofTartudidanexcellentjobandIwouldbehappy
toattendmorecoursesofthiskind."
"FirstofallIwouldliketothanktheUTteamforthe
pastsixweeks,Ihadablast!
Ididn'tknowanythingaboutmeasurementuncertainty
whenIstartedthiscourse,butnowsuchfigures
suddenlymakesensetome.Interesting,usefulandfun
SusanneJacobsson,
(believeitornot)!"
Sweden
April2014
"Iwasneitheranabsolutebeginnernoranexpertin
measurementuncertainty,butafterthiscourseIfeelIimprovedalot.The
courseintroducedmetoconceptsIwasnotveryfamiliarwith,and
clarifiedsomequestionsIhadinmymindforsometimebuthadnooneto
askto.Thematerialsexplainedtheideasinafewsimplewords,andthe
supportfromteachersthroughoutthecoursewasverygood.Itwasa
pleasureparticipatinginthecourseandIwillbelookingforwardtohearing
LuisCabrita,
ofnewcourses(validation,etc)fromtheUniversityofTartuteam.
Portugal
ThankyouandCongratulations!"
April2014
"Itwasagoodpleasureformetoattendsuchawell
organizedandperfectlypresentedcourse.Ihave
read/listenedalotonthetopicbefore,butthiswasthebest
presentationwithoutanyfrustrationsandmassof
calculations/formulaeetc.Iwouldliketothankto
teammembers,UniversityofTartuandalltheparticipants
KevserTopal,
withtheirvaluablequestionsanddiscussions."
Turkey
April2014
ThefullparticipantfeedbackfromSpring2014isavailablefromthislink:
u_mooc_participants_feedback_2014.pdf 467KB
https://sisu.ut.ee/measurement 88