Quality Assurance and Quality Control in the Analytical Chemical Laboratory
A Practical Approach

.© 2009 by Taylor & Francis Group, LLC

A N A LY T I C A L C H E M I S T R Y S E R I E S
Series Editor

Charles H. Lochmüller
Duke University

Quality and Reliability in Analytical Chemistry, George E. Baiulescu, Raluca-Ioana Stefan, Hassan Y. Aboul-Enein HPLC: Practical and Industrial Applications, Second Edition, Joel K. Swadesh Ionic Liquids in Chemical Analysis, edited by Mihkel Koel Environmental Chemometrics: Principles and Modern Applications, Grady Hanrahan Quality Assurance and Quality Control in the Analytical Chemical Laboratory: A Practical Approach, Piotr Konieczka and Jacek Namie´ snik

.© 2009 by Taylor & Francis Group, LLC

A N A LY T I C A L C H E M I S T R Y S E R I E S

Quality Assurance and Quality Control in the Analytical Chemical Laboratory
A Practical Approach
Piotr Konieczka • Jacek Namie´ snik

Boca Raton London New York

CRC Press is an imprint of the Taylor & Francis Group, an informa business

.© 2009 by Taylor & Francis Group, LLC

but the author and publisher can‑ not assume responsibility for the validity of all materials or the consequences of their use. 2. ‑‑ (Analytical chemistry) Includes bibliographical references and index. MA 01923. or in any information storage or retrieval system. Jacek Namiesnik. and are used only for identification and explanation without intent to infringe. Analytic‑‑Qualitative.S. Government works Printed in the United States of America on acid‑free paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number‑13: 978‑1‑4200‑8270‑8 (Hardcover) This book contains information obtained from authentic and highly regarded sources. Chemistry. including photocopying.CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW.4. Trademark Notice: Product or corporate names may be trademarks or registered trademarks. Series. Chemical laboratories‑‑Quality control.1‑‑dc22 Visit the Taylor & Francis Web site at http://www.com . II. microfilming.com (http://www. transmitted. and recording. please access www.copyright. without written permission from the publishers. (CCC). Quality assurance and quality control in the analytical chemical laboratory : a practical approach / Piotr Konieczka. ISBN 978‑1‑4200‑8270‑8 (hardcover : alk.© 2009 by Taylor & Francis Group.S.Q34K66 2009 542’. 222 Rosewood Drive. now known or hereafter invented. Piotr. or utilized in any form by any electronic.com/) or contact the Copyright Clearance Center. For organizations that have been granted a photocopy license by the CCC. paper) 1. For permission to photocopy or use material electronically from this work. If any copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint.crcpress. Jacek. Danvers. cm. no part of this book may be reprinted. The authors and publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained. QD75. an Informa business No claim to original U. Reasonable efforts have been made to publish reliable data and information. Copyright Law.taylorandfrancis. Title. III. Namiesnik.com and the CRC Press Web site at http://www. Inc. I. LLC 2009000685 . a separate system of payment has been arranged. CCC is a not‑for‑profit organization that pro‑ vides licenses and registration for a variety of users. LLC CRC Press is an imprint of Taylor & Francis Group. or other means. FL 33487‑2742 © 2009 by Taylor & Francis Group. mechanical. Except as permitted under U. Library of Congress Cataloging‑in‑Publication Data Konieczka.copy‑ right. reproduced. 978‑750‑8400. Suite 300 Boca Raton. p.

....................................4 Measures of Dispersion.....8...............8........................... 39 1.......................................................... .........................................................8...............................................................15 Z Score [10.................3 Measures of Location..................1 Introduction.............xi List of Abbreviations...............16 En Score [10.............................. 23 1..................10 Linear Regression..........9............. 4......................................................... 4].......................8..........................8... 29 1.............................................26 1......................................... 21 1.............. 38 References................3 1................5 Snedecor’s F Test [3.10 Cochran-Cox Test [3]..........................8 Statistical Tests............7 1.......1 1............Contents Preface........ 29 1......1 Shewhart Charts........5 Measures of Asymmetry...........................................................12 Cochran’s Test [6]............8........3 Dixon’s Q Test [3.... 19 1.. 11]......1 1.....7 Statistical Hypothesis Testing..........................................6 Hartley’s Fmax Test [3].............6 Measures of Concentration........................ix About the Authors... 27 1............1 Confidence Interval Method [3]...... 13 1............................ LLC ........... 17 1..............14 Hampel’s Test............................................................................................ 5]..........................24 1........... 12 1......... 16 1...................................................8..... ......8 Morgan’s Test [3]....................8............. 14]. 4]................................ 36 1.......1 1........4 Chi Square Test [3]................................ 16 1...................................................17 Mandel’s Test [6............. ............ 13].........7 Bartlett’s Test [3]........................8..............9 Student’s t Test [3............28 1.... 10 1.................8. 11]......................... 27 1.........................................................9 Control Charts............................................................... 15 1...8.....................................................8...................................... xiii Chapter 1 Basic Notions of Statistics........................................ 12..............................1 Characterization of Distributions.................2 v ...................................© 2009 by Taylor & Francis Group..........................................8..18 Kolmogorov-Smirnov Test [2....... 18 1.....................2...................8......................8................... 22 1....................................8...............13 Grubbs’ Test [6...8........ 7]........7 1...................................9........................................... 29 1......8........5 1................8 1......11 Significant Digits: Rules of Rounding......2 Shewhart Chart Preparation...................11 Aspin-Welch Test [3]................2 Critical Range Method [3].............. 30 1.....................1 Distributions of Random Variables............ 10 1.........................................................

..................................................................3........................ 41 2................. 55 References.....3..............3.......... 73 References........................ 55 Chapter 4 Uncertainty......................................................................46 Chapter 3 Traceability...... 85 5...... 95 References..... 67 4................ 49 Role of Traceability in Quality Assurance/Quality Control System...........1 Definitions [1–3]...................5 Certified Value.......... 67 4...........................................................3 Homogeneity....3 4..............................................................................................................................5 Conclusion........2 3........ 77 Parameters that Characterize RMs.........................................................1 4.................vi Contents Chapter 2 Quality of Analytical Results............4 Conclusions.......5 Uncertainty and Confidence Interval.....6 Calibration Uncertainty............................... 59 4................................................. ................................................................................................................................................. 77 Introduction..........4 Practical Application of CRM...................................................2 Introduction..........3 Definitions [1]..............................................................80 5.......3................................................................................2 4.................................................. 69 4............... 74 Chapter 5 Reference Materials......... 41 2.............1 General Information........................................................................................................................................2 5................................................................... 49 Introduction.................4 Tools Used for Uncertainty Estimation........................... 57 Definitions [1–3].............................. 57 Introduction....................................................................................... 51 3...............4 Stability........46 References..... 42 2....2 Representativeness..................................................................................................© 2009 by Taylor & Francis Group.............................. 49 3.........1 Procedure for Estimating the Measurement Uncertainty According to Guide to the Expression of Uncertainty in Measurement........... ......3.. 2]..................7 Conclusion.................3 Quality Assurance System..84 5............................ 58 4.................................. 77 Definitions [1.....................................................1 5................................................. LLC .....................................3 .................................................................................................4 Conclusion.......3.. 82 5.................................... 82 5.............................. 82 5................1 3....................................................... 95 5...................... ....80 5........................... 57 Methods of Estimating Measurement Uncertainty.................................................. 41 2.....................................................

....1 7.............2.............4 Definitions [1..............................................................3......... 160 7...5 Presentation of Interlaboratory Comparison Results: Statistical Analysis in Interlaboratory Comparisons....... 148 7.............................4 Range..............................................................................3..................6 Precision...... 98 Characteristics and Organization of Interlaboratory Comparisons. 143 7.....2 Comparison of Measurement Results Obtained in a Two-Level Study (for Two Samples with Various Analyte Concentrations)....3....................1 Manners of Estimating the Standard Deviation...............4 Graphical Method................ 145 Determinations for Blank Samples...........2 Calculation of LOD Based on the Numerical Value of the S/N Ratio....................3 6.....................................6 Calculation of LOD Based on a Given LOQ....................................................................2.....................................97 Introduction........ 134 7..... .....................................5............................. 164 7.......2..................2 Introduction................ 130 Chapter 7 Method Validation... 120 6... LLC ........3 Limit of Detection and Limit of Quantitation...........2.....2 6...3 Calculation of LOD Based on .....................3..........................7 Testing the Correctness of the Determined LOD....................................... 131 7................1 Selectivity................3...........2......................................................................2...........................5 Sensitivity................................Contents vii Chapter 6 Interlaboratory Comparisons...................................2 Linearity.................. 164 7...........................................6 Conclusions....2.......................... 123 6............... 144 7........................................................ 145 7....... 131 Characterization of Validation Parameters... 129 References.. 166 .........5 Calculating LOD Based on the Standard Deviation of Signals and the Slope of the Calibration Curve.........97 Classification of Interlaboratory Studies..... 146 7......3...... 101 6...... 102 6...1 Comparisons of Results Obtained Using Various Procedures.......3.....2........ ....2.......2..................1 Visual Estimation...5...............................................97 6..................6. 7.........................2.... ..................2. 147 7.......................2....2....... ..© 2009 by Taylor & Francis Group........1 6..... 146 7...................... 136 7..................................................... 134 7............... 2]....

.... LLC .........1 Measurement Errors..........................2................................ 174 7...............................7...................................................2........................................... ......... 195 7.......................................viii Contents Accuracy and Trueness..........................2... 196 7. 231 7.............................© 2009 by Taylor & Francis Group............ 173 7.......................................4 Conclusions.......... ........ ................ 214 Appendix......8 Robustness and Ruggedness........7 .................2.......................................................................... 217 Index..........204 References............9 Uncertainty........................................................................ ..

calculation of the margin of error. and an increase in knowledge about the practical application of statistical tools during analytical data treatment. which makes problem solving easier. it is necessary to use it appropriately. With its comprehensive coverage. For all examples. ix . in some cases. a constructed calculation datasheet (Excel) is attached. each consisting of three main components: problem. it may also prove useful to the scientific community. data. we can contribute to a better understanding of all problems connected with QA/QC. etc. Although this book is primarily designed for students and academic teachers. We hope that with this book. particularly among those who are interested in QA/QC.Preface The aim of this book is to provide practical information about quality assurance/ quality control (QA/QC) systems. LLC . The accompanying CD contains more than 60 Excel datasheet files. It should be noted that in order to obtain correct calculations. Solution data will be calculated and can be read from green marked cells. After saving an Excel file on the hard disk. estimating uncertainty. including definition of all tools. understanding of their uses.© 2009 by Taylor & Francis Group. this book can be of particular interest to researchers in the industry and academia. it is possible to use it on different data sets. as well as government agencies and legislative bodies. additional data such as graphs and conclusions are also included. using statistical tests. The practical part includes more than 60 examples relating to validation parameter measurements. The user’s own data should be copied only into yellow marked cells (be sure that your data set fits the appropriate datasheet). The theoretical part of the book contains information on questions relating to quality control systems. and solution.

His research interests include environmental analytics and monitoring and trace analysis. LLC . 1998). Jacek Namieśnik (MSc 1972-GUT. and more than 350 lectures and communications published in conference proceedings. PhD 1994. He has been the head of the Department of Analytical Chemistry since 1995. and trace analysis. has been employed at Gdańsk University of Technology since 1972. born in 1965. the Jan Hevelius Scientific Award of Gdańsk City (2001). xi . has been employed at Gdańsk University of Technology since 1989 and is currently working as a tutor. he has also served as vice dean of the Chemical Faculty (1990–1996) and dean of the Chemical Faculty (1996–2000 and 2005–present). DSc 1985-GUT. and Fellow of the International Union of Pure and Applied Chemistry (IUPAC) since 1996.© 2009 by Taylor & Francis Group. His research interests include metrology. more than 300 papers. and the Prime Minister of Republic of Poland Award (2007). as well as chairman of the Committee of Analytical Chemistry of the Polish Academy of Sciences since 2007. Among his published scientific papers are 7 books. He was director of the Centre of Excellence in Environmental Analysis and Monitoring in 2003–2005. he has 7 patents to his name. environmental analytics and monitoring. PhD 1978-GUT. He is the receipient of various awards.About the Authors Piotr Konieczka (MSc 1989. and more than 40 papers. Prof. Currently a full professor. including Professor honoris causa from the University of Bucharest (Romania) (2000). as well as more than 70 lectures and communications. 6 book chapters. born in 1949. DSc 2008-GUT). His published scientific output includes 1 book.

List of Abbreviations AAS ANOVA BCR CDF CITAC CL CRM CV CVAAS EN GC GLP GUM IAEA ICH IDL ILC IQR IRMM ISO IUPAC JCGM LAL LOD LOQ L-PS LRM LWL MB M-CS MDL M-PS NIES NIST NRCC PRM PT QA(1) QA(2) QA/QC Atomic Absorption Spectrometry ANalysis Of VAriance Bureau Communautaire de Reference (Standards. LLC . Measurements.© 2009 by Taylor & Francis Group. and Testing Programme–European Community) Cumulative Distribution Function Cooperation on International Traceability in Analytical Chemistry Central Line Certified Reference Material Coefficient of Variation Cold Vapor Atomic Absorption Spectrometry European Norm Gas Chromatography Good Laboratory Practice Guide to the Expression of Uncertainty in Measurement International Atomic Energy Agency International Conference on Harmonization Instrumental Detection Limit InterLaboratory Comparisons InterQuartile Value Institute for Reference Materials and Measurements International Organization for Standardization International Union of Pure and Applied Chemistry Joint Committee for Guides in Metrology Lower Action (control) Limit Limit of Detection Limit of Quantification Laboratory-Performance Study Laboratory Reference Material Lower Warning Limit Method Blank Material-Certification Study Method Detection Limit Method Performance Study National Institute for Environmental Studies National Institute of Standards and Technology National Research Council of Canada Primary Reference Material Proficiency Test Quality Assessment Quality Assurance Quality Assurance/Quality Control xiii .

xiv List of Abbreviations QC QCM RH RM RSD S/N SD SecRM SI SOP SRM UAL USP UWL VIM VIRM Quality Control Quality Control Material Relative Humidity Reference Material Relative Standard Deviation Signal-to-Noise ratio Standard Deviation Secondary Reference Material Le Systeme Internationale d’Unités Standard Operating Procedure Standard Reference Material Upper Action (control) Limit United States Pharmacopea Upper Warning Limit Vocabulaire International des Termes Fondamentaux et Généraux de Métrologie European Virtual Institute for Reference Materials . LLC .© 2009 by Taylor & Francis Group.

• How many determinations should be conducted to increase the precision of a measurement. • Whether the investigated product fulfills the necessary requirements and/or norms. Statistics presents these regularities by means of numbers. 1 . a CDF is (not necessarily strictly) right-continuous.© 2009 by Taylor & Francis Group. LLC . in practice.1  Introduction Mathematical statistics is a branch of mathematics that applies the theory of probability to examining regularities in the occurrence of certain properties of material objects or phenomena that occur in unlimited quantities.2. because it may clear many doubts and answer many questions associated with the nature of an analytic process. • A density function that is the derivative of the CDF: f(x) = F X  ′ (x). Each defined distribution is characterized by the following parameters: • A cumulative distribution function (CDF) X is determined by FX and represents the probability that a random variable X takes on a value less than or equal to x. Statistics is especially helpful for analysts. with its limit equal to 1 for arguments approaching positive infinity. for example: • How exact the result of determination is. 1. Yet. and equal to 0 for arguments approaching negative infinity.2 Distributions of Random Variables 1. A result is a consequence of a measurement. It is a very useful tool that can help us find answers to many questions.1 Basic Notions of Statistics 1. Statistics is not only art for art’s sake. here treated as independent random variables. it is important to remember that statistics should be applied in a reasonable way. The set of obtained determination results creates a distribution (empirical). a CDF is described shortly by: FX (x) = P(X ≤ x).1  Characterization of Distributions The application of a certain analytical method unequivocally determines the distribution of measurement results (properties).

Because this distribution is continuous. defined by two parameters: mean (location) and standard deviation (scale).. These parameters can be divided into four basic groups: • • • • measures of location measures of statistical dispersion measures of asymmetry measures of concentration .2 Quality Assurance and Quality Control Below are the short characterizations of the most frequently used distributions: • normal distribution • uniform distribution (rectangular) • triangular distribution Normal distribution. Uniform distribution is characterized by: • an expected value μx = 0 • a median Me = 0 • a variance SD2 = a2/3 Triangular distribution over the interval 〈−a. physiochemical property). N (μx. The distribution is determined by a pair of parameters – a and +a. also called the Gaussian distribution (particularly in physics and engineering). Statistical parameters are numerical quantities used in the systematic description of a statistical population structure. Unfortunately. +a〉 is characterized by: • an expected value μx = 0 • a median Me = 0 • a variance SD2 = a2/6 The distribution of a random variable provides complete information on an investigated characteristic (e. characteristic inference is drawn using the analysis of a limited number of elements (samples) representing a fragment of the whole set that is described by the distribution. Normal distribution. SD) is characterized by the following properties: • an expected value μx • a median Me = μx • a variance SD2 Uniform distribution (also called continuous or rectangular) is a continuous probability distribution for which the probability density function within the interval 〈−a. one may infer a characteristic using an estimation of some of its parameters (statistical parameters) or its empirical distribution. As a rule. concentration. +a〉 is constant and not equal to zero. it is not important whether the endpoints – a and +a are included in the interval. but outside the interval is equal to zero. LLC . such complete information is seldom available. Then. It is an infinite family of many distributions. content.© 2009 by Taylor & Francis Group. is a very important probability distribution used in many domains.g.

1) Here are the selected properties of the arithmetic mean: • The sum of the values is equal to the product of the arithmetic mean and the population size. • The arithmetic mean fulfills the condition: x min < x m < x max (1.Basic Notions of Statistics 3 1.4) • The arithmetic mean is sensitive to extreme values of the characteristic.3) • The sum of squares of deviations of each value from the mean is minimal: ∑( x − x ) i m i =1 n 2 = min (1. .2) • The sum of deviations of individual values from the mean is equal to zero: ∑ (x − x ) = 0 i m i =1 n (1. LLC . The most popular measures of location are the following: • • • • arithmetic mean truncated mean mode quantiles: • quartiles • median • deciles Arithmetic mean is the sum of all the values of a measurable characteristic divided by the number of units in a finite population: xm = ∑x i =1 n i n (1. estimator) of the expected value.3 Measures of Location Measures of location use one value to characterize the general level of the value of the characteristic in a population [1].© 2009 by Taylor & Francis Group. • The arithmetic mean from a sample is a good approximation (estimation.

and 100-quantiles are percentiles. other means have been proposed. Quantiles are data values marking boundaries between consecutive subsets.© 2009 by Taylor & Francis Group. the truncated mean. there may be more than one value that can be a mode. less sensitive to outliers than the standard mean (only a large number of outliers can significantly influence the truncated mean) and standard deviation. The third quartile (designated Q3) divides the population in a such a way that 75% of the population units have values lower than or equal to the third quartile Q3. A quartile is any of three values that divide a sorted data set into four equal parts. and 75% units have values higher than or equal to the first quartile. is calculated using all results. because the same maximum frequency can be attained at different values. but sometimes may also be regarded as a flaw. Contrary to the arithmetic mean. even immense differences between outliers and the arithmetic mean do not affect its value. among which the extrema (minima or maxima) have a high uncertainty concerning their actual value [2]. 4-quantiles are called quartiles. so that each part represents 1/4 of the sampled population.4 Quality Assurance and Quality Control The truncated mean x wk is a statistical measurement calculated for the series of results. and 90% of the results have values greater than or equal to it. The second quartile Q2 is the median. This mean. Quantiles Q are values in an investigated population (a population presented in the form of a statistical series) that divide the population into a certain number of subsets. The first quartile (designated Q1) divides the population in a such a way that 25% of the population units have values lower than or equal to the first quartile Q1. A median separates the higher half of a population from the lower half. or the mean of the two middle values (for those with an even number of observations). LLC . The median measurement is the middle number in a population arranged in a nondecreasing order (for a population with an odd number of observations). which transfers the extreme to an accepted deviation range — thanks to the application of appropriate iterative procedures. and half of them have values higher than or equal to the median. Hence. Its value is calculated according to the formula: x wk = n − k −1  1  k + 1 x( k +1) + x ( i ) + k + 1 x( n − k )  n  i= k +2   ( ) ∑ ( ) (1. This is usually perceived as its advantage.5) where xwk = truncated mean n = number of results in the series k = number of extreme (discarded) results Mode Mo is the value that occurs most frequently in a data set. In a set of results. The 2-quantile is called the median. for example. The first decile represents 10% of the results that have values lower than or equal to the first decile. . the median is not sensitive to other units in a population. half of the units have values smaller than or equal to the median. and 25% units have values higher than or equal to the quartile. 10-quantiles are deciles.

• If each measurement value is multiplied or divided by any constant value. the greater the dispersion of results. . for example. is the measure of dispersion of individual results around the mean. It must be remembered that dispersion of results occurs in each analytical process. the standard deviation is also multiplied/divided by that same constant. In all other cases it has positive values. It is described by the equation: SD = ∑( x − x ) i m i =1 n 2 n −1 (1. Yet it is not always observed.6) It is a measure characterizing the empirical variability region of the examined characteristic. because of the resolution of a measuring instrument being too low.Basic Notions of Statistics 5 1. The most popular measures of dispersion are: • • • • • range variance standard deviation average deviation coefficient of variation (CV) The range R is a difference between the maximum and minimum value of an examined characteristic: R = x max − x min (1. the greater the value of the standard deviation SD.4 Measures of Dispersion Measures of dispersion (variability) are usually used to determine differences between individual observations and mean value [1]. Variance SD2 is an arithmetic mean of the squared distance of values from the arithmetic mean of the population. but does not give information on the variability of individual values of the characteristic in the population. the square root of the variance. LLC . Thus. Its value is calculated according to the formula: SD 2 = 1 n −1 ∑( x − x ) i m i −1 n 2 (1.© 2009 by Taylor & Francis Group.7) Standard deviation SD. Properties of the standard deviation: • If a constant value is added to or subtracted from each value.8) Standard deviation equals zero only when all results are identical. the standard deviation does not change.

9) Relative standard deviation (RSD) is obtained by dividing the standard deviation by the arithmetic mean: RSD = SD xm (1. and it is always expressed in the same units as the results. the standard deviation is calculated according to the following formula: SD = ∑( x − µ ) i x i =1 n 2 n (1. It determines the mean difference between the results in the population and the arithmetic mean: D= 1 n ∑x −x i i =1 n m (1.14) .13) The mean absolute deviation D is an arithmetic mean of absolute deviations of the values from the arithmetic mean. For series with equal numbers of elements.10) Obviously. If an expected value µx is known. the formula is simplified to the following equation: SDg = 1 k ∑ SD i =1 k 2 i (1. The standard deviation of the arithmetic mean SD is calculated according to the following equation: SD = SD n (1. xm ≠ 0. LLC .12) where k is the number of series of parallel determinations.11) The standard deviation of an analytical method SDg (general) is determined using the results from a series of measurements: SDg = 1 n−k ∑ SD ( n − 1) 2 i i i =1 k (1.© 2009 by Taylor & Francis Group.6 Quality Assurance and Quality Control • Standard deviation is always a denominate number.

The greater the value of the coefficient. give the following values: • • • • • • • • • • mean standard deviation relative standard deviation mean absolute deviation coefficient of variation minimum maximum range median mode . These are absolute numbers. The CV is usually applied in comparing differences: • Between several populations with regard to the same characteristic • Within the same population with regard to a few different characteristics 1. The skewness coefficients are applied in comparisons to estimate the force and the direction of asymmetry. the greater their value. LLC .1 Problem: For the given series of measurement results. The quartile skewness coefficient shows the direction and force of result asymmetry located between the first and third quartiles. the greater asymmetry.© 2009 by Taylor & Francis Group.Basic Notions of Statistics 7 The relationship between the mean and standard deviations for the same set of results can be presented as D < SD.6 Measures of Concentration A concentration coefficient K is a measure of the concentration of individual observations around the mean.15) The CV is the quotient of the absolute variation measure of the investigated characteristic and the mean value of that characteristic. 1. the more slender the frequency curve and the greater the concentration of the values about the mean.5 Measures of Asymmetry A skewness (asymmetry) coefficient is an absolute value expressed as the difference between an arithmetic mean and a mode. The CV is obtained by multiplying RSD by 100%: CV = RSD ⋅ 100% (1. usually presented in percentage points. Example 1. It is an absolute number.

53 12.91 12.7 Statistical Hypothesis Testing A hypothesis is a proposition concerning a population. based on probability.57% 12.34 12. The alternative hypothesis is contrasted with the null hypothesis.52 mg/dm3 12.0257 0.67 12. A hypothesis requires testing. Formulating the null hypothesis and the alternative hypothesis The null hypothesis H0 is a simple form of the hypothesis that is subjected to tests.02 12.xls Quality Assurance and Quality Control 12. law. assumed in order to explain some phenomenon. or fact.52 12.32 mg/dm3 0.© 2009 by Taylor & Francis Group.98 mg/dm3 0.00 mg/dm3 12.98 mg/dm3 12. .98 12.79 12. Statistical hypothesis testing means checking propositions with regard to a population that have been formulated without examining the whole population.48 mg/dm3 0.34 12.34 mg/dm3 1. LLC .8 Data: result series.67 12. mg/dm3: 1 2 3 4 5 6 7 8 9 10 11 12 13 Solution: Mean Standard deviation Relative standard deviation Mean absolute deviation Coefficient of variation Minimum Maximum Range Median Mode Excel file: exampl_stat01.34 12.00 12.264 2. The plot of the testing procedure involves: 1.12 12.

LLC . Statistica). hence. then the null hypothesis should be rejected as false. the null hypothesis H0 is rejected. 4. statistical hypothesis testing is usually carried out using various pieces of software (e. according to the procedure of the selected test and are the basis for the calculation of the test statistic. It means that the value of the calculated parameter is not greater than the critical value of the test (read from a relevant table). and its location is determined by the alternative hypothesis.© 2009 by Taylor & Francis Group. the procedure is limited to calculating the parameter p for a given set of data. the conclusion that the null hypothesis may be true. Nonparametric tests are used to test various hypotheses on the goodness of fit in one population with a given theoretical distribution. 5. the null hypothesis is not rejected. and the randomness of sampling. after selecting an appropriate statistical test. If the calculated p value is smaller than the α value ( p < α). 6.g. is compared with the critical value of the test: −− If the value falls within the critical region. Usually. Determining the critical region of a test The size of the critical region is determined by any low level of significance α. Parametric tests serve to verify parametric hypotheses on the distribution parameters of the examined characteristic in a parent population. −− If the value is outside the critical region. The p value is then compared with the assumed value of the level of significance α. Conclusion The test statistic. the goodness of fit in two populations. . it means that there is not enough evidence to reject the null hypothesis. In this case. It means that the value of the calculated test parameter is greater than the critical value of the test (read from a relevant table). Calculation of a test’s parameter using a sample The results of the sample are processed in an appropriate manner. The basic classification of a statistical test divides tests into parametric and nonparametric ones. 3. Type II error — accepting the null hypothesis H0 when it is false. determined using the sample. Nowadays. The choice of an appropriate test The test serves to verify the hypothesis.Basic Notions of Statistics 9 2.. Determination of the level of significance α Errors made during verification: Type I error — incorrectly rejecting the null hypothesis H0 when it is true. Otherwise. they are used to test propositions concerning arithmetic mean and variance. The tests are constructed with the assumption that the CDF is known for the parent population.

tcrit is the critical parameter of the Student’s t test. Calculate the endpoints of the confidence interval for a single result based on the following formula: g = x m ± tcrit n SD n−2 (1. read for f = n – 2 degrees of freedom (Table A. SD is the standard deviation for an unbiased series.10 Quality Assurance and Quality Control 1. and the values of xm and SD are calculated again. and inference based on these tests are presented below. application. various statistical tests can be used. LLC .8 Statistical Tests During the processing of analytical results.16) Course of action Inference where xm is the mean for an unbiased series.1. it is rejected. Calculate the value of the parameter tcalc according to the following formula: tcalc = xi − x m SD (1. it is compensated for in further calculations.8. together with an uncertain result. and SD the standard deviation for the unbiased series. n is the entire size of a series. otherwise. Appropriate tables with critical values for individual tests are given in the attachments at the end of the book (Appendix).1  Confidence Interval Method [3] Test Aim Requirements Confidence interval method Test whether a given set of results includes a result(s) with a gross error • set size 3–10 • unbiased series — an initially rejected uncertain result • only one result can be rejected from a given set Exclude from a set of results the result that was initially recognized as one with a gross error.17) Requirements Course of action where xi denotes uncertain result. 1. Their description. xm the mean value for the unbiased series. If an uncertain result falls outside the limits of the confidence interval.© 2009 by Taylor & Francis Group. • set size 3–10 • unbiased series — an initially rejected doubtful result • only one result can be rejected from a given set Exclude from a set of results the result that was initially recognized as one with a gross error. . Appendix).

tcrit is the critical parameter of the Student’s t test. If tcalc ≤ tcrit(corr). for α = 0. wα is the critical parameter determined for the number of degrees of freedom f = n – 2 (Table A. the initially rejected result is considered to have a gross error. • set size 3–10 • unbiased series — an initially rejected uncertain result • only one result can be rejected from a given set Calculate the endpoints of the confidence interval for an individual result using the formula: g = x m ± wα ⋅ SD (1. otherwise.Basic Notions of Statistics 11 Compare the value of tcalc with the critical value calculated according to the formula: tcrit (corr ) = tcrit ⋅ n n−2 (1.33. read for f = n – 2 degrees of freedom (Table A. SD is the standard deviation for the unbiased series. then the initially rejected result is included in further calculations and xm and s are calculated again. Appendix).1.© 2009 by Taylor & Francis Group. If the uncertain result falls outside the endpoints of the determined confidence interval.19) Requirements Course of action Inference where xm is the mean for the biased series.18) Inference where n is the entire size of a series. LLC . Requirements Course of action Inference .01. kα = 2.65. and n is total number of a series. together with an uncertain result. Appendix). it is rejected and xm and SD are calculated again. SD is the standard deviation for the unbiased series.2.05. • set size >10 • biased series Calculate the endpoints of the confidence interval for an individual result using the formula: g = x m ± kα ⋅ SD (1. If the uncertain result(s) falls outside the endpoints of the determined confidence interval.20) where xm denotes the mean for the biased series. kα = 1. kα is the confidence coefficient for a given level of significance α. from a normal distribution table: for α = 0. it is rejected and xm and SD are calculated again.

3. otherwise.33. If R > Rcrit. 1.8.65. • known value of the mean range for the series — Rm • known results of k series of parallel determinations.21) Inference where xm is the mean for the unbiased series. k ≥ 30) Inference Requirements . and z is the coefficient from the table for a given level of confidence α and n parallel measurements and f degrees of freedom (Table A. the extremum result is rejected and the procedure is conducted anew.22) where SDg is the standard deviation of the method. SDg is the standard deviation of the method.12 Quality Assurance and Quality Control Requirements Course of action • set size >10 • unbiased series — an initially rejected uncertain result • known value of the method’s standard deviation Calculate the endpoints of the confidence interval for an individual result using the formula: g = x m ± kα ⋅ SDg n n −1 (1. with n determinations in each series (most often. it is rejected. from a normal distribution table: for α = 0. kα = 1.01.05. for α = 0.2  Critical Range Method [3] Test Aim Requirements Course of action Critical range method Test whether a given set of results includes a result(s) with a gross error • set size >10 • known value of the method’s standard deviation — SDg Calculate the value of the range result according to the formula: R = xn – x1 Calculate the value of the critical range according to the formula: Rcrit = z ⋅ SDg (1. kα is the confidence coefficient for a given level of significance α.© 2009 by Taylor & Francis Group. If the uncertain result falls outside the endpoints of the determined confidence interval. n = 2 or 3. it is included in the series. Appendix). LLC . and xm and SD are calculated again. kα = 2.

Appendix). then the result from which it was calculated (xn or x1) should be rejected as a result with a gross error and only then should xm and SD be calculated. If one of the calculated parameters exceeds the critical value Qcrit. .5. the ith series of the measurement results is rejected. In some studies [1].24) Inference where zα is the coefficient from a table for a given level of confidence α and n parallel measurements in a series (Table A. 4] Test Aim Hypotheses Requirements Dixon’s Q test Test whether a given set of results includes a result with a gross error H0: In the set of results there is no result with a gross error. xn. .25) Course of action Inference Compare the obtained values with the critical value Qcrit (Table A. read for the selected level of significance α and the number of degrees of freedom f = n. 1.© 2009 by Taylor & Francis Group. Calculate the value of parameters Q1 and Qn according to the formulas: Q1 = x 2 − x1 R Qn = x n − x n−1 R (1.4. Calculate the value of the range R according to the formula: R = Xn − x1. Appendix).3  Dixon’s Q Test [3. the authors use a certain type of Dixon’s Q test that makes it possible to test a series comprising up to 40 results. • set size 3–10 • test whether a given set of results includes a result with a gross error Order the results in a nondecreasing sequence: x1. LLC . . . If Ri > Rcrit.23) Calculate the value of the critical range according to the formula: Rcrit = zα ⋅ Rm (1. H1: In the set of results there is a result with a gross error.8.Basic Notions of Statistics 13 Course of action Calculate the value of the range for each series according to the formula: Ri = x ni − x1i (1. .

 . Appendix). read for the selected level of significance α and the number of degrees of freedom f = n. . .6. read for the selected level of significance α and the number of degrees of freedom f = n.28) .© 2009 by Taylor & Francis Group. . Calculate the value of parameters Q1 and Qn according to the formulas: Q1 = x3 − x1 x n− 2 − x1 Qn = x n − x n− 2 x n − x3 (1. Appendix). • set size 3–7 • test whether a given set of results includes a result with a gross error Order the results as a nondecreasing sequence: x1.6. H1: In the set of results there is a result with a gross error. . . xn. . Calculate the value of parameters Q1 and Qn according to the formulas: Q1 = x 2 − x1 x n−1 − x1 Qn = x n − x n−1 xn − x2 (1. Calculate the value of parameters Q1 and Qn according to the formulas: Q1 = x 2 − x1 R Qn = x n − x n−1 R (1. . • set size > 12 • test whether a given set of results includes a result with a gross error Order the results as a nondecreasing sequence: x1.14 Quality Assurance and Quality Control Test Aim Hypotheses Requirements Course of action Dixon’s Q test Test whether a given set of results includes a result with a gross error H0: In the set of results there is no result with a gross error. xn. LLC . . xn. . • set size 8–12 • test whether a given set of results includes a result with a gross error Order the results as a nondecreasing sequence: x1. . .27) Requirements Course of action Compare the obtained values with the critical value Qcrit (Table A. Calculate the value of the range R according to the formula: R = Xn − x1.26) Requirements Course of action Compare the obtained values with the critical value Qcrit (Table A.

• normal distribution of results in a series Calculate the standard deviation for the series of results. read for the selected level of significance α and the number of degrees of freedom f = n. If one of the calculated parameters exceeds the critical value Qcrit. then it may be inferred that the compared values of the standard deviation differ in a statistically significant manner — rejection of H0. then the result from which it was calculated (xn or x1) should be rejected as a result with a gross error and only then should xm and SD be calculated. Appendix). If the calculated χ2 value does not exceed the critical value 2 ( χ 2 ≤ χcrit ). LLC . Inference . 2 Compare the calculated value χ2 with the critical value χcrit for the assumed level of significance α and the calculated number of degrees of freedom f = n – 1 (Table A. and n is the number of results in an investigated set.8. 1.6.7. H1: The variance calculated for the series of results is different from the set value in a statistically significant manner. Appendix).4  Chi Square Test [3] Test Aim Hypotheses Chi square (χ2) test Test if the variance for a given series of results is different from the set value H0: The variance calculated for the series of results is not different from the set value in a statistically significant manner.© 2009 by Taylor & Francis Group.29) Requirements Course of action where SD is the standard deviation calculated for the set of results. SDo is the set value of the standard deviation. then it may be inferred that the calculated value of the standard deviation does not differ in a statistically significant manner from the set value — acceptance of H0.Basic Notions of Statistics 15 Inference Compare the obtained values with the critical value Qcrit (Table A. If the calculated χ2 value is greater than the critical value read 2 from the tables ( χ 2 > χcrit ). Calculate the chi square test parameter χ2 according to the formula: χ2 = n ⋅ SD 2 2 SDo (1.

1. LLC . 4.8. • normal distributions of results in a series Calculate the standard deviations for the compared series of results. H1: The variances calculated for the compared series of results differ in a statistically significant manner. Appendix). Calculate the Snedecor’s F test parameter according to the formula: n1 ⋅ SD12 n1 − 1 F= n2 2 ⋅ SD2 n2 − 1 Requirements Course of action (1.30) Inference where SD1. n2 are the number of results for two sets.6 Hartley’s Fmax Test [3] Test Aim Hartley’s Fmax test Compare the standard deviations (variances) for many sets of results . then it may be inferred that the calculated values for the standard deviation do not differ in a statistically significant manner — acceptance of H0. then it may be inferred that the compared values of the standard deviation differ in a statistically significant manner — rejection of H0. and n1.© 2009 by Taylor & Francis Group. SD2 denote the standard deviations for the two sets of results. If the calculated F value does not exceed the critical value (F ≤ Fcrit).8.5 Snedecor’s F Test [3. 5] Test Aim Hypotheses Snedecor’s F test Compare the standard deviations (variances) for two sets of results H0: The variances calculated for the compared series of results do not differ in a statistically significant manner. If the calculated F value is greater than the critical value read from the tables (F > Fcrit).16 Quality Assurance and Quality Control 1.8. Note that the value of the expression should be constructed in such a way so that the numerator is greater than the denominator — the value F should always be greater than 1. Compare the calculated value with the critical value for the assumed level of significance α and the calculated number of freedom degrees f1 and f2 (where f1 = n1 – 1 and f2 = n2 – 1) (Table A.

9. Compare the calculated value with the critical value of the parameter for the assumed level of significance α. • the number of results in each series of the sets is greater than 2 Requirements .Basic Notions of Statistics 17 Hypotheses Requirements Course of action H0: The variances calculated for the compared series of results do not differ in a statistically significant manner.31) Inference where SDmax. H1: The variances calculated for the compared series of results differ in a statistically significant manner. the calculated number of degrees of freedom f = n – 1.© 2009 by Taylor & Francis Group. LLC . H1: The variances calculated for the compared series of results differ in a statistically significant manner.8. • normal distributions of results in a series • numbers of results in each series of the sets greater than 2 • set sizes are identical • the number of series not greater than 11 Calculate the standard deviations for the compared series of results. SDmin are the greatest and smallest value from the calculated standard deviations for the sets of results. and the number of the compared series k (Table A. then it may be inferred that calculated standard deviations do not differ in a statistically significant manner — acceptance of H0. 1. Calculate the value of the Fmax test parameter according to the formula: Fmax = 2 SDmax 2 SDmin (1. then it may be inferred that the compared standard deviations differ in a statistically significant manner — rejection of H0. Appendix). If the calculated Fmax value does not exceed the critical value (Fmax ≤ Fmaxo). If the calculated Fmax value is greater than the critical value read from the tables (Fmax > Fmaxo).7 Bartlett’s Test [3] Test Aim Hypotheses Bartlett’s test Compare the standard deviations (variances) for many sets of results H0: The variances calculated for the compared series of results do not differ in a statistically significant manner.

If the calculated Q value is greater than the critical value read 2 from the tables (Q > χcrit ). then it may be inferred that the calculated standard deviations do not differ in a statistically significant manner — acceptance of H0. Appendix). ni is the number of parallel determinations in a given series.18 Quality Assurance and Quality Control Course of action Calculate the standard deviation for the compared series of results.34) where n is the total number of parallel determinations. k is the number of the compared method (series).7.© 2009 by Taylor & Francis Group. Calculate the value of a Q test parameter according to the formula: Q= 2 2.32) in which: c = 1+ 1 3 k −1 2 1 1 −  ( ) ∑ n −1 n − k   i =1 k i     k  (1.33) SD o = 1 n−k ∑ SD ( n − 1) 2 i i n =1 (1. and SDi is the standard deviation for the series i. H1: The variances calculated for the compared series of results differ in a statistically significant manner. LLC .8  Morgan’s Test [3] Test Aim Hypotheses Morgan’s test Compare standard deviations (variances) for two sets of dependent (correlated) results H0: The variances calculated for the compared series of results do not differ in a statistically significant manner. Inference 1. 2 Compare the calculated value with the critical value of χcrit parameter for the assumed level of significance α and the calculated number of degrees of freedom f = k – 1 (Table A. • number of results in each series of the sets is greater than 2 Requirements . If the calculated Q value does not exceed the critical value 2 (Q ≤ χcrit ).303   n − k log  SD o  −    c   ( ) ∑ ( n − 1) log ( SD ) i 2 i i =1 k   (1. then it may be inferred that the compared standard deviations differ in a statistically significant manner — rejection of H0.8.

If the calculated t value does not exceed the critical value tcrit.1. x2i denote the individual values of results for the compared sets. Compare the calculated t value with the critical value tcrit.36) Calculate the value of parameter t according to the formula: t= (1 − L )( k − 2) L (1. H1: The calculated means for the compared series of results differ in a statistically significant manner. If the calculated t value is greater than the critical value read from the tables (t >  tcrit). then it may be inferred that the calculated standard deviations do not differ in a statistically significant manner — acceptance of H0. Calculate the regression coefficient r according to the formula: k r=  k    ∑x x −∑x ∑x 1i 2 i 1i i =1 i =1 i =1 k k k 2i 2 ∑ i =1 k  2 x1 i −   ∑ i =1 k  x1i    2   k     ∑ i =1 k  2 x2 i −   ∑ i =1 k  x 2i         (1. . then it may be inferred that the compared standard deviations differ in a statistically significant manner — rejection of H0. 1.© 2009 by Taylor & Francis Group.8. Appendix). 4] Test Aim Hypotheses Student’s t test Compare means for two series (sets) of results H0: The calculated means for the compared series of results do not differ in a statistically significant manner.Basic Notions of Statistics 19 Course of action Calculate the standard deviations for the compared series of results. a parameter for the assumed level of significance α and the calculated number of degrees of freedom f = k – 2 (Table A. and x1i.35) Calculate the value of test L parameter according to the formula: L= ( 2 4 SD12 SD2 1− r2 2 SD12 + SD2 ) 2 ( ) 2 − 4r 2 SD12 SD2 (1. so that the relation t ≤  tcrit is satisfied.9 Student’s t Test [3.37) Inference where k is the number of pairs of results. LLC .

20 Quality Assurance and Quality Control Requirements Course of action • normal distributions of results in a series • number of results in each series of the sets greater than 2 • insignificant variance differences for the compared sets of results (Snedecor’s F test. certified value). Student’s t test Compare the mean with the assumed value H0: The calculated mean does not differ in a statistically significant manner from the assumed value. and SD1.8. x2m denote the means calculated for the two compared sets of results. If the t value does not exceed the critical value tcrit (t ≤ tcrit). LLC . then it may be inferred that the obtained means do not differ in a statistically significant manner — acceptance of H0.39) Test Aim Hypotheses Requirements Course of action where xm is the mean calculated for the set of results. μ is the reference (e. SD2 are the standard deviations for the sets of results.g.. If the calculated t value is greater than the critical value read from the tables (t > tcrit).© 2009 by Taylor & Francis Group. Calculate the Student’s t test parameter according to the equation: t= (x − x ) ( n − 1) SD + ( n − 1) SD 1m 2m 1 2 1 2 n1n2 n1 + n2 − 2 n1 + n2 ( 2 2 ) (1. Appendix). Compare the calculated value with the critical value of a parameter for the assumed level of significance α and the calculated number of degrees of freedom f = n1 + n2 – 2 (Table A. SD is the unit of deviation. then it is inferred that the compared means differ in a statistically significant manner — rejection of H0. Calculate the Student’s t test parameter according to the equation: t= xm − µ SD n (1. .1. Section 1.5) Calculate the means and standard deviations for the series of results. • normal distribution of results in a series • the number of results in a series of sets is greater than 2 Calculate the mean and standard deviation for the series of results. H1: The calculated mean differs in a statistically significant manner from the assumed value.38) Inference where x1m .

.41) where x1m. If the calculated t value is greater than the critical value read from the tables (t > tcrit).1. and SD1. If the t value does not exceed the critical value tcrit (t ≤ tcrit).42) . then it may be inferred that the obtained mean is not different from the set value in a statistically significant manner — acceptance of H0. Compare the calculated value with the critical value of a parameter for the assumed level of significance α and the calculated number of degrees of freedom f = n – 1 (Table A.© 2009 by Taylor & Francis Group. x2m denote the means calculated for the two compared sets of results.10  Cochran-Cox C Test [3] Test Aim Cochran-Cox C test Compare the means for the series of sets of results. and z2 = n1 − 1 n2 − 1 Hypotheses Requirements Course of action C= x1m − x 2 m z1 + z2 (1. Appendix). Calculate the critical value of the parameter C (Ccrit) according to the formula: Ccrit = z1t1 + z2t2 z1 + z2 (1. LLC . it is inferred that the mean is different from the set value in a statistically significant manner — rejection of H0. and n is the number of results.g.8.40) (1. SD2 are the standard deviations for the sets of results. for which the standard deviations (variances) differ in a statistically significant manner H0: The calculated means for the compared series of results do not differ in a statistically significant manner.Basic Notions of Statistics 21 Inference e. H1: The calculated means for the compared series of results differ in a statistically significant manner. the standard deviation of the set of results which the mean was calculated based on. • normal distribution of results in a series • the number of results in a series of sets is greater than 2 Calculate the means and standard deviations for the compared series of results. Calculate the value of parameter C according to the formula: in which: z1 = 2 SD12 SD2 . 1.

45) .43) SD12 n1 c= 2 SD12 SD2 + n1 n2 2 SD12 SD2 < n1 n2 (1. • normal distribution of results in a series • the number of results in a series of sets is greater than 6 Calculate the means and standard deviations for the compared series of results. and the level of significance. the number of degrees of freedom.22 Quality Assurance and Quality Control Inference where t1 and t2 denote the critical values read from the tables of the Student’s t distribution (Table A. Compare the calculated C value with the calculated critical value Ccrit.© 2009 by Taylor & Francis Group. LLC .1. Appendix).11 Aspin-Welch Test [3] Test Aim Aspin-Welch test Compare the means for the series of sets of results for which the standard deviations (variances) differ in a statistically significant manner H0: Calculated means for the compared series of results do not differ in a statistically significant manner. If C does not exceed Ccrit (C ≤ Ccrit).44) in which (1. for f1 = n1 – 1 and f2 = n2 – 1. respectively. Calculate the values of expressions described using the following equations: ν= x1m − x 2 m 2 SD12 SD2 + n1 n2 Hypotheses Requirements Course of action (1.8. then it is inferred that the obtained means differ from one another in a statistically significant manner — rejection of H0. then it may be inferred that the obtained mean values do not differ from one another in a statistically significant manner — acceptance of H0. 1. If the calculated C value is greater than the calculated critical value (C  >  Ccrit). H1: Calculated means for the compared series of results differ in a statistically significant manner.

© 2009 by Taylor & Francis Group.46) 2 i where SDmax is the maximum standard deviation in the investigated set (among the investigated laboratories). If the calculated v value is greater than the calculated critical value (ν > νo ). SDi is the standard deviation for a given series (data from a laboratory). LLC . but only when the number of compared laboratories is greater than 2 • sets of results (series) with the same numbers • it is recommended to apply the tests before Grubbs’ test (Section 1. f 2.10. Compare the calculated value ν with the critical value νo for the corresponding level of significance α. SD2 are the standard deviations for the sets of results. If the value of ν does not exceed the critical value νo (ν ≤ νo ). and the calculated values of c. f 2 = n2 – 1. the number of degrees of freedom f1 = n1 – 1. and thus νo (α.13) Calculate the standard deviations for each of the compared sets of results. x2m denote the means calculated for the two compared sets of results.8. and SD1. c) (Table A. then it may be inferred that the obtained means do not differ from one another in a statistically significant manner — acceptance of the hypothesis H0.Basic Notions of Statistics 23 Inference where x1m . 1. and p is the number of standard deviations (the number of compared laboratories). Appendix). Calculate the value of parameter C using the formula: C= 2 SDmax Requirements Course of action ∑ SD i =1 p (1. .8.12  Cochran’s Test [6] Test Aim Cochran’s test Detection of outliers in a given set — intralaboratory variability test One-sided test for outliers — the criterion of the test examines only the greatest standard deviations • the number of results in a series (set) greater than or equal to 2. f1. it is inferred that the obtained means differ from one another in a statistically significant manner — rejection of the hypothesis H0.

47) Course of action Inference where xp is the value in the set of results considered to be an outlier. 2. If the value of the calculated test parameter is less than or equal to the critical value corresponding to the level of significance α = 0. . 1. .01. then the investigated result is an uncertain value. Compare the calculated value Gp with the critical value for a given p value. xm is the mean. Calculate the value of parameter Gp according to the relation: Gp = (X p − Xm ) SD (1.12. the number of results in a series. then the investigated result is considered correct. Appendix).12) • with a single use. Appendix).24 Quality Assurance and Quality Control Inference Compare the calculated C value with the critical value for a given n value. the number of laboratories (Table A. . it enables the detection of one outlier. If the value of the test parameter is greater than the critical value corresponding to the level of significance α = 0.13 Grubbs’ Test [6.05. then the investigated result is considered correct.05. but only when the number of compared laboratories is greater than 2 • the same number of results in the sets (series) of results • it is recommended to apply this test before Cochran’s test (Section 1.01.8. then the investigated result is considered an outlier. Order the set of data xi for i = 1. . LLC .11. 7] Test Aim Requirements Grubbs’ test Detect outliers in a given set — interlaboratory variability test • the number of results in a series (set) is greater than or equal to 2. If the numerical value of a respective test parameter is greater than the critical value corresponding to the level of significance α = 0. p in an increasing sequence. and p the number of laboratories (Table A.© 2009 by Taylor & Francis Group. .8. If the value of the calculated test parameter is less than or equal to the critical value corresponding to the level of significance α = 0.05 and less than or equal to the critical value corresponding to the level of significance α = 0. and SD is the standard deviation. thus it should be repeated until no outliers are observed in the remaining results Calculate the standard deviation for the set of results.

50) SD(22. .01.2) for two of the lowest results.2) according to the equations: SD(2p−1. then the investigated result is considered an outlier.05. p) when testing two of the highest results or xm(1. Calculate the values of parameter SDo according to the equation: SD = 2 o Course of action ∑( x − x ) i m i =1 p 2 (1. after rejection of this value from the set of results. and the course of action should be continued until there are no more outliers in the set of results.1) = ∑( x − x ) m (1.© 2009 by Taylor & Francis Group.8. or x m (1. or 2 (1. then the investigated result is an uncertain value. the test for the series of p – 1 results may be conducted again.2 ) 2 . .49) Calculate the values of respective parameters: SD( p –1. p) = 1 p −1 ∑ i =1 p− 2 xi . but only when the number of compared laboratories is greater than 2 • the same number of results in the sets of results (series) • it is recommended to apply this test before Cochran’s test (Section 1. 2. p in an increasing sequence. according to the equations: x m ( p−1.2) = 1 p −1 ∑x i i =3 p (1. LLC .12) • in a given course of action. Test Aim Requirements Grubbs’ test Detect outliers in a given set — interlaboratory variability test • the number of results in a series (set) is greater than or equal to 2. p) = ∑( x − x i i =1 p i i =3 p− 2 m ( p−1. p ) ) . p) or SD(1. two (the greatest or the smallest) results may be rejected from the set of results Order the set of data xi for i = 1.Basic Notions of Statistics 25 If the numerical value of a corresponding test parameter is greater than the critical value corresponding to the level of significance α = 0. . and less than or equal to the critical value corresponding to the level of significance α = 0.48) Calculate the values of parameters xm( p –1.01. If the value of the test parameter is greater than the critical value corresponding to the level of significance α = 0. .

then the investigated results are uncertain. If the following condition is satisfied ri ≥ 4.© 2009 by Taylor & Francis Group. Calculate the deviations of ri from the median for each result using the formula: ri = xi − Me ( ) (1. and the course of action should be continued until there are no more outliers in the set of results. Calculate the median deviations Me ri .12. then the investigated results are considered correct. Calculate the median Me for all the results xi.26 Quality Assurance and Quality Control Calculate the value of parameter G according to the equations: G= SD(2p−1. If the value of the calculated test parameter is less than or equal to the critical value corresponding to the level of significance α = 0. after rejection of these values from the set of results.14 Hampel’s Test The Hampel test by some authors is called Huber’s test [8. LLC . 1. Order the values of |ri| in an increasing sequence.01. or G = SD(2 1. Compare the values of |ri| with 4. Test Aim Requirements Course of action Hampel’s test Detect outliers in a given set • the number of results in a series (set) is greater than 2 Order the values in an increasing sequence.8. .53) then the result xi is considered an outlier. the number of laboratories (Table A.05.05 and less than or equal to the critical value corresponding to the level of significance α = 0. the test for the series of p – 2 results may conducted again. where xi includes the interval from x1 to xn.5 ⋅ Me ri . If the value of a test parameter is greater than the critical value corresponding to the level of significance α = 0.01. p) 2 SDo .52) Inference Calculate the absolute values |ri|. 9]. If the numerical value of a corresponding test parameter is greater than the critical value corresponding to the level of significance α = 0.2 ) 2 SDo (1.5 ⋅ Me ri (1. then the investigated results are considered outliers. Appendix).51) Inference Compare the calculated value of G with the critical value for a given p value.

57) where x lab is the value obtained by a given laboratory.8. 11] Test Aim Requirements Course of action En score Estimation of results of interlaboratory comparisons • the number of results in a series (set) is greater than 2 Calculate En score using the formula: En = xlab − xref u(2xlab ) + u(2xref ) (1. the estimation is unsatisfactory. The modified standard deviation calculated using the relation: SDmod = SD 2 + u(2xref ) (1. Inference If En ≤ 1.16  En Score [10. a result is considered satisfactory.54) Requirements Course of action where xlab is result obtained by a given laboratory. If Z ≥ 3. xref is the assumed value/the reference value. xref is the reference value.15  Z Score [10. If 2 < Z < 3 .55) where u( xref ) is the standard uncertainty of the accepted value/ reference value.8. If En > 1. and u( xref ) the combined standard uncertainty of the reference values. u( xlab ) is the combined standard uncertainty result obtained by a given laboratory. Combined standard uncertainty calculated using the relation: u = u(2xlab ) + u(2xref ) (1. a result is considered unsatisfactory. . and SD is the deviation unit: The standard deviation calculated using all the values in a set. Inference If Z ≤ 2.Basic Notions of Statistics 27 1. 11] Test Aim Z score Detect uncertain results and outliers Applied during the processing of results of interlaboratory comparisons • the number of results in a series (set) is greater than 2 Calculate the Z score using the formula: Z= x lab − x ref SD (1. the estimation is satisfactory. LLC . a result is considered uncertain.© 2009 by Taylor & Francis Group.56) where u( xlab ) is the standard uncertainty of a value obtained by a given laboratory. 1.

mark the horizontal lines to test the data’s configuration. Mandel’s k Determine the interlaboratory traceability of results • the number of results in a series (set) is greater than 2 Calculate the standard deviations SDi for each series of results for each laboratory. corresponding to the Mandel h coefficients for a given level of significance (α = 0. according to the formula: hi = x mi − x m 1 ( p − 1) ∑ (x i =1 p − x m )2 (1. Calculate the mean for results from a given series according to the formula: xm = ∑n ⋅ x i i =1 p mi ∑n i =1 p (1.01 or 0.8. LLC .28 Quality Assurance and Quality Control 1. Appendix). 13] Test Aim Requirements Course of action Mandel’s h Determine the interlaboratory traceability of results • the number of results in a series (set) is greater than 2 Calculate the means xmi for each series of results for each laboratory.© 2009 by Taylor & Francis Group. Calculate the values of parameter hi for a given series and for a given laboratory.13a or Table A. and p is the number of laboratories. On the graph of the values of parameter h.17  Mandel’s Test [6.59) mi Inference Make a graph of the values of parameter hi for each series in the sequence of laboratories. 12. according to the formula: ki = SDi p Test Aim Requirements Course of action ∑ SDi2 (1. The value of parameter hi greater than h needs to be checked from an analytical viewpoint.13b.58) i where ni is the number of results for a given series obtained by a given laboratory.60) .05) (Table A. Calculate the values of parameter ki for a given series and for a given laboratory.

13b.© 2009 by Taylor & Francis Group. The value of parameter ki bigger than k value needs to be checked from an analytical viewpoint.13a or Table A. On the graph of the values of parameter k. and Dn1n2 is the greatest value of differences between empirical distribution functions. according to the formula: λn = n1n2 Dn n n1 + n2 1 2 (1. Appendix). then it may be inferred that there are statistically significant differences in distribution functions for both compared series. then it may be inferred that there are no statistically significant differences in distribution functions for both compared series.9 Control Charts 1. If the λn value does exceed the critical value λα (λn > λα ). 14] Test Aim Requirements Course of action Kolmogorov-Smirnov test Compare the distribution of two series of results • two series of results Calculate the empirical distribution functions for each series of results.61) Inference where n1.14.9. If the λn value does not exceed the critical value λα (λn ≤ λα ). the most frequently used charts are Shewhart charts.05) (Table A. Appendix). It enables fast and simple .8. n2 denote the number of results for a given series. mark the horizontal lines to test the data’s configuration. 1. This method of monitoring and regulating processes is a graphic procedure minimizing the number of necessary numerical operations and allowing systematic monitoring of the course of the process being subjected to control. In practice. LLC .1 Shewhart Charts Control charts are used to test the stability of research results conducted in a given laboratory.01 or 0. Calculate the values of parameter λn.18 Kolmogorov-Smirnov Test [2. Compare the λn value with critical value — λα for a given level of significance (Table A. 1. corresponding to the Mandel k coefficients for a given level of significance (α = 0.Basic Notions of Statistics 29 Inference Make a graph of the values of parameter ki for each series in the sequence of laboratories.

• If the hypothesis is not rejected..30 Quality Assurance and Quality Control detection of abnormalities in the configuration of the marked points. start preparation of the first chart: Mark the consecutive numbers of result determinations on the x-axis of the graph.g. it is recommended that action be taken on the chart. where SD is the standard deviation of the investigated characteristics. or in other words the upper and lower warning limits. respectively). however. charts are prepared separately for each procedure. usually a graph with control limits depicted. The course of action in preparing the control chart is as follows: • Conduct 10–20 measurements for a standard sample. therefore. and two statistically determined control limits. Limits of ±2SD are also marked. that is. both values should be determined for the unbiased series. one line on either side of the central line.© 2009 by Taylor & Francis Group. • Test the hypothesis about a statistically insignificant difference between the obtained mean and the expected value using Student’s t test (Section 1.9. when a point appears outside the control limits ±3SD. the occurrence of any value from a sample falling outside these limits is simply warning about a possible transgression of the control limits. hence. • Calculate the mean xm and the standard deviation SD. after the initial rejection of outliers. Both the upper and lower limits on the chart are found within ±3SD from the central line. every batch).g.8. provided that the process is statistically ordered. expressed either in time (e. the upper and lower control limits (UAL and LAL. The measurement is obtained from a series of measurement results obtained at approximately regular intervals. The possibility of transgressing the control limits as a result of random incident is insignificantly small. 1. and thus fast correction and confirmation of the reliability of the research.2 Shewhart Chart Preparation Preparation of a chart depicting mean and standard deviation (xm – SD) will be described as an example. but the results obtained for control samples should be marked parallel to the received results for the investigated samples: . the values of a certain statistics measurement are registered. Limits of ±3SD (so-called action limits) show that approximately 99.9). The main role is played by an appropriate control chart. Mark the obtained measurement results for 20 consecutive samples.7% of the values fall in the area bounded by the control lines. the limits of ±2SD are called warning limits (UWL and LWL).. Mark a central line CL on the graph corresponding to the reference values of the presented characteristic. LLC . and the values of the observed characteristics (the mean) on the y-axis. every hour) or quantity (e. On such a graph.

Basic Notions of Statistics 31 −− If a determination result is located within the warning limits.© 2009 by Taylor & Francis Group. Otherwise. and the process may then be halted or corrected. but within the action limits. a new mean is calculated for the next chart as the arithmetic mean for the compared chart. Once the cause has been located and eliminated. If the standard deviations do not differ in a statistically significant manner. one should detect the cause. and a new chart is prepared for the newly calculated values of xm and SD. If the mean values differed only in a statistically significant manner. If a value marked on the chart falls outside any of the control limits or the series of values reflects unusual configurations. −− Two subsequent measurement points being outside warning limits. When the difference between these values is statistically significant. For each new chart. calibration should be carried out again. testing whether the process is not changing and remains statistically regulated. or seven consecutive results create a trend (decreasing or increasing). LLC . it is necessary to compare the mean obtained for test samples with the expected value. albeit the comparison should always involve two last charts. If the process is statistically regulated. the process is not statistically regulated. however.5). −− If a result for a test sample is found outside the action limits. but in the interval determined by the action limits. −− Ten consecutive measurement points being found on the same side of the mean value. −− There exist three other signs indicating the occurrence of a problem in the analyzed arrangement. a new chart should be prepared for the values of the penultimate chart. it is considered satisfactory. If the standard deviations differ in a statistically significant manner. −− The occurrence of results between the warning limits and action limits is also acceptable. a new chart should be prepared for parameters identical to those in the compared chart.8. In this situation. one should compare the standard deviations obtained for the investigated chart and those obtained for a previous chart using the Snedecor F test (Section 1. on the same side of the mean value. namely: −− Three consecutive measurement points occurring outside the warning limits. the standard deviation is calculated for the next chart as the arithmetic mean of the s values for the compared charts (the mean values are compared using the Student’s t test). If the means do not differ in a statistically significant manner. not more often than two results per 20 determinations. then a control chart is the method used for continuously testing the statistical null hypothesis. . the process may be resumed and continued. the results from this series (chart) should be rejected.

23 4.40 4.32 4.1005 4.14 4.22 .20 4.12 4.32 Quality Assurance and Quality Control Example 1.2 Problem: Draw a Shewhart chart for the 20 given measurement results obtained for the test samples.23 4.34 4. Mark the central line.17 4.36 4.20 0.10 4.22 4.07 4.90 Data 4.50 4.32 4. LLC .00 3.30 4.© 2009 by Taylor & Francis Group. Data: result series: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Solution: Mean SD UAL UWL LWL LAL 4.03 4.21 4.27 4. and the warning and action lines.04 4.11 4.

15 4.34 4.60 4.33 4.00 4. Data: result series: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Data 4.08 4.© 2009 by Taylor & Francis Group.11 4.32 4.41 4.50 4.Basic Notions of Statistics Graph: 4.00 3.35 4.10 4.20 4.40 4.32 4.23 4.xls Example 1.17 4.90 3.80 0 5 10 15 LWL LAL 20 CL UAL UWL 33 Excel file: exampl_stat02. LLC .3 Problem: Mark the following data from the previous example on the chart.11 4.12 4.01 4.11 Conclusion ! ok ok ok ok ok ok ! ok ok ok ok ok ok ok ok ok ok ok ok .18 4.44 4.20 4.12 4.30 4.

128 4.20 4.21 0.10 4.80 0 5 10 Quality Assurance and Quality Control 4.xls .34 Solution: Mean1 SD1 Mean2 SD2 UAL UWL LWL LAL Graph: 4.60 4.90 UAL UWL CL LWL LAL 15 20 Excel file: exampl_stat03.00 3.50 4.00 3.101 4. LLC .30 4.90 3.50 4.© 2009 by Taylor & Francis Group.40 4.20 0.40 4.

98 3. and then (with variances not differing in a statistically significant way) the mean were compared using the Student’s t test.Basic Notions of Statistics 35 Example 1.013 For the new chart.200 0.50 4. The variances were compared using the Snedecor’s F test. LLC .106 4.40 3.20 4. the values have been calculated as the means of the two previous charts.10 4.00 3.85 0. 36) t < tcrit 20 0. Solution: Values 1 and 8 have been removed from the set of data.05.80 0 5 10 15 LWL LAL 20 CL UAL UWL 4.90 3. The remaining values were used to calculate the means and the standard deviation.88 Excel file: exampl_stat04.19 0. 19) F < Fcrit t tcrit(0.05. Mean SD UAL UWL LWL LAL Graph: 4.011 1.60 4. Series 1 Series 2 No results. SD Mean n/(n – 1)SD2 F Fcrit(0.51 4.111 4.xls .© 2009 by Taylor & Francis Group.101 4. 17. n Standard deviation.184 0.031 18 0.4 Problem: Draw a new chart based on the data from the previous example.22 1.40 4.30 4.456 2.

such as: • accuracy — through the determination of systematic errors • linearity • limits of detection Therefore. in which the values of the output signal are assigned to corresponding values of analyte concentration. LLC . It is also applied in determining some of the validation parameters of the analytical procedure. we present below the course of action for the linear regression method. b: b= ∑ ∑ xi n i =1 i =1 n n yi − n 2 ∑x y i =1 n 2 i i =1 n i i     ∑ i =1 n  xi  − n   (1. a: a= ∑ y − b∑ x i i =1 i =1 n i n (1.© 2009 by Taylor & Francis Group.64) . a linear regression method is applied.10 Linear Regression Linear correlation is the most frequently used correlation in analytical chemistry.62) where y = dependent variable (output signal of the measuring instrument) x = independent variable (concentration of the determined analyte) a = intercept b = slope The following regression parameters are calculated [3]: • Slope. A decisive majority of analytical measurements uses the calibration stage. To determine the functional dependency that connects the output signal with analyte concentration.63) ∑x • Intercept value.36 Quality Assurance and Quality Control 1. The equation of the linear regression is: y = b⋅x + a (1. together with a presentation of the determination method for the calibration chart parameters.

66) ∑ i =1 n  xi    • Intercept. LLC .© 2009 by Taylor & Francis Group.68) where: n = number of independent determination results for the standard solution samples from which the calibration curve has been determined yi = the value determined experimentally Yi = the value calculated from the determined regression equation . r: n r=  n    n ∑ i =1 n i =1 n xi yi −  2 ∑ ∑y xi i =1 i =1 n 2 i i =1 n n i 2 ∑ x −  ∑ x  2 i i i =1     ⋅ n       ∑ y −  ∑ y  i i =1  n       (1. SDb : SDb = SDxy 2 ∑ i =1 n 1 xi2 −  n  (1.67) • Residuals.65) • Values of standard deviations for: • Slope.Basic Notions of Statistics 37 • Regression coefficient. SDxy : SDxy = ∑( y − Y ) i i i =1 n 2 n−2 (1. SDa: SDa = SDxy n ∑x i =1 n 2 i n ∑ i =1 n  xi2 −    ∑ i =1  xi    2 (1.

© 2009 by Taylor & Francis Group. the number should be “read” from left to right until reaching the first digit that is not zero.2 Then it should be presented with three significant digits: 73.0010823 20.123 349.63 × 104 34. LLC . the value of a result should be presented with the same number of digits after the decimal point as the value with the fewest number of digits after the decimal point.38 Quality Assurance and Quality Control 1. the number of significant digits in a result should be the same as in the value with the fewest significant digits.8 For multiplication and division. That digit and all the subsequent digits are called significant. strictly dependent on the notation of the values applied in the calculation(s). Significant digits in the decimal notation of a given number are all the digits without initial zeros. A value obtained from a calculation(s) should be recorded in an appropriate way.123 349.2113 0. After addition or subtraction. If a result is a product of the following numbers: 11.23 15.1200 507.80 0. the significant digits are underlined: 230.70 Calculations very often use values with different numbers of significant digits and different number of digits after the decimal point.2 then it should be presented with one digit after the decimal point: 375.546 0.4 × 102 .11 Significant Digits: Rules of Rounding A problem with correct notation of the measurement results is usually associated with issues related to significant digits and the rules of rounding.2113 0. if a result is the sum of numbers: 11. For example.23 15. To determine how many significant digits there are in a number. In the example below.

” Accred.” Warszawa. Anal. M. . P. J. Assur. V. Bożyk... Delanote. L.” Warszawa. “Spectrometric determination of silicon in food and biological samples: an interlaboratory trial. The notation of the determination requires presentation of the uncertainty value with two significant digits and a result with the same precision (same number of figures after the decimal point) as the uncertainty value. ISO/IEC Guide 43-1 Proficiency testing by interlaboratory comparison — Part 1: Development and operation of proficiency testing schemes. and Roskams. Scheldeman. Chem. 15. “Statystyka dla chemików analityków.. L.: Analiza ilościowa związków organicznych. “The influence of different evaluation techniques on the results of interlaboratory comparisons. Czermiński. Lugowski.J. 2001 (in Polish).: Zygmunt B. F.P......E.. “Statystyczne kryteria oceny wyników i metod analitycznych w: Bobrański B. W. At. 11.” Warszawa. Claes.V.. Robberecht. Koronacki.. 12. Prof.” Warszawa. H. Ferreira. 2. M. and Kafarow.. “Metody statystyczne dla chemików. Van Cauwenbergh.. Centeno. WNT. Davies. Willemyns. “Quality assurance and quality control in forest soil analyses: a comparison between European soil laboratories. R. K. Verheyen. D’Haese. W. P. 9. WNT. P.Basic Notions of Statistics 39 It must be remembered that the number of significant digits given in the value of a result is strictly dependent on the calculated uncertainty value (see Chapter 5). J. 8.” Accred. 1988. 14. “Optymalizacja eksperymentu w chemii i technologii chemicznej. 1989 (in Polish). 1977 (in Polish). L. Spectrom. 2004 (in Polish).© 2009 by Taylor & Francis Group... “Metody statystyczne w badaniu jakości produktów żywnościowych i chemicznych. G. J. M. M... Benijts. De Vos. 1998..L.” Warszawa. Soares.W. Accuracy (trueness and precision) of measurement methods and results — Part 2: Basic method for the determination of repeatability and reproducibility of a standard measurement method. Kandel.. Quataert. J. 5.” Warszawa. H. PWN. H.J.. and Uytterhoeven M. PWN. 2000. P. 1982 (in Polish). Arnaud. Qual.. X.. 2001..A. Aut. and Mielniczuk. Dobecki... LLC .A. Doerffel. J.. Qual. Van Grieken. K. WNT. 513–519. W.B. Paszek. 6.. S. T.. N. Knapp. Iwasiewicz. Z. “Statystyka dla studentów kierunków technicznych i przyrodniczych.” Fresenius Z. and Grasserbauer. Hoenig. 322–327. M. Lamberts. “Statistical evaluation of interlaboratory tests.... “Zapewnienie jakości analiz chemicznych. 735–743. 688–694. and Sikorski.. 3... E. 1986 (in Polish). ISO 5725-2:1994. i tłum.. Krska. This requirement frequently makes it necessary to round the obtained values down to the appropriate number of digits. Cools.” J. R. Kozłowski... 2004. Van Dyck. 3..C. R.. Nofera.. M. 4..Ł. Clement. Linsinger. 9.. and Rudzki. L... Łódź.. Bastos. Moens. References 1... WNT. Z. 7. A.. 331.. Use of LRM in quality control: interlaboratory testing — EC Growth Projects TRAP-LRM/TRAP-NAS. Achnazarowa.. J. L. M. B. Riondato.. Taylor..L. Assur. 1979 (in Polish). A.” Instytut Medycyny Pracy im. 13. 10. J. Cortez. Deelstra. Anal. S. R.

Measurement results must be reliable.© 2009 by Taylor & Francis Group. Quality control — a complex system of actions to obtain measurement (determination results) with the required quality level. that — in other words — are characterized by the greatest information capacity possible. This information is not usually obtained through an analysis of the whole object. but is based on the analyses of appropriate samples. more and more intense research is taking place with the aim of developing new methodologies and devices so that analytical results are a source of as much information as possible. Analytical data on researched material objects are a specific type of information. quality of the instruments.” Access to a variety of information sources facilitates decision making not only in politics. and quality of the work and organization. LLC . This leads to the conclusion 41 . To satisfy the growing demand for analytical data. where information is bought and sold. Therefore. The quality of information can be divided into components: quality of results. samples have to be collected in such a way that the most important criterion — that is. but also in the economy and technology (related to control over the processes of manufacturing consumer goods). A new type of market arose. quality of the process. Analytical quality — consistency of the obtained results (chemical analysis) with the accepted assumptions. A program of quality control includes: • • • • Assuring a suitable level of staff qualifications Assuring the proper calibration of instruments and laboratory equipment Good laboratory practice (GLP) Standard procedures 2.2  Introduction The past decade or so was undoubtedly a period of “information hunger.1 Definitions [1–3] Quality — the realization of specific requirements (which include the standards established by the quality control system in addition to accepted in-house requirements). representativeness — is met.of Analytical 2 Quality Results 2. which means that they must accurately (both truly and precisely) reflect the real content (amount) of analytes in a sample that is representative of the material object under research.

For the obtained result to be comparable (authoritative.42 Quality Assurance and Quality Control that all developments in analytical chemistry are derived from the desire to obtain in-depth analytical data [4]. In analytics.3  Quality Assurance System One of the basic trends in the recent development of analytical chemistry is the determination of lower and lower concentrations of analytes in samples with a complex matrix. that determines and confirms its reliability. The notion of reliability is closely associated with the notion of quality. LLC . In addition. Both manufactured products and analytical results must have an appropriate quality. the quality of analytical measurements appears to have its own accumulative requirement: the quality of every product is a result of comparison of the obtained value with the reference value. Results of analytical measurements are a type of a product of the chemical analyst’s work. reliable) to the reference value. the notions of quality have a specific meaning. together with its control and assurance. and assuring the quality of analytical results. its (high) quality must be documented and maintained. The need for a uniform and defined control system. is a consequence of the following trends in analytics: • Decrease in the concentrations of analytes • Increase in the complexity of the matrix composition of the sample • Introduction of new notions associated with the application of metrology principles in analytics • Necessity of traceability documentation and estimating uncertainty as requisite parameters of an analytical result • Globalization and the associated necessity of comparing results in different laboratories This task poses a great challenge for analysts and draws attention to quality control and quality assurance (QC/QA) of the obtained results. the expected or the standard one.© 2009 by Taylor & Francis Group. estimation. 2. The quality of results of analytical measurements must be assured in the first place to draw conclusions about the quality of the examined products. The system of quality estimation usually includes the following elements: • Tracking and estimating the precision of obtained results by periodic analysis of test samples • Estimation of accuracy by: • Analyses of certified reference samples • Comparison of obtained results with results obtained for the same sample using the reference method • Sample analyses after the addition of a standard • Comparative interlaboratory (intercomparison) exercises • Control charts • Suitable audit system . It is the quality of a result.

The problem relating to quality assurance and control of measurement results is primarily associated with the insufficient amount of information concerning instruments used in the process and their application.1. LLC . . These are first of all statistical instruments based on metrology.1  Position and role of elements of a quality control/quality assurance (QC/QA) system for obtaining a reliable analytical result.Quality of Analytical Results 43 At present. is in principle voluntary. In Figure 2. it is indispensable to use both the certified reference materials and the analytical procedures subject to prior validation. RELIABLE ANALYTICAL RESULT UNCERTAINTY TRACEABILITY REQUIREMENTS METHOD VALIDATION REFERENCE MATERIALS INTERLABORATORY COMPARISONS TOOLS FIGURE 2. there are three systems of quality assurance in analytical laboratories [5]: • GLP • Accreditation of a laboratory according to EN 45001 or ISO Guide 17025 • Certification according to norms ISO of series 9000 The selection of the quality system. Quality assurance of analytical measurement results is a system comprising five interdependent elements [7]: • • • • • Assurance of measuring traceability of the obtained results Evaluation of uncertainty in obtained results of measurement Use of certified reference materials Participation in various interlaboratory comparisons Validation of the applied analytical procedures Only when the aforementioned tools are used is it possible to provide the authoritative (reliable) results of analytical measurements.© 2009 by Taylor & Francis Group. The elements of the quality system are interdependent. a schematic presentation of the elements of a quality control/quality assurance system used for obtaining reliable analytical results is shown [7]. To assure measuring traceability. although increasing attention is paid to the procedures of accreditation [6]. introduced by a given laboratory.

it is obvious that determination of uncertainty increases the reliability of the obtained results.44 Quality Assurance and Quality Control During validation of an analytical procedure. tolerance (elasticity) • Estimate uncertainty — which enables the control of the entire analytical procedure Interlaboratory comparisons involve both reference materials and analytical procedures.© 2009 by Taylor & Francis Group. In the production of reference materials. compels the precise and very attentive “tracking” of the entire analytical procedure.” it is requisite to determine the influence of all possible parameters of an analytical procedure on the value of the combined uncertainty. Interrelations among the components of a QA/QC system are presented in Figure 2. TRACEABILITY UNCERTAINTY METHOD VALIDATION QA/QC SYSTEM REFERENCE MATERIALS INTERLABORATORY COMPARISONS FIGURE 2.2  Components of the QC/QA system for an analytical process. it is necessary to: • Use certified reference materials — determine the accuracy • Participate in interlaboratory research — determine the traceability. showing interrelationships between components.2. Reference material is also characterized by the uncertainty value. Although uncertainty is not one of the validation parameters. is indispensable in the production of reference materials. On the other hand. those enabling the control of the procedure. . This. this type of research serves to determine certified values for the manufactured reference materials. as noted earlier. LLC . in turn. It is because during the design of the so-called “uncertainty budget. Estimation of measurement uncertainty. analytical procedures are applied during the determination of homogeneity and stability of materials.

© 2009 by Taylor & Francis Group. Accordingly. a special case of such mixtures are “zero” mixtures used for: • Testing the zero position on the measuring scale of the instrument • Diluteness of standard mixtures. A schematic representation of this concept is shown in Figure 2. along with the determination and the control of the elements of the quality control. These two parameters are the basic requirements for a reliable measurement result. in order to be applicable. containing strictly defined concentrations of analytes • Testing the reliability of the whole plot of the analytical conduct UNCERTAINTY X ± U SI TRACEABILITY FIGURE 2. This can be achieved by: • Defining the basic notions of the quality system • Determining the simple and intelligible procedures used when using individual elements of the quality system • Providing clear and transparent dependencies (in which elements of metrology and mathematical statistics are used) enabling the “numerical” or “parametric” determination of each characteristic. must be defined in a way that is intelligible for the user. LLC .3.Quality of Analytical Results 45 Each element of the quality control system concerning the results of analytical measurements must be applied by any laboratory that wishes to obtain reliable results. The necessity of presenting the result together with these two basic parameters must be remembered by every “producer” of analytical results. . and the determination of the quality of the control system elements • Helping users to derive inferences on the quality system. Each of these elements. based on determined values for each of its elements Every analyst should be aware that the basic and requisite parameters characterizing an analytical result are traceability and uncertainty. A requisite condition of assuring the appropriate quality of analytical results is the verification of the reliability of the used gauges and checking of the range of application and calibration of the analytical procedures. It must also be clearly and intelligibly presented.3  Necessary parameters for a reliable analytical result. analytical procedures usually involve two operations associated with calibration: • Periodic reliability test of indications of the instruments used by means of standard mixtures.

Hence. and also their transport. 2007 (in Polish). if stages of sample preparation (extraction. the quality management system does not take into account these weak links of the analytical process.4 Conclusions For a laboratory to be able to deliver reliable and repeatable results. actions. P. WNT. albeit an important one. precision. LLC . chromatographic extraction of mixtures or spectrometric detection).46 Quality Assurance and Quality Control Realization of this operation can be achieved in two ways: • By addition of a standard to the analyzed sample • As a result of applying reference material samples Chemical analysis of any material can be described as a chain of decisions. etc. it is necessary to perform systematical calibration of analytical instruments and subject all analytical procedures to validation. . reference material samples are subject to the same processing and determinations as real samples.).© 2009 by Taylor & Francis Group.. and procedures [8]. As in the case of any chain.g. Moreover. but rather the stages that take place outside the analytical laboratory. accuracy. maintenance. covering the previous notion of “method applicability range” (selectivity. can cause serious misinformation. This process must be an integral process. This notion means the determination of the methodology characteristics. purifying extracts) have not been carried out properly. limit of detection. where the applicability test for the analytical method (validation) is only one stage. J. Kontrola i zapewnienie jakości wyników pomiarów analitycznych. range. and Namieśnik. such as: • Selection of materials to be sampled • Preparation of the sampling plan • Selection and use of techniques and devices necessary in sampling. For the purpose of quality control in a laboratory work. In general. Warsaw. and storage If a given analytical laboratory is not responsible for the sampling stage. 2.. Konieczka. in a chemical analysis the power of the entire chain also depends on the power of its weakest link. instead of being a source of information. control and assurance of quality of the analytical results should involve all stages of the analytical process. Comparison of the obtained result with the real analyte concentration in the reference material sample may give conclusions concerning the reliability of analytical works conducted in a given laboratory [9]. (eds. then even the most modern analytical instruments and complex computer techniques cannot improve the situation. repeatability. References 1. the weakest links in the analytical process are not the elements acknowledged as components of chemical analysis (e. Such analytical results have no value and. linearity.).

9. 266–268. Assur. Instytut Medycyny Pracy im. Weinheim. “How to achieve international comparability for chemical measurements. A.B. Valcárcel. Richter. Otto.. PWN. Wiley-VCH. 7.” Accred. 1994. ... 13. 173–190. Assur. “GLP and other quality assurance systems — a comparison. and Ponomareva. Chem. “The role of and place of method validation in the quality assurance and quality control (QA/QC) System. 4.-W. 2007.. 2001 (in Polish)..I.” Accred. Nofera. Dobecki.. M. LLC . J. 1999.. M. and Valcárcel. 4. P.Quality of Analytical Results 47 2. H. 2002. 7. Zapewnienie jakości analiz chemicznych. Prof.. Współczesna chemia analityczna. W. Wybrane zagadnienia. “Quality assurance in analytical measurements.. M.M. Hembeck. 2000. Chem. Hulanicki.. Warsaw. Assur. Qual. Paneva. Qual. 17–23. 2004. 418–422. 3.” Crit. Mermet. J.” Trends Anal. and Rios.. 6. M. 37.© 2009 by Taylor & Francis Group. Analytical Chemistry: a Modern Approach to Analytical Science.. “Analytical chemistry and quality.. 177–184.. 5. O. Qual.. Rev. Anal. V.” Accred. (ed). A. 2004 (in Polish).. Łódź. 8. Konieczka. 5.

The problem of traceability appeared with the first measurements carried out by man. in association with the development of metrological infrastructure.3 Traceability 3. or application of an item by means of recorded identification. traceability is defined not only as a property of a measurement result. National (measurement) standard — measurement standard recognized by national authority to serve in a state or economy as the basis for assigning quantity values to other measurement standards for the type of quantity concerned. (Measurement) standard (etalon) — realization of the definition of a given quantity. ISO 8402 [3]. but also as a 49 . traceability is defined as “the ability to verify the history. Traveling standard — measurement standard. 3.” In International Vocabulary of Basic General Terms in Metrology (VIM) [1]. initially in reference to measurements of physical properties. sometimes of special construction. Traceability — property of a measurement result whereby the result can be related to a reference through a documented unbroken chain of calibrations. Reference standard — measurement standard designated for the calibration of other measurement standards for quantities of a given type in a given organization or at a given location. used as a reference. Primary standard — measurement standard established using a primary reference measurement procedure. International (measurement) standard — measurement standard recognized by signatories to an international agreement and intended to serve worldwide. location. each contributing to the measurement uncertainty. but later with relation to chemical measurements [2].2  Introduction Comparison of measurement results is sensible only when they are expressed in the same units or on the same scale. with stated quantity value and associated measurement uncertainty. In the handbook Quality Management and Quality Assurance Vocabulary. the notion of traceability itself was formulated much later. LLC . Secondary standard — measurement standard established through calibration with respect to a primary measurement standard for a quantity of the same type.1 Definitions [1] Measurand — quantity intended to be measured. However. intended for transport between different locations. or created as an artifact. Working standard — measurement standard that is routinely used to calibrate or verify measuring instruments or measuring systems.© 2009 by Taylor & Francis Group. chosen by convention.

and the rationale and meaning behind it is presented in Figure 3. . a balance should be used that is regularly calibrated via weights with a calibration certificate that describes a reference to higher-order standard weights. For example. Knowing uncertainty values at each step of this chain of comparisons. in turn. Every day throughout the world. 1 kg mass standard (Sevres. SI FIGURE 3.2. one can qualify the uncertainty of the value measured.2  Rationale and meaning of traceability. millions of chemical analyses are carried out. should be calibrated against the national standards related to the international prototype kilogram. Such a series of comparisons is an uninterrupted chain illustrating the very property of traceability. The schematic presentation of traceability meaning is shown in Figure 3. In a general meaning. 6].1 [7]. for mass determination.50 Quality Assurance and Quality Control property of a reference standard. The obtained measurement results should be traceable to respective international standards [5. These. LLC . France) Official copy of mass standard National mass standard (1 kg) Mass measurement FIGURE 3. it can be described as a continuous and logical process that discourages weak or missing activity at any step of an analytic process.1  The idea of traceability — an example of mass determination [6].© 2009 by Taylor & Francis Group. and each has its own requirements concerning the quality of an obtained result [4]. which could burden or lower the effectiveness of the entire process.

traceability is one of the most important elements of the quality of a result. which means the necessity of obtaining a representative dose of the examined . because every physical or chemical operation can fracture this chain. if the determined substance is available as a certified reference material. This result must be related to a reference standard. Because results of measurements of physical and chemical properties are the basis of many decisions. then it can be treated as the last cell of an uninterrupted chain of comparisons. the science of measurement. Thus. apart from the scale calibration of a gauge.© 2009 by Taylor & Francis Group. In compliance with the content of VIM [1]: “traceability is a property of the result of a measurement or the value of a standard whereby it can be related to stated references. but to the value (property) represented by or produced using its [8]. and at every step uncertainty should be defined. respective scientific and legal metrology centers have emerged on an international scale. and in principle does not depend on the type of examined object. In compliance with the requirements of metrology. In measuring chemical values. 3. that is. In measuring physical properties. For chemical measurements.3 Role of Traceability in Quality Assurance/Quality Control system The accuracy of an analytical result depends directly on the material used for calibration.Traceability 51 The traceability could be achieved by comparison of result value with [2]: • • • • • SI unit value represented by well-stated standard value obtained by primary (absolute) method value obtained by reference (excellence) laboratory value obtained by group of laboratories in systematic PT scheme It should be noted that value of result is traceable not to the reference material (RM) or primary method. the most important feature of reliable measurement result is its traceability in relation to the recognized standard with well-known metrological characteristics. A critical step of assuring traceability in chemical measurements is the applied analytical procedure.” The quoted definition of traceability underlines the elements that are especially significant in chemical measurements. so that it may be expressed in suitable units. Moreover. Chemical measurements usually require a sample preparation step. usually national or international standards. traceability. the connection should be realized by means of an uninterrupted chain of comparisons. the result of a measurement depends substantially on the quality of the measuring instruments (rule. LLC . scale) used. through an unbroken chain of comparisons all having stated uncertainties. Assurance of traceability is realized by comparing given properties to a higher-order standard. Traceability is primarily a feature of a result of measurement obtained with the use of a given measurement procedure. that is. the result of measurement depends to a significant degree on the type of the sample and how the analytical procedure is conducted. thermometer.

for solid samples. in chemical measurements. and therefore the effectiveness of creating free atoms in the determined element. Thus the results can only be compared in the same measurement conditions. The homogeneity of a sample (heterogeneity of composition) significantly influences the determination of the representative portion of the examined material. the value of the measurand depends on the applied methodology and/or on the measurement conditions.© 2009 by Taylor & Francis Group. LLC . For example. . This effect is extremely important with consideration to the entire chemical measurement and must be taken into account when validating a given measurement procedure. for example. and extraction (just to mention the most important physicochemical processes). calibration of instruments is not a significant source of problems. an analytical sample should be sufficiently large so that grain size is not a source of heterogeneity. and proving traceability is considerably more difficult. and. While planning analytical conduct. Interferences. In the case of chemical measurements. it is well known that a type of acid mixture used for mineralization influences the atomic absorption signal. As has been noted earlier. an extremely important element that assures the quality of chemical measurement results is the validation of the entire measurement procedure and the estimated influence of sample components on the ultimate result of the measurement. an analyst must allow for the heterogeneity of a sample’s composition. depend on the type of determined substance (analyte) and the type of matrix of the sample. the influence of sample components on the analytical signal.52 Quality Assurance and Quality Control material. The greatest problem is assuring the traceability of the entire analytical process. In chemical measurements. The most important difficulties are: • • • • • • • identifying the object of measurement (object of determination) interferences homogeneity of a sample (heterogeneity of composition) persistence of the sample sample preparation correctness of measurement realization determination of uncertainty Determination of the measurand is a crucial element in the selection of an analytical procedure. As noted earlier. Therefore. a result of determination depends on the components accompanying the analyte. when measuring physical properties. In chemical measurements. the sample type does not have a significant influence on the measurement result. The type of acid affects the process of atomization. the chain of connections with standards is always broken when a sample is physically or chemically modified in the analytical process. hence. dissolution or mineralization of a sample. that is. enrichment. In most cases. Traceability determination in chemical measurements is associated with many difficulties resulting from the need for sample preparation before the measurement process itself. there is no organized metrological system similar to physical measurements realized by a system of standardizing laboratories. For this reason. the notion of accuracy is difficult to define.

which can assure traceability to standards. It is necessary to use reference standards for which traceability may be shown and which have known uncertainty. The preparation of a sample is the most important element causing difficulty in maintaining traceability. In some cases. For example. which can be assured by calibration using suitable calibrants. and consequently international agreement on measurements [9].© 2009 by Taylor & Francis Group. Traceability should be shown for each parameter of a given procedure and should be carried out by calibration with suitable standards. a pH measurement requires calibration of an instrument and measurements at a suitable temperature. The traceability of measurement results depends on. Correct realization of a measurement depends primarily on the efficiency of the measuring instrument used and the maintenance of suitable measurement conditions. For trace analysis. which in turn have a connection with the value obtained in reference measurements Reference values should come from specialist laboratories with good international reputations. 12]: • Assuring the correct performance of an instrument (instrument calibration) • Determining a clear dependence between a determined signal and a determined property (analytical calibration) For reference materials reproducing the chemical properties. LLC . the problem of traceability assurance involves the accessibility of standards with a required level of analyte concentration. the composition of a sample can change even over several minutes. among other things.Traceability 53 The stability of a sample determines the measurement duration. which implies a necessity for a detailed plan of action. hence the necessity for exact knowledge concerning how the sample behaves over time. Every physicochemical operation disrupts the chain of traceability. which obtain traceability. the proper functioning of instruments. traceability for chemical measurements can be determined in two ways [10]: • By comparing an obtained value with reference measurements • By referring an obtained value to reference standards. . A procedure enabling the determination of correlations between the value of a signal (indicated by an instrument) and the concentration of the examined substance in the sample is called calibration. Determination of the uncertainty of a result is an integral part of traceability assurance. An important part at this step is played by reference materials. Uncertainties in reference standards comprise the uncertainty of a result obtained by a comparison with these standards. that is to say. determined with a suitable accuracy (higher than the accuracy in the applied analytic methodology). In practice. fulfilling the traceability requirement for a typical analytical procedure demands the use of matrix reference materials. The calibration step is used for [11.

it is necessary to: • Determine the measurand. calibration. the assurance of the obtained analytical results referred to specific reference materials by an uninterrupted chain of comparisons of uncertainties associated with suitable reference materials (certification and history of their production). LLC . For the purpose of obtaining a full and correct picture. nor determined by means of precisely defined physical and chemical measurement methods. that is. is an element of analytical chemistry that is currently given considerable attention. in order to determine the traceability of a given analytical procedure. analytes in samples occur in trace or ultratrace amounts. • Determine a strategy for proving traceability by selecting suitable standards and determining procedure calibration. and hence assuring the reliability of measurements. • Determine the uncertainty of the applied measurement procedure. Assuring traceability. • Prove (by validation) the correctness of the selected measuring conditions and the model equation. and other parameters associated with the specific instrument. the properties of standard values that can be related to reference materials by an uninterrupted chain of comparisons of uncertainties associated with suitable reference materials and supplied documentary evidence giving the history of their production (in which significant properties such as homogeneity. • Select a suitable measurement procedure and record a respective model equation.© 2009 by Taylor & Francis Group. In compliance with requirements stated in the EURACHEM /CITAC Guide [15]. the possibility of obtaining traceable results after a correct process of validating all analytical conduct. with special attention paid to maintenance. • The traceability of an instrument. • The traceability of analytical methodology (procedures). and origin must be clearly presented). Many reference materials may have properties that for various reasons cannot be measured in units of mass or quantity.54 Quality Assurance and Quality Control Here it must be remembered that very frequently. That is why the notion of traceability and the associated notion of uncertainty are also two key problems in present-day metrology in analytical chemistry. Examples of such reference materials are biological reference materials attributed to a respective international unit by the World Health Organization. traceability should be considered in four ways [14]: • The traceability of analytical results. and preparing suitable reference materials poses an immense challenge. number of hours used. stability. . that is. and repairs. that is. damage. that is. a detailed and up-to-date history of the instruments containing descriptions of their installation. This has an undoubtable influence on the cost of preparing reference materials. sample processing. and also technological reference materials [13]. • The traceability of the applied standards.

Traceability

55

As noted earlier, one of the most important tools used for the purpose of traceability assurance in chemical measurements is the use of certified reference materials, which are extremely useful for: • Estimating the accuracy of new analytical procedures • Comparing different methods • Comparing and testing the competence of different laboratories Realization of traceability in chemical measurements by means of reference materials can be attained by using pure standard substances for calibration or suitable certified reference materials. It is very important to purchase reference material from a reputable distributor, which will assure the maintenance of traceability for a given value together with a given uncertainty value. The most important criteria for the selection of reference materials are primarily the agreement of matrix and concentrations of the determined substance. Moreover, it is necessary to allow for uncertainty provided by the manufacturer and to estimate to what extent this will be important in the uncertainty budget of the applied measurement procedure.

3.4 Conclusion
The main sense of traceability is to enable comparability of measurement results — either compare results of the measurements on the same sample or compare results on different samples [12]. In theory, all measurements can be tracked back to the base seven SI units [16]. Traceability is highly connected with uncertainty, comparability, utility, reliability, and validity. Traceability and uncertainty are necessary parameters for obtaining reliable results.

References
1. International Vocabulary of Metrology — Basic and General Concepts and Associated Terms (VIM), Joint Committee for Guides in Metrology, JCGM 200, 2008. 2. Valcárcel, M., and Rios A., “Traceability in chemical measurements for the end users,” Trends Anal. Chem., 18, 570–576, 1999. 3. Quality Management and Quality Assurance Vocabulary, ISO 8402, Geneva, 1994. 4. Walsh, M.C., “Moving from official to traceable methods,” Trends Anal. Chem., 18, 616–623, 1999. 5. De Bièvre, P., Kaarls, R., Peiser, H.S., Rasberry, S.D., and Reed W.P., “Protocols for traceability in chemical analysis: Part II. Design and use,” Accred. Qual. Assur., 2, 270– 274, 1997. 6. De Bièvre, P., and Taylor, P.D.P., “Traceability to the SI of amount of substance measurements: from ignoring to realizing a chemist’s view,” Metrologia, 34, 67–75, 1997. 7. Thomson, M., “Comparability and traceability in analytical measurements and reference materials,” Analyst, 122, 1201–1205, 1997. 8. King, B., “The practical realization of the traceability of chemical measurements standards,” Accred. Qual. Assur., 5, 429–436, 2000. 9. ISO Guide 35, Certification of reference materials. General and statistical principles, ISO, Geneva, 1989.

.© 2009 by Taylor & Francis Group, LLC

56

Quality Assurance and Quality Control

10. Bulska, E., and Taylor, Ph., “On the importance of metrology in chemistry,” in Namieśnik, J., Chrzanowski, W., and Żmijewska, P., eds., New Horizons and Challenges in Environmental Analysis and Monitoring, CEEAM, Gdańsk, 2003. 11. Valcárcel, M., and Rios, A., “Is traceability an exclusive property of analytical results? An extended approach to traceability in chemical analysis,” Fresenius J. Anal. Chem., 359, 473–475, 1997. 12. Williams, A., “Traceability and uncertainty — a comparison of their application in chemical and physical measurement,” Accred. Qual. Assur., 6, 73–75, 2001. 13. ISO Guide 30, Trends and definitions used in connections with reference materials, ISO, Geneva, 1992. 14. Marschal, A., Andrieux, T., Compagon, P.A., and Fabre, H., “Chemical metrology — QUID?” Accred. Qual. Assur., 7, 42–49, 2002. 15. EURACHEM/CITAC Guide, Traceability in Chemical Measurements. A guide to achieving comparable results in chemical measurement, 2003. 16. Buzoianu, M., and Aboul-Enein, H.Y., “The traceability of analytical measurements,” Accred. Qual. Assur., 2, 11–17, 1997.

.© 2009 by Taylor & Francis Group, LLC

4 Uncertainty
4.1 Definitions [1–3]
Uncertainty of measurement — nonnegative parameter characterizing the dispersion of the quantity values being attributed to a measurand, based on the information used. Definitional uncertainty — component of measurement uncertainty resulting from the finite amount of detail in the definition of a measurand. Standard uncertainty u( xi ) — uncertainty of a result xi of a measurement expressed as a standard deviation. Combined standard uncertainty uc( y) — standard measurement uncertainty that is obtained using the individual standard measurement uncertainties associated with the input quantities in a measurement model; standard uncertainty of a result y of a measurement when the result is obtained from the values of many of other quantities equal to the positive square root of a sum of terms, the terms being the variances or covariances of these other quantities weighted according to how the measurement result varies with these quantities. Uncertainty budget — statement of a measurement uncertainty, of the components of that measurement uncertainty, and of their calculation and combination. Expanded uncertainty U — product of a combined standard measurement uncertainty and a factor larger than the number 1. Coverage factor k — number larger than 1 by which a combined standard measurement uncertainty is multiplied to obtain an expanded measurement uncertainty; a coverage factor is typically in the range of 2–3, and for an approximately 95% level of confidence, k = 2. Type A evaluation (of uncertainty) — evaluation of a component of measurement uncertainty by a statistical analysis of measured quantity values obtained under defined measurement conditions. Type B evaluation (of uncertainty) — evaluation of a component of measurement uncertainty determined by means other than a type A evaluation of measurement uncertainty. Relative uncertainty ur ( xi ) — standard measurement uncertainty divided by the absolute value of the measured quantity value.

4.2  Introduction
Decisions made in many fields of science and other domains of life are based on the results of analytical studies. It is therefore obvious that their quality is increasingly important. Uncertainty of measurement is a component of uncertainty for all individual steps of an analytical procedure [4–7]. Hence, it is necessary to determine the sources and types of uncertainty for all these steps [8–10].
57
.© 2009 by Taylor & Francis Group, LLC

58

Quality Assurance and Quality Control

The main sources of uncertainty during sample analysis while using an appropriate analytical procedure may be [11]: • Inaccurate or imprecise definition of the measurand • Lack of representativeness at the step of collecting a sample from an examined material object • Inappropriate methodology of determinations • Personal deviations in reading the analog signals • Not recognizing the influence of all the external factors on the result of an analytical measurement • Uncertainty associated with the calibration of an applied measurement instrument • Insufficient resolution of the applied measurement instrument • Uncertainties associated with the applied standards and/or reference materials • Uncertainties of parameters determined in separate measurements and which are used in calculating the final result, such as physicochemical constants • Approximations and assumptions associated with using a given instrument, applied during measurement; or • Fluctuations of the measurement instrument gauge, over the course of repeated measurements, with seemingly identical external conditions There is a difference between measurement error and uncertainty. The error is a difference between the determined and expected values, and uncertainty is a range into which the expected value may fall within a certain probability. So the uncertainty cannot be used to correct a measurement result. This difference is schematically presented in Figure 4.1 [11].

4.3 Methods of Estimating Measurement Uncertainty
There are several approaches for uncertainty estimation [12, 13]: • Bottom-up — based on an identification, quantification, and combination of all individual sources of uncertainty of measurement. The overall uncertainty is derived from the uncertainties of individual components. This method has high complexity and because of that it needs considerable time and effort; this approach is adapted by EURACHEM [2, 14].
Error

µx

xm Uncertainty

FIGURE 4.1  Schematic presentation of difference between error and measurement uncertainty [11].

.© 2009 by Taylor & Francis Group, LLC

and the standard uncertainty for each of them must be determined.© 2009 by Taylor & Francis Group. such as in the case of a standard’s purity. the following conditions must be satisfied: 1. value. 2. LLC . in order to determine the uncertainty of analysis result. and describes the relation between uncertainty and analyte content.1) where y is the value of a result and x1. The observed quantity and the searched parameter (result) must also be clearly described. The relation is described as follows: y = f ( x1 . Type B uncertainty is strictly associated with the probability distribution described by the distribution of a variable. unit. and its number of degrees of freedom. which has the form of algebraic expression. . there are two methods for calculating standard uncertainty. .or intralaboratory validation studies (precision. trueness. . • Validation-based — based on inter. x 2  x n ) (4. Type A uncertainty is equal to a standard deviation of an arithmetic mean. According to the Guide to the Expression of Uncertainty in Measurement [2]. along with the unit in which it is expressed. when a variable has a rectangular distribution. • Top-down — based on data obtained from interlaboratory studies (precision). standard uncertainty. x2. Values must be assigned to all the possible parameters that could affect the final result of the analysis. Each measurand has a name. robustness). An appropriate mathematical model ties the value of a determination result (the one to be determined) with the observed values (measurement values). 4. xn — measurement values.Uncertainty 59 • Fitness-for-purpose — based on a definition of single parameter called the fitness function. The measurement procedure and the measurand must be defined. Modeling (usually mathematical modeling) must be applied to calculate the analysis result based on the measured parameters. and in turn allows comparison of results obtained in interlaboratory studies and helps users to decide the significance of any difference between the obtained result and the reference value.3. The measurand in a given measurement must be clearly defined. As noted before. 3.1 Procedure for Estimating the Measurement Uncertainty According to Guide to the Expression of Uncertainty in Measurement Determining the uncertainty of a measurement increases its reliability. For example. • Robustness-based — based on robustness tests from interlaboratory studies. Calculation of uncertainty for the result of measurement is very easy and less time-consuming than a bottom-up approach. . the variable may assume (with equal .

Uncertainty given by the manufacturer have been calculated for coverage factor k = 2. Cst = 1001 ± 2 mg/dm3 Solution: Because value for k is given. Data: Standard solution concentration.15 mg/dm3 Excel file: exampl_uncert01. standard uncertainty is calculated accordingly: 2 u(Cst ) = k Excel file: exampl_uncert02. the calculated standard uncertainty is a 6 Example 4. Cst = 1001 ± 2 mg/dm3 Solution: Due to no additional information.xls Example 4. based on data given by the manufacturer. but the occurrence of the mean value from the range is the most probable. +a〉. which means that the value is in the range 〈– a. +a〉. When a variable has a triangular distribution.60 Quality Assurance and Quality Control probability) a value in the range 〈– a. LLC . we assume a rectangular distribution.xls u(Cst ) = 1 mg/dm3 .2 Problem: Calculate the standard uncertainty for the concentration of magnesium in a standard solution. Data: Standard solution concentration. u(Cst ) = 2 3 u(Cst ) = 1.© 2009 by Taylor & Francis Group.1 Problem: Calculate standard uncertainty for the concentration of magnesium in a standard solution. based on data given by the manufacturer. +a〉 . and the calculated standard uncertainty is a 3 where a is the midpoint of the range 〈– a.

Vfl = 500 ± 0.1). . For a given mathematical model that binds the final results of analysis with measured parameters (Equation 4.8 6 u(Vfl ) = 0.Uncertainty 61 Example 4.© 2009 by Taylor & Francis Group. Data: Volume.5) If the value of the analytical result is a quotient/product of the measurement values.xls 0.32 cm3 4. LLC . standard uncertainty is calculated by using the principle of uncertainty propagation expressed in the following formula: 2 uc ( y) = ∑ i =1 n  δf  2 ux   δxi   ( i) 2 (4.3) then the value of the combined uncertainty is described by the following equation: uc( y) = u(2x1 ) + u(2x2 ) + … + u(2xn ) (4. it is more convenient to apply relative uncertainties.2) When the value of an analytical result is the sum or difference of the measurement values y = x1 + x 2 + … + x n (4. Assume triangular distribution.3 Problem: Calculate the standard uncertainty for the determination of volume 500 cm3 using volumetric flask.4) Due to the very frequent occurrence of individual measurement values being expressed in different units. based on data given by the manufacturer. The applied principles of uncertainty propagation in calculating the standard uncertainty of an analytical result. A relative uncertainty is described by the following relation: ur ( xi ) = u( xi ) xi (4.8 cm3 Solution: u(Vfl ) = Excel file: exampl_uncert03.

and present indication results.5 mg/dm3.6 (%) . a standard solution was obtained with the predetermined concentration. calculate the following: • The value of the combined and expanded uncertainty (for k = 2) for the obtained standard solution concentration. Therefore. LLC .7) 5. the basic solution was diluted as follows: We took 1 cm3 of basic sample solution by using a pipette with a volume of 1 cm3 and transferred it to a volumetric flask with a volume of 100 cm3.0017 (mol/dm3) cNaOH ± U(k = 2) = 0. cNaOH ± U(k = 2) = 0. With the aim of obtaining a standard solution with a concentration of ~0.1038 mol/dm3 ± 1. the standard uncertainty should be multiplied by an appropriate coverage factor k.1038 ± 0.© 2009 by Taylor & Francis Group. a correctly presented result of an analysis should be as follows: or Example 4. To establish a uniform distribution for each of the measured parameters. for which the expanded uncertainty has been calculated Thus. the final result of an analysis comprises the following: • Determination of the measured value and its unit • The result with the expanded uncertainty value ( y ± U. After filling the flask to the line and mixing the solution. along with units for y and U ) • k factor value. 5 cm3 of solution was taken from it with the help of a pipette with a volume of 5 cm3 and was transferred to a volumetric flask with a volume of 100 cm3 and after being filled to the line. with a basic diluted solution with a concentration of 1001 ± 2 mg/dm3. Presentations of the final result of the analysis as: result ± expanded uncertainty (after using an appropriate k factor).6) then the value of the combined relative uncertainty is described by the following equation: ur ( y) = ur2( x1 ) + ur2( x1 ) + … + ur2( xn ) (4. Uncertainty calculated according to the aforementioned equation is a combined standard uncertainty of the final determination. To calculate the value of the expanded uncertainty.62 Quality Assurance and Quality Control y= x1x 2… x 3… (4.4 Problem: A standard Mg2+ solution was prepared.

25 ur Cst Vp1 Vf1 Vp2 Vf2 k 0.012 mg/dm3 Example 4.25 18.501 ± 0. the basic solution was diluted as follows: A basic sample solution of 10 cm3 was taken using a pipette with a volume of 10 cm3 and was transferred to a volumetric flask with a volume of 100 cm3.4% 0. % 6.0122 0.5 mg/dm3.2 u(Vp2) 0. LLC .0012 0. With the aim of obtaining a standard solution with a concentration of ~0.03 u(Vf2) 0.02 u(Vf1) 0.75 6.0115 0.0122 mg/dm3 2.0035 0.5 Problem: A standard Mg2+ solution was prepared. with a basic diluted solution having a concentration of 1001 ± 2 mg/dm3.0012 2 c ur(c) U(c) U Result Excel file: exampl_uncert04.© 2009 by Taylor & Francis Group.0012 0. Data: Standard solution concentration Pipette 1 volume Flask 1 volume Pipette 2 volume Flask 2 volume Uncertainty of single measurement Cst 1001 Vp1 1 Vf1 100 Vp2 5 Vf2 100 u(Cst) 2 u(Vp1) 0.50 6.5005 mg/dm3 0.2 Rectangular (R) or triangular (T) mg/dm3 cm3 cm3 cm3 cm3 mg/dm3 cm3 cm3 cm3 cm3 R 63 Distribution Solution: Relative Uncertainty Contribution.xls 0.25 62. After .Uncertainty • The participation percentage of each of the standard uncertainty values in the determined values of the combined uncertainty.

08 18.18 9.18 9.0023 0. and after being filled to the line a standard solution was obtained with the predetermined concentration. Data: Standard solution concentration Pipette 1 volume Flask 1 volume Pipette 2 volume Flask 2 volume Pipette 3 volume Flask 3 volume Uncertainty of single measurement 1001 Cst Vp1 10 Vf1 100 Vp2 5 Vf2 100 Vp2 10 Vf2 100 u(Cst) 2 u(Vp1) 0. To establish a uniform distribution for each of the measured parameters.09 18. and present indication results.04 u(Vf2) 0.03 u(Vf2) 0.04 u(Vf1) 0. 10 cm3 of solution was taken from it with the help of a pipette with a volume of 10 cm3 and was transferred to a volumetric flask with a volume of 100 cm3.0012 0.0023 0.0012 2 .2 u(Vp2) 0.09 ur Cst Vp1 Vf1 Vp2 Vf2 Vp3 Vf3 k 0.2 u(Vp2) 0.64 Quality Assurance and Quality Control filling the flask to the line and mixing the solution. After filling the flask to the line and mixing the solution.28 9.0012 0. • The participation percentage of each of the standard uncertainty values in the determined values of the combined uncertainty.0035 0. 5 cm3 of solution was taken from it with the help of a pipette having a volume of 5 cm3 and was transferred to a volumetric flask with a volume of 100 cm3. calculate the following: • The value of the combined and expanded uncertainty (for k = 2) for the obtained standard solution concentration.09 27. % 9.0012 0. LLC .© 2009 by Taylor & Francis Group.2 Rectangular (R) or triangular (T) mg/dm3 cm3 cm3 cm3 cm3 cm3 cm3 mg/dm3 cm3 cm3 cm3 cm3 cm3 cm3 R Distribution Solution: Relative Uncertainty Contribution.

LLC .5005 mg/dm3 0.72 mg).© 2009 by Taylor & Francis Group.6 Problem: A standard sample was weighed for the preparation of a standard solution.5005 ± 0.72 mtarra 332. Calculate the combined standard uncertainty of the obtained standard solution concentration.0053 mg/dm3 65 Example 4. 332.83 mg 0. Assuming .408 mg 2 0.5 mg.55 mbrutto 0. The mass measurement was carried out using an analytical scale.83 ± 0. for which its manufacturer gave a measurement uncertainty of 0.0053 mg/dm3 1.5 u(mtarra) 0.xls 144.5 u(mbrutto) Rectangular (R) or triangular (T) mg mg mg mg R Solution: mnetto u(mnetto) k U(mnetto) Result Excel file: exampl_uncert06.82 mg Example 4. calculate the expanded uncertainty of the mass measurement.6) was put into a measurement flask (250 cm3) for which the manufacturer provided an uncertainty value equal to 0. Give a correct presentation of the mass measurement result.82 mg 144. Calculate the standard uncertainty of the mass measurement.0053 0. Assume a rectangular distribution of the parameters. Assume a rectangular distribution of the parameter.55 mg) and net (container.1% 0.4 cm3.7 Problem: The weighed standard sample (Example 4. Assuming the value of the coverage factor to be 2. Data: Mass (tarra) Mass (brutto) Uncertainty of single measurement Distribution 187.xls 0. 187. The mass was calculated as the difference of two mass measurements: gross (container with a sample.Uncertainty c ur(c) U(c) U Result Excel file: exampl_uncert05.

5793 ± –0. LLC . calculate the expanded uncertainty of the concentration.66 Quality Assurance and Quality Control the value of the coverage factor to be 2.72 332. sampling 1 cm3 of the original solution using a pipette for which the manufacturer provided an uncertainty value of 0.0023 mg/cm3 Example 4. Calculate the combined standard uncertainty of the obtained standard solution concentration.5793 mg/cm3 0.4 u(Vflask) Rectangular (R) or triangular (T) mg mg cm3 mg mg cm3 R Solution: Concentration ur(concentration) k U(concentration) Result Excel file: exampl_uncert07. Assume a rectangular distribution of parameters. Give a correct presentation of the result.7) was dissolved by a 1:10 ratio. for which the manufacturer provided an uncertainty value 0.2 Rectangular (R) or triangular (T) mg mg cm3 cm3 cm3 mg mg cm3 cm3 cm3 R Distribution .05 u(Vpipette) 0.72 332. Give a correct presentation of the result. and dissolving in a measurement flask (10  cm3).5 u(mbrutto) 0. calculate the expanded uncertainty of the concentration.5 0.2 cm3.xls 0.5 u(mbrutto) 0. Data: Mass (tarra) Mass (brutto) Flask volume Flask volume Pipette Uncertainty of single measurement mtarra 187. Assuming the value of the coverage factor to be 2.0023 mg/cm3 0.00199 2 0.05 cm3.8 Problem: The obtained standard solution (Example 4.55 mbrutto 250 Vflask1 Vflask2 10 Vpipette 1 u(mtarra) 0.5 0.4 u(Vflask1) u(Vflask2) 0.© 2009 by Taylor & Francis Group. Data: Mass (tarra) Mass (brutto) Flask volume Uncertainty of single measurement Distribution mtarra 187.55 mbrutto 250 Vflask u(mtarra) 0.

Weighing of pure substance Dilution STANDARD SOLUTION FIGURE 4. the value of uncertainty can be estimated as a confidence interval.0579 mg/cm3 0.5 Uncertainty and Confidence Interval In some cases. and n is the number of measurements.2  Flow diagram for the procedure of preparation of standard solution. .013 mg/cm3 67 4.3. The flow diagram and the Ishikawa diagram for the procedure of preparation of a standard solution are presented in Figures 4.Uncertainty Solution: Concentration ur(concentration) k U(concentration) Result Excel file: exampl_uncert08. LLC . respectively. if one of the parameters has a dominating influence over the uncertainty budget.4 Tools Used for Uncertainty Estimation Correct estimation of uncertainty needs an understanding of whole analytical procedure by analyst. 4. then the expanded uncertainty may be calculated according to the following relation: U=k SD n (4. The basic principle of the uncertainty propagation is underlining the influence of the quantity with the highest value.2 and 4. Therefore.1155 2 0. or cause-and-effect.xls 0.© 2009 by Taylor & Francis Group. If that dominating parameter is the repeatability of measurements.134 mg/cm3 0.058 ± 0. 16]. 6]: • Flow diagram — drawn on the basis of information presented in detail in a standard operating procedure.8) where SD is the standard deviation. calculation of uncertainty may be limited to the calculation based on the value of that parameter. • Ishikawa. or fishbone diagram — shows the influence parameters (sources of uncertainty) of a whole analytical procedure [15. The most helpful tools used for that are [4.

54 .68 Mass of pure substance Quality Assurance and Quality Control Balance calibration Balance calibration Gross mass Tare mass Concentration of analyte in standard solution Flask calibration Temperature Volume of solvent Purity of substance FIGURE 4.9 Problem: The concentration of mercury was determined in water by using the cold vapor atomic absorption spectrometry (CVAAS) technique. the parameter t ≈ 2 (Table A.© 2009 by Taylor & Francis Group.05.14 77.05 and the number of degrees of freedom f →∞.9) For a level of significance of α = 0. Given this condition. the coverage factor k = 2. Appendix).53 72.1.13 76.3  Ishikawa diagram for the procedure of preparation of standard solution. calculate the expanded uncertainty of the determination result for k = 2. The series involved four determinations. LLC . the aforementioned equations are thus consistent. μg/dm3:   1 2 3 4 Data 71. f ) SD n (4. the value of confidence interval could be calculated as: ∆x sr  = t (α . Considering the unrepeatability as the main component of the uncertainty budget. Example 4. Provide a correct presentation of the determination result. Data: result series. On the other hand. For a level of significance of α = 0.

Uncertainty Solution: Mean SD k U Result Excel file: exampl_uncert09. it is possible to determine and identify the uncertainty of the determined regression curve through the determination of confidence intervals. Those intervals are determined using a correlation that is described by the following equation: where ∆yi = Y ± SDxy ⋅ t(α . y ) • Uncertainty due to the determination of the reference value for standard samples — u( xsmpl .9 μg/dm3 69 4. drawn based on Equations (1. f = n− 2) 1 ( xi − x m )2 + n Qxx (4.63)–(1. a calibration curve technique is usually used.9075 μg/dm3 74. Using a calibration curve.9075 μg/dm3 2 2. usually using a method of consecutive dilutions • Incorrect approximation of measurement points using a regression curve Figure  4.10) ∆yi = confidence interval of the calculated value Y for a given value xi Y = values calculated based on the regression curve equation for given values xi .68) in Chapter 1. Standard uncertainty due to this step should be included in the uncertainty budget. This step of the analytical procedure has influence on the combined uncertainty of the determination result for the real sample.6 Calibration Uncertainty A decisive majority of analytical measurements involve a calibration step.xls 74. xstd ) i • The influence of the manner of preparing the standard samples. At the calibration step.4 presents an example of a calibration graph along with the marked uncertainty values associated both with the reading of the signal values and the reference values.335 μg/dm3 2.© 2009 by Taylor & Francis Group. 17–19]: • Repeatability of reading the value of a signal y both for standard samples (based on measurements for which the calibration curve is determined) and for study samples — u( xsmpl .3 ± 2. LLC . which is associated with the relative (comparative) character of measurements. which is determined using linear regression. There are four sources of uncertainty due to the calibration step that can influence the standard uncertainty of a single measurement u( xsmpl ) [8.

© 2009 by Taylor & Francis Group. y ) may be calculated using the determined regression parameters according to the following relationship: u( xsmpl . y ) = SDxy b 1 1 ( xsmpl − x m ) + + p n Qxx 2 (4.9989 6 8 Content.564x – 0. ppm 10 12 14 FIGURE 4. SDxy = residual standard deviation t(α. LLC . f = n − 2) = Student’s t test parameter n = total number of standard samples used for the determination of the calibration curve (number of points) xi = value x for ∆yi is calculated xm = mean value x (x is most frequently the analyte concentration and is the mean of all the concentrations of a standard solution for which the measurement was made to make a standard curve) Qxx = parameter calculated according to a relation described by the equation: Qxx = ∑ (x − x ) i m i =1 n 2 (4.11) Standard uncertainty for xsmpl due to the uncertainty of calibration and linear regression method u( xsmpl . y ) = standard uncertainty for the determination of the xsmpl concentration due to the application of the determined calibration correlation b = direction coefficient of the calibration curve p = number of measurements (repetitions) carried out for a given sample .4  An example of a calibration graph along with the marked uncertainty values associated both with the reading of the signal values and the reference values.0291 r = 0.70 8 7 6 Signal 5 4 3 2 1 0 0 2 4 Quality Assurance and Quality Control y = 0.12) where u( xsmpl .

Figure 4. standard uncertainty due to the application of standard solutions at the calibration step may be described by the following equation: u( xsmpl .5  A calibration curve along with the marked confidence intervals and the determined uncertainty value for the determination of an analyte’s concentration in an examined sample. ppm 10 12 14 y = 0.564x – 0. y ) . If each standard sample is prepared by consecutive dilutions. y ) i (4.14) Such an uncertainty value does not allow for the uncertainty associated with the manner of standard sample preparation. its value may be estimated by only considering the number of standard samples used at the calibration stage. Because usually only one basic standard is used and then appropriate standard solutions are made (consecutive dilutions). xstd ) << u( xsmpl .5 presents a calibration curve along with the marked confidence intervals and the determined uncertainty value for the determination of an analyte’s concentration in an examined sample. LLC . then the uncertainty budget must allow for the standard uncertainties associated with the step of standard sample preparation. The value of uncertainty for the determination of analyte concentration in the applied standard samples is usually significantly smaller compared to the uncertainty associated with the calculation of analyte concentration based on the determined calibration function: u( xsmpl . xstd ) ≈ i u( xstd ) i n (4.13) Therefore. Usually. the standard uncertainty of a result.0291 r = 0.Uncertainty 8 7 6 Signal 5 4 3 2 1 0 0 2 4 6 8 Content. associated with an applied calibration technique. . requires only the value u( xsmpl .© 2009 by Taylor & Francis Group.9976 71 FIGURE 4.

41 4.54 3. ppm 2 2 2 4 4 4 6 6 6 8 8 8 10 10 10 12 12 12 Data y. ppm:     1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Result for sample Number of measurements for sample x.44 5.66 7. Calculate: • Regression parameters of the calibration curve • Confidence intervals • Uncertainty value of the determination value for the real sample due to calibration.32 2.72 Quality Assurance and Quality Control Example 4. making three independent measurements for each of the solutions.51 6.10 Problem: A calibration curve was determined using determinations of analyte concentration in samples of six standard solutions.67 5.97 6.2 1.12 1. Signal 1.78 6. for which three independent measurements were made and the result was calculated using the determined curve Data: results.76 5.33 3.59 ppm 3 .08 2.12 4.© 2009 by Taylor & Francis Group.11 2. LLC .32 4.23 3.

7 Conclusion Each analytical result derives from a conducted measurement.9976 210 0.9951 73 18 0. b Intercept.12 Excel file: exampl_uncert10.0291 0. Appendix) Graph: 8 7 6 Signal 5 4 3 2 1 0 0 2 4 6 8 Content.5642 −0. LLC . . The ultimate goal for an analyst is to obtain a result that will most reliably reflect the expected (actual. r Qxx Uncertainty for result due to calibration Relative uncertainty for result due to calibration t(α = 0.Uncertainty Solution: n Slope.16 ppm 2.© 2009 by Taylor & Francis Group.1. f = n − 2) (from Table A. a Residual standard deviation. SDx.1% 2.05.0291 R2 = 0.5642x – 0.1433 0. the basic tool for any analyst.xls 4.y Regression coefficient. The certainty of the analytical result depends on the uncertainties occurring at all the steps of an analytical procedure. real) value. ppm 10 12 14 y = 0.

.L. “The role of and place of method validation in the quality assurance and quality control (QA/QC) System. Qual. A. 7. Lis.” Accred. Therefore. 1998.. “Uncertainties related to sampling and their impact on the chemical analysis of groundwater. J. Bioanal. 7. A. Risk.-M. and Williams. “Chemical metrology. 15–24. “Application of cause-and-effect analysis to potentiometric titration. Uncertainty occurs always and at any step of a measurement procedure. 14. and Fouillac. Springer. G. Qual... 2nd edn.... S... 2007. Uncertainty is a basic property of each measurement. S. Guide to the Expression of Uncertainty in Measurement (GUM). 2005. E. 37. and Mecozzi.. Rev. 1158. and more exactly for each measurand. Populaire. R.. Geneva 1993. 6.J. 2004. V. Chem.. Assur. Ellison. Ellison.. Chem. 1652–1661. A.” J.. Kadis.L. References 1. “Trends in quality in the analytical laboratory: I.. Kufelnicki. 3. M.” Accred. 8. Traceability and measurement uncertainty of analytical results. and Campos Gimenez. Armishaw... chemistry. G. Konieczka. 2. 3. 10. International vocabulary of metrology — Basic and general concepts and associated terms (VIM). Rosslein...” Accred. A.© 2009 by Taylor & Francis Group. 15.” Crit.L. V.. S. LLC . Taverniers. Qual. and Barwick.. “Uncertainty and traceability: the view of the analytical chemist. Manag. 2005... “Uncertainty in environmental analysis: theory and laboratory studies. 1998. “Evaluating uncertainty in analytical measurements: the pursuit correctness. I.” in Fajgelj. A. Part 1. 23. Principles of an approach using cause and effect analysis. A. Belli.” Analyst. S. U.. 2002. O..J. and Rauret. and De Loose. E.” Accred. 218–224. 2007. Muse. Combining and Reporting Analytical Results. M....R. 2007. 2008. 173–190. 8.” Trends Anal. 10. 92–94. 2000. RSC. Conti. “Introduction to measurement uncertainty in chemical analysis. and Sansone. 485–493. Chem. 9. and the uncertainty of chemical measurements. 3. 1998.. “A simplified approach to the estimation of analytical measurement uncertainty. eds. 16. Assur. 237–241.. EURACHEM. JCGM 200. A..” Anal. Combined uncertainty covers all sources of uncertainty that are relevant for all analyte concentration levels. Roy. Love. 1387–1392... Assess.. P.E..” Accred. Anal. 2004. 382. Van Bockstaele. M. P. 311–335. “Estimating measurement uncertainty in an afternoon. it is necessary to determine the sources and types of uncertainty for individual steps of an analytical procedure. 11. Quantifying Uncertainty in Analytical Measurements.” Int. 13. Chromatogr. 480–490.. 185–193. 123. 5.R. 12. A.. “Measurement uncertainty. Meyer. Williams. 95–100. ISO. J.74 Quality Assurance and Quality Control The most crucial parameter affecting a measurement result’s uncertainty is the parameter with the highest uncertainty value. 5. 4. It is a “key indicator” of both fitness-for-purpose and reliability of results. A case study in the practical application of measurement uncertainty. Sahuquillo. it is not a property that should result in additional difficulties during the measurement procedure. M. 2005. Assur. M. 23.” Trends Anal. and Meinrath. Joint Committee for Guides in Metrology. Qual. Berlin.. “Using validation data for ISO measurements uncertainty estimation. Chem. ... Assur. Hence. Qual.R. 2003. Assur.

“Guidelines for calibration in analytical chemistry. K. 1991. J. .” LC-GC.” Analyst. Danzer.N. 70. 19. 531–532.A.. Calibration and regression methods. 116. Prediction and confidence intervals. 1998. “Basic statistical methods for analytical chemistry: Part 2. 993–1014. Miller. 10.” Pure Appl. 3–14. 18. Chem. P.© 2009 by Taylor & Francis Group. LLC . and Currie. Bonate. 1992.. “Concepts in calibration theory: Part IV..L..Uncertainty 75 17.. L.

sufficiently homogeneous and stable with reference to specified properties. accompanied by documentation issued by an authoritative body and providing one or more specified property values with associated uncertainties and traceabilities. Stability — ability of a reference material. and a detailed classification of RMs is presented in Table 5. which has been established as fit for its intended use in measurement or in examination of nominal properties. 2] Reference material — material. RMs may be divided into pure substances. RM is said to homogeneous with respect to a specified property if the property value.5 Reference Materials 5. to maintain a stated property value within specified limits for a specified period. is found to lie within the specified uncertainty limits. packages.1 [8]. Homogeneity — condition of having a uniform structure or composition with respect to one or more specified properties. those that have a high and strictly defined level of purity. and standard solutions.© 2009 by Taylor & Francis Group.1 [9]. when stored under specified conditions. where RMs are used to determine precision and accuracy • Interlaboratory comparisons. 77 . where they are applied as subject matter for studies • Estimating the uncertainty of a measurement • Documenting traceability With regard to the function that is used in a measurement process. The range of their application is very wide [3–7]: • Validation of analytical procedures. LLC .) — between-bottle homogeneity. etc. using valid procedures. as determined by tests on samples of specified size. the samples being taken either from different supply units (bottles. 5.1 Definitions [1.2  Introduction Reference materials (RMs) play a significant role in all the elements of a quality assurance system that evaluates the reliability of measurement results. Certified reference material — reference material. The general classification of RMs is presented in Figure 5. or from a single supply unit — within-bottle homogeneity.

Standard solutions FIGURE 5. SecRMs NONCERTIFIED Primary reference materials. TABLE 5.1 Classification of Reference Materials Suitable for Chemical Investigations [9] Parameter Property Chemical composition Additional Remarks Reference materials (RMs). PRM MATRIX-FREE .78 Quality Assurance and Quality Control REFERENCE MATERIALS CERTIFIED MATRIX . LRMs .Quality control materials.. melting point. either natural or with added analytes (e.Pure substances . e.. viscosity.Secondary reference materials. or compound) stochiometrically and isotopically certified in amount of substance ratios with total impurities <10 µmol/mol Primary chemicals As above.Laboratory reference materials. characterized for one or more chemical or physicochemical property values Materials characterized for one or more biochemical or clinical property values Materials characterized for one or more physical property values. Other subcategories can be added at any time to address the needs of applicants seeking recognition of competence in producing types of RMs not currently listed High purity Pure specific entity (isotope. but with limits of <100 µmol/mol Defined purity As above.g.1  Classification of reference materials [8]. being either pure chemical compounds or representative sample matrices. or surface characteristics) These principal categories are subdivided into subcategories as indicated in the following draft list..g. element.g. density Materials characterized for one or more engineering property values (e. tensile strength. but with limits of <50 µmol/mol Biological and clinical properties Physical properties Engineering properties Miscellaneous Chemical nature Single major constituent . hardness.© 2009 by Taylor & Francis Group. animal fats spiked with pesticides for residue analysis). LLC . QCMs .

LLC . etc. labels. I. appropriate fraction grain size) Initial examination of the material’s homogeneity Determination of main components Putting the materials into containers Final examination of the material’s homogeneity Disinfection of the material (ensuring its biological stability) Determining humidity Organization of an interlaboratory comparison.Reference Materials Parameter Matrix types Additional Remarks Major constituents 79 Traceability 0 Primary class I class II class III class IV class V class Uncertainty of With uncertainty determination value of analyte Without concentration uncertainty value Field of application Major constituents (in matrix) >100 mmol/kg or >100 mmol/dm3 Minor constituents Minor constituents (in matrix) <100 mmol/kg or <100 mmol/dm3 Trace constituents Trace constituents <100 µmol/kg or <100 µmol/dm3 Ultra trace Ultra trace constituents <100 nmol/kg constituents or <100 nmol/dm3 Pure specified entity certified to SI at the smallest achievable uncertainty Certified by measurement against class 0 RM or SI with defined uncertainty (no measurable dependence on matrix) Verified by measurement against class I or 0 RM with defined uncertainty Described linkage to class 0. in order to carry out a certification process . II Described linkage other than to SI No described linkage Primary reference materials Certified reference materials Laboratory reference materials Quality control materials Validation of analytical method Establishing measurement traceability Calibrating an instrument Assessment of a measurement uncertainty Assessment of a measurement method Recovery studies Quality control Preparation of the RM involves the following: • • • • • • • • • • • Material selection Obtaining an appropriate amount of the material Selection and purchase of appropriate containers.© 2009 by Taylor & Francis Group. Initial material preparation (grinding. sifting.

A certification process involves preparation of a great number of homogeneous. and the uncertainty in determination of given parameters. RMs should be prepared in such a way that they are homogeneous. including the laboratory RM (cheaper and more available). Uncertified RMs. calculating means. and printing the attestation certificate A general procedure for preparing RMs is shown schematically in Figure 5. LLC . and its availability.2 [8].” The requirements at the production stage.3. 5.80 Quality Assurance and Quality Control • Statistical analysis of the obtained results (rejection of deviating results. are used mainly for the calibration of measuring instruments and checking analytical procedures [9. stable since the moment of production until their use) [9]. One should take into account microbiological degradation.© 2009 by Taylor & Francis Group. It is very important to pay special attention not only to the preparation of stable and homogeneous primary materials.3 Parameters That Characterize RMs 5. which are representative parts of a given production batch. which can be minimized by decreasing the content of water in the material to the level of 1–3% of relative humidity. stable. stable. and the confidence intervals) • Determination of values attested to on the basis of hitherto formulated criteria. but also to sampling [11]. which is reflected in their price and thus their availability. or ampoules). It may be achieved in two ways: by sending the same material sample or sending material samples with the same parameters (homogeneous.1 General Information Certification of RMs is something more than just performing a series of accurate and precise measurements traceable to SI standards or to any other metrological system. the type of analytical measurements in which it is going to be used. 10]. stable during storage. The selection of the RM depends on the needs at a given time. RM can perform its function only when each of its users receives a material with exactly the same parameters. and have constant characteristics over a sufficiently long period. The parameters that characterize CRMs [12–18] are as follows: • • • • representativeness homogeneity stability certified value . are more rigorous. standard deviations. and appropriately packaged samples. penicillin vials. It is also recommended to pack the RM samples into appropriate containers in the argon atmosphere (bottles with fillers. precision. No certified reference materials (laboratory reference material and material for quality control) and certified reference materials (primary reference material and certified reference material) differ in accuracy. according to ISO recommendations. That is why CRMs occupy a higher position in the “metrological hierarchy.

2  A general procedure for certified reference materials preparation — example for solid CRMs [8]. labels. . Preliminary evaluation of homogeneity Content determination of main components Distribution of material into containers Final homogeneity determination (inside one container and among containers) Sterilization of material (securing biological stability) Water content determination Organizing interlaboratory comparison conducting certification procedure Statistical analysis of obtained results Determining certified values Printing of certificate CERTIFIED REFERENCE MATERIAL Continuation of studies on long-term stability 81 FIGURE 5.g. LLC . sieving) Preliminary determination of material’s stability Choosing and purchasing of proper containers.Reference Materials Choosing material Obtaining appropriate amount of desired material Preliminary preparation of material (e.. breaking down. etc. drying.© 2009 by Taylor & Francis Group.

A material should be homogeneous and stable. This value should be determined by the manufacturer of the RM and taken into account in the uncertainty budget of the certified value. but in the process of homogenization and stabilization a change may occur in the connection between the analyte and the matrix. the RM is exposed to the influence of various external factors (temperature. light. microbiological activity) that may affect its composition [16]. the manner of processing.3 Homogeneity Homogeneity study is a comparison of the obtained results for the random samples of the RM. next to the homogeneity study.© 2009 by Taylor & Francis Group. However.4 Stability A stability study.3. A user has no influence on between-bottle heterogeneity of the material. In such cases.2 Representativeness Representativeness is a property that describes a similarity between individual samples with regard to: • • • • • Matrix composition Analyte concentration Manner of the connection between the analytes and the matrix Type and concentration of interfering substances Physical state of the material For practical reasons. .82 Quality Assurance and Quality Control 5.3.3. Both sources of heterogeneity of reference materials are presented in Figure 5. the user should be informed about the actual state of the material. LLC . There are two types of homogeneity [13]: • Within-bottle homogeneity • Between-bottle homogeneity The influence of within-bottle heterogeneity of the material on the result of the certified value may be eliminated by sampling a greater amount of the material. and how to achieve a representative sample of the material for further analysis. 5.3. the value of a given parameter of the material should be stable over the whole validity period. achievement of the required similarity is not always possible. It is carried out at the stage of distributing the RM into the appropriate containers. oxygen. The stability of the RM is determined by using the analysis of the certified parameters in the samples of materials stored in a so-called reference temperature (with an assumption that in that temperature the composition of the RM does not change) in relation to samples stored in temperatures recommended for a given RM. humidity. During storage and transportation. That is why it is necessary to define the minimum amount (mass) of the RM samples for the study. plays a decisive role in the production of RMs. 5.

. stability during transportation) Stability studies require the application of fast measurement methods.g. for example. shelf life) • Short-term stability (e.3  Sources of reference materials heterogeneity: (A) within-bottle. usually stored in a lower temperature.Reference Materials 83 A B FIGURE 5. LLC ... (B) betweenbottle. and the high repeatability of the measurements. Studying the stability of RMs may be considered in two aspects: • classical (long term) • isochronous In case of the classical stability study. There are two types of RM stability: [14–16] • Long-term stability (e. −40°C. low-mass samples.g. The studies are carried out for various temperatures and storage durations. stability is determined by comparing the results obtained for samples stored in the recommended conditions and for the reference samples.© 2009 by Taylor & Francis Group.

according to the guidelines presented by Guide to the Expression of Uncertainty in Measurement [25]. Thus the following solutions are applied [23]: • Measurements at a single laboratory. as described in an appropriate ISO norm [19–22]. Certification is based on material sample analyses. The aim of material certification is to ascribe certain values of individual properties to a group or individual units. 5. matrix RMs cannot be certified using direct gravimetric measurement.1) where ucert = uncertainty of determining the certified value ubott = uncertainty associated with the within-bottle homogeneity uls = uncertainty associated with the long-term stability uss = uncertainty associated with the short-term stability . essential for certification [24]. An isochronous stability study is based on deducing the stability of the RM on the basis of analyses of samples stored over a short period (several weeks) and at various temperatures (usually higher than the recommended storage temperature) [16].84 Quality Assurance and Quality Control Such studies are carried out a short time before the hitherto determined expiry date and may result in extending the validity period. in which each of the measurement series is carried out with the highest accuracy and traceability. by two or more analysts • Interlaboratory studies using one or several various methods. In this case. should include all the uncertainty sources described in the following equation [26]: 2 2 2 2 uCRM = ucert + ubott + uls + uss (5. an additional stage is required: a complete change or the removal of the matrix. that is. The reliability of the obtained results of analytical measurements is a self-evident condition.3. using the absolute methods.© 2009 by Taylor & Francis Group.5  Certified Value RM certification is carried out according to strictly determined rules. and must be documented by a complete uncertainty budget. LLC . The final uncertainty value of the CRM. including the absolute methods It must be remembered that certification studies should be carried out by the laboratories with supreme and proven competence. using one or more methods at one or several laboratories. In contrast to pure substances and calibration solutions. methods that give the results directly in units of measurement or methods that allow the result to be expressed in those units through the application of mathematical equations from the appropriate physical and chemical theories • Measurements at a single laboratory using two or more methods.

. therefore.comar. concerning the minimum mass of the RM sampled.4 Practical Application of CRM These are the main issues associated with the application of the RMs [27–30]: • Determination of validation parameters — first of all their precision and accuracy • Examining the skills of an analyst or a laboratory • Routine control of precision and accuracy of the performed determinations • Laboratory accreditation • The quality control of performance of a given laboratory • Estimating measurement uncertainty • Monitoring and ensuring traceability • Calibration of measuring instruments It is not possible to prepare appropriate RMs for all the analytical tasks. for example. and the manner of storage. the validity period.net Using RMs requires compliance with the rules of good laboratory practice at laboratories that determine the trace components in the examined samples: • It is necessary to comply with the recommendations of the RM manufacturer.htm http://www.© 2009 by Taylor & Francis Group. a key to the right choice. can be found in the following databases available at the Internet websites: http://www-naweb. Selection of the RM should allow for the following criteria: • • • • • • Availability (the issue of the matrix composition) Concentration range of the reference value Uncertainty value of the reference value Traceability of the reference value Required uncertainty value of the measurement Influence of the CRM uncertainty on the combined uncertainty of the measurement • Quality of the CRM manufacturer (competence. and help in finding an appropriate RM.de http://www.iaea.bam. A good knowledge of analytical procedures and the available materials is.org/nahu/nmrm/nmrm2003/browse. LLC .Reference Materials 85 5. due to the high heterogeneity of matrix compositions and the wide spectrum of analytes present in the examined samples.virm. reputation) • Composition of the sample matrix • Price Detailed information concerning the RMs.

RMs are an essential tool for the determination of accuracy and/or precision. LLC .2. It seems practical to provide a graphical comparison of the reference (certified) value with the value obtained during the measurement (determined one). TABLE 5. together with the associated conclusions are presented in Table 5. depending on the information on the two compared values. Possible situations.86 Quality Assurance and Quality Control • It is necessary to determine the concentration of water (in case of solid materials) for the RM samples taken simultaneously with the RM sample for the study. Because one of the main problems associated with this process is the interpretation and numerical presentation of the determined parameter.© 2009 by Taylor & Francis Group.2 A Suitable Way of Graphically Comparing the Reference (Certified) Value with the Determined Value Conditions Reference value without providing the uncertainty (not a certified value) Graphical presentation Conclusions Determined value agreed with the reference value Reference value Determined value Conclusion impossible Reference value Reference value with the uncertainty Determined value Determined value agreed with the reference value Certified value Determined value . this book presents the basic formulas and correlations that help in selecting the manner of documenting the values of the determined parameters. • The used and nonused RM cannot be placed back into the container.

Data: result series. The certified value given by the manufacturer is 4.64 ± 0.94 5. LLC .57 4. Using a graphical method.1 Problem: Five independent determinations of total mercury were carried out for the samples of the certified reference material NRCC-DORM-2 — dogfish muscle. μg/g: 1 2 3 4 5 4.Reference Materials Conditions Determined value without a provided uncertainty Graphical presentation Conclusions Conclusion impossible 87 Certified value Reference value with the uncertainty Determined value Determined value agreed with the certified value Certified value Determined value with a provided uncertainty Determined value Determined value not agreed with the certified value Certified value Determined value Example 5.28 μg/g.76 4.82 .© 2009 by Taylor & Francis Group. test the agreement of the obtained value with the certified value.04 4.

34 0.023 μg/g 0.35 0.18 μg/g 0.40 . LLC .37 0.16 μg/g Determined Conclusion: An obtained value agreed with the certified one.372 μg/g 0.019 μg/g 0. Data: result series.© 2009 by Taylor & Francis Group.83 μg/g 0.xls Example 5. Excel file: exampl_RM01. μg/g: 1 2 3 4 5 6 Solution: Mean SD U (k = 2) 0.36 µg/g Using a graphical method. The assigned value given by the manufacturer is 0.39 0.38 0. test the agreement of the obtained value with the assigned value.88 Solution: Mean SD U (k = 2) Graph: 6 5 4 3 2 1 0 CRM Quality Assurance and Quality Control 4.2 Problem: Six independent determinations of total mercury were carried out for the samples of the reference material GBW 07601 — powdered human hair.

05 0 89 RM Determined Conclusion: An assigned value is in the range of obtained value ± uncertainty.1 0. A comparison of the standard deviation values in the series of measurements for CRM. The following condition must be fulfilled: where SDdet = standard deviation for the measurement series for CRM n = number of measurements for CRM UCRM = expanded uncertainty for CRM and xCRM − UCRM < x det < xCRM + UCRM where xdet = determined value xCRM = certified value (5.15 0.4 0.xls An alternative solution is to determine the conformity of the reference value with the determined value using appropriate tests. Excel file: exampl_RM02.2) .3 0. LLC . with the value of expanded uncertainty for CRM.25 0.3) SDdet n < UCRM (5. The following options are feasible: 1.Reference Materials Graph: 0. and the comparison of the determined values with the certified value.© 2009 by Taylor & Francis Group.45 0.35 0.2 0.

2 71.76 4. LLC . Using the aforementioned method. Data: result series.18 µg/g 5 0.xls Example 5.36 µg/g 4.90 Quality Assurance and Quality Control Example 5.92 µg/g 4. μg/g: 1 2 3 4 5 Solution: xdet SDdet n SDdet n UCRM xCRM xCRM − UCRM xCRM + UCRM 4. test the agreement of the obtained value with the certified value. Data: result series.83 µg/g 0.4 µg/g.3 Problem: Five independent determinations of total mercury were carried out for the samples of the certified reference material NRCC-DORM-2 — dogfish muscle. test the agreement of the obtained value with the certified value.28 µg/g 4. The certified value given by the manufacturer is 4.82 SDdet n < UCRM xCRM − UCRM < xdet < xCRM + UCRM Conclusion: An obtained value agreed with the certified one.080 µg/g 0.64 ± 0. µg/g: 1 2 3 4 70.04 4.64 µg/g 4.6 .8 70.57 4.© 2009 by Taylor & Francis Group. The certified value given by the manufacturer is 68. Excel file: exampl_RM03.94 5.4 69.4 Problem: Four independent determinations of lead were carried out for the samples of the certified reference material NIST-SRM 1633b — coal fly ash. Using the aforementioned method.2 ± 1.28 μg/g.

4) The calculated value should be compared with the critical value from the distribution values for an appropriate significance level (α) and the number of degrees of freedom f = n − 1. Comparison of the certified value with the determined value.68 µg/g 4 0. Formula (5. The following correlations are examined: x det − xCRM < 2 u(2xdet ) + u(2xCRM ) (5.4 µg/g 68.6) t= x det − xCRM u(2xdet ) + u(2xCRM ) n (5.8 µg/g 69. LLC . Application of Student’s t test.4) does not allow for the uncertainty of the certified value. using the uncertainty values for both the values.© 2009 by Taylor & Francis Group.5) . that is why it is recommended to use its modified version: where u( xdet ) = combined uncertainty of the determined value u( xCRM ) = combined uncertainty of the certified value 3.342 µg/g 1. Excel file: exampl_RM04. The value of parameter t is calculated according to the formula: t= x det − xCRM SDdet n (5.Reference Materials Solution: xdet SDdet n SDdet n UCRM xCRM xCRM − UCRM xCRM + UCRM 70.5 µg/g 0.2 µg/g 66.6 µg/g 91 SDdet n < UCRM xCRM − UCRM < xdet < xCRM + UCRM Conclusion: An obtained value did not agree with the certified one.xls 2.

LLC . Data: result series.6 Problem: Four independent determinations of lead were carried out for the samples of the certified reference material NIST-SRM 1633b — coal fly ash.2 ± 1.323 Conclusion: An obtained value agreed with the certified one. Example 5. test the agreement of the obtained value with the certified value.92 Quality Assurance and Quality Control x det − xCRM ≥ 2 u(2xdet ) + u(2xCRM ) (5.© 2009 by Taylor & Francis Group.82 4.7) Satisfying the first relation implies conformity of the determined value with the certified value.080 0.04 4.28 μg/g. .76 4.4 μg/g. The certified value given by the manufacturer is 68.57 4. μg/g: 1 2 3 4 5 Solution: xdet SDdet n u( xdet ) u( xCRM ) xdet − xCRM 2 u 2 ( xdet ) 4. Using the aforementioned method.186 2 ( xCRM ) xdet − xCRM < 2 u(2xdet ) + u(2xCRM ) xdet − xCRM ≥ 2 u(2xdet ) + u(2xCRM ) +u 0. The certified value given by the manufacturer is 4. and satisfying the second relation denotes the lack of conformity between these values.5 Problem: Five independent determinations of total mercury were carried out for the samples of the certified reference material NRCC-DORM-2 — dogfish muscle.94 5.xls Example 5.64 ± 0. Excel file: exampl_RM05.18 5 0.83 0.140 0.

xls 4.700 2. then the determined value agreed with the reference value. μg/g: 1 2 3 4 Solution: xdet SDdet n u( xdet ) u( xCRM ) xdet − xCRM 2 u(2xdet ) + u(2xCRM ) 70.Reference Materials Using the aforementioned method. The reasoning is carried out using the following relations: • If Z ≤ 2. LLC .9) . which can be calculated as the combined uncertainty of the certified value and the determined value. The value of the Z score is calculated using the following formula: Z= x det − xCRM s (5. due to application of CRMs. • If Z > 2.8) where s is the value of a deviation unit.30 1.© 2009 by Taylor & Francis Group. then the determined value did not agree with the reference value.5 0. Data: result series.6 93 xdet − xCRM < 2 u(2xdet ) + u(2xCRM ) xdet − xCRM ≥ 2 u(2xdet ) + u(2xCRM ) Conclusion: An obtained value did not agree with the certified one. test the agreement of the obtained value with the certified value.342 0.68 4 0.2 71. can be presented as recovery and should be calculated according to the following equations: R= x det ⋅ 100% xCRM (5. The application of Z score.4 69. Trueness value. Excel file: exampl_RM06.56 70.8 70.

28 μg/g. the calculated value of trueness is acceptable. Example 5.94 Quality Assurance and Quality Control U = k⋅ (u 2 ( xdet ) + u(2xCRM )  x det + xCRM    2   ) (5. Using the obtained result. LLC . calculate trueness as a recovery value for k = 2.140 2 104.83 4.82 Conclusion: A value of 100% is in the range of calculated trueness value.xls .080 0.57 4. μg/g: 1 2 3 4 5 Solution: xdet xCRM SDdet n u( xdet ) u( xCRM ) k R U 4. The value of trueness is usually given as: Trueness = R ± U (5.94 5. Data: result series.04 4.7 Problem: Five independent determinations of total mercury were carried out for the samples of the certified reference material NRCC-DORM-2 — dogfish muscle.10) The reasoning should be based on the following: if the range R ± U includes value 100%.18 5 0.8% 4.64 0.11) and is most frequently expressed in %. Excel file: exampl_RM07. The certified value given by the manufacturer is 4.0% 6.64 ± 0.© 2009 by Taylor & Francis Group.76 4.

123.. 5. LLC . .. “A need for clearer terminology and guidance in the role of reference materials in method development and validation.” Chemosphere. A.. H. Konieczka. 2307–2315. 2004. and Namieśnik. 2004. Due to financial limitation.. P.” Accred. nor in interlaboratory comparisons. Rev. “Reference materials — an industry perspective.5 Conclusion Production and certification of RM is very costly. Qual.” Accred. “Reference materials: terminology and use.Reference Materials 95 Due to a limited number of CRMs. 8. 2007. and ensuring the quality of analytical measurement results. 23(6). 11. “Homogeneity and stability of reference materials. and do not nullify the remaining elements of the quality system. 108–122. CRMs play a crucial role in the system of estimation. Majcen. 20–25 (2001). 5. and Lamberty. LGC/VAM. 1998. J. T. 10. H. and Otson. Chem. 27. 1996.. However. however. eds... RMs should be stored in conditions that guarantee the stability of their composition over the whole period of use. Van der Veen. 360. Schimmel. Fellin.D. 6. “A test atmosphere generation system for particle-bound PNA: development and use for evaluation of air sampling methods. 370. Pauwels. J. 2007 (in Polish). “The role of and place of method validation in the quality assurance and quality control (QA/QC) system..N.P. it is not recommended to use CRMs for a routine intralaboratory statistical control.. Konieczka. in competence tests. Can’t one see the forest for the trees?” Trends Anal. Assur. It is recommended. Anal. Assur.. and Lamberty. 6. as noted above.. JCGM 200. G. Warszawa. RMs must be applied in a rational way.” Mikrochim. “Reference materials in the world of tomorrow..M. 87–93.H. 2.© 2009 by Taylor & Francis Group. Chem. 111–114.. “CRMs for the 21st century: new demands and challenges. 8. Acta. References 1. Rasberry. Anal. P. 37. 2003.. 539–542. Chem.. Qual. is necessary in any laboratory. A.” Accred. monitoring.. Lipp.P.. Kontrola i zapewnienie jakości wyników pomiarów analitycznych. Assur. 3. 4. 442–449. Joint Committee for Guides in Metrology. which is why application of CRMs is usually limited to the verification of analytical procedures and only in some exceptional case to calibration (in comparative methods). Chem... N.. 7.. a widely known standard addition method is applied as an alternative means of determining trueness. A. 12. Pauwels. 277–281. WNT. J. “The preparation of biological and environmental reference materials.” Crit. International vocabulary of metrology — Basic and general concepts and associated terms (VIM). and Gawlik. M. it must be said that using CRM in a laboratory does not automatically ensure the obtainment of reliable results. R. 9. T. Kramer.J.” Fresenius J. 1993. Emons. Qual. P.. J. and Pauwels.. Linsinger...... S. Guidelines for the in-house production of reference materials. Anal.” Fresenius J. version 2. Linsinger. 9. Their application. 2001. 1998. 2008.M. 173–190.J. B.

.. 359–361.. T. 95–99. “The role of reference materials and reference methods in chemical analysis. A.. 2000. A. and Pauwels. and Schimmel. Stability study. 361–411. ISO Guide 31. G.H. and Pauwels. 6. J. J. 136–143.. 26.. Qual. Chem... H. S.” Accred.. T. 368.” Accred. Van der Veen. 290–294. Pauwels. Uriano.. “Some difficult problems still existing in the preparation and certification of CRMs. Reference materials — Contents of certificates and labels... 257–263. Pb. ISO.... J.. Qual. Characterisation and certification. Qual. Linsinger. A. “Uncertainty calculations in the certification of reference materials. Schumann.. Linsinger. Guide to the Expression of Uncertainty in Measurement (GUM). A. H.P... 14. Chem. and Iamiceli.” Fresenius J. Van der Veen. G. Geneva. and Gravatt. J. Qual. Chem. 2000. and Pauwels.P. 1999.” Fresenius J. 2001.1Lamberty. Geneva.H..H. Lamberty. G. 1977.. A. 6.. Chem.. ISO. A.” Microchem. and Zn in certified reference materials using the optimized BCR sequential extraction procedure..H.” Fresenius J. L.A. J. Schimmel. Anal. 361.M.J. ISO Guide 30..” Accred.H. 6. and Kramer. and Tack. H. 1998. Lamberty. Van der Veen. Sutherland..A. A. 1996.J.. J. O. A. Linsinger.. A. Geneva.. S. T. P. 2. Van der Veen. Schimmel. 4..L.. 2000. “Uncertainty calculations in the certification of reference materials. 30.© 2009 by Taylor & Francis Group. A. Geneva. Senofonte. 2002.” Fresenius J. C. 15. 2000. Dybczyński. 18. 24. 21. 23. J.. 26–30.” Crit. 464– 469. 589. 1998.” Accred.. Chem. 244–250.M. Chim. Anal. F.” Microchem. Lamberty.. 2001. 1993. Certification of reference materials. Fe.. 62.. Anal.. Robouch. “The study of the stability of reference materials by isochronous measurements. 1989. Caroli. A. ISO Guide 34. 6.C. J. 19.” Accred. 2001. 29. A. 1992.. 370. 25. J. 20. Veen. Homogeneity study. Forte. Caimi. Pauwels. “Certified reference materials for research in Antarctica: the case of marine sediment. S. Assur. Assur... Quality system guidelines for the production of reference materials.G. Linsinger. and Pauwels. 360. 16. 59. “Determination of Al. .. J. 1. Rev..M. Pauwels. Cu.. 28.. Principles of analysis of variance.” Anal. H. ISO Guide 35. Trends and definitions used in connections with reference materials. Anal. Acta. Schimmel.. R.N. 22. “Estimation of the CRMs in accordance with GUM: Application to the certification of four enzyme CRMs.P. Mn. and Schimmel. 249–257. Assur. a new candidate certified reference material... Lamberty. and Siekmann... 5.. 1998. 2001.M.. 126–130. General and statistical principles. 454.. and Polkowska-Motrenko. G. 395–399..M. H. Lamberty.M. Anal. “Uncertainty calculations in the certification of reference materials. Assur... Danko. “ICP-AES and ICP-MS quantification of trace elements in the marine macroalga Fucus sample. Assur. “Uncertainty calculations in the certification of reference materials. ISO. Van der Veen. T. LLC . Pauwels.. “Evaluation of uncertainty of reference materials. J.96 Quality Assurance and Quality Control 13. H. 5.. Qual. and Pauwels. R. Geneva. A... 3. B.J. 27. 17. Caroli. “Quantification of the expected shelf-life of certified reference materials.

1 [5]. One of the most crucial means of that monitoring is participation in various interlaboratory studies [3].. Participation in these programs gives a laboratory a chance to compare its results with those obtained by other laboratories and to prove its competence. and evaluation of tests on the same or similar test items by two or more laboratories in accordance with predetermined conditions.2  Introduction Demand for results as a source of reliable analytical information poses new challenges for analytical laboratories: they need to be especially careful in documenting the results and the applied research methods. 6. Moreover. A generalized scheme for conducting interlaboratory studies is shown in Figure 6. 97 . usually with a determined uncertainty. and in the case of error detection. which can be especially significant for laboratories with accreditation or those applying for accreditation.g.1 Definitions [1. LLC . Method performance study — interlaboratory research in which all participants act according to the same protocol and use the same test procedures to determine the characteristic features in a batch of identical test samples. performance. 2] Interlaboratory comparisons — organization.6 Interlaboratory Comparisons 6. undertake rectifying action [4]. The way to realize this goal is to implement a suitable quality assurance system at a laboratory through constant monitoring of the reliability of the analytical results and calibration.© 2009 by Taylor & Francis Group. analyte concentration) in a tested material or a given sample. Ensuring a suitable quality of analytical results is essential because of the negative implications of presenting unreliable measurement results. participation in analytical interlaboratory comparative studies gives a laboratory a chance to search and detect unexpected errors using comparison with external standards and its own previous results. Certification study — a study that assigns a reference value to a given parameter (e. Proficiency testing — determination of laboratory testing performance by means of interlaboratory comparisons.

6.1  A generalized outline for conducting interlaboratory studies [5].3 Classification of Interlaboratory Studies Interlaboratory studies are organized to: • • • • • • Assess the reliability of measurement results Gain experience Increase the quality of conducted analytical determinations Create possibilities for proving the competence of a given laboratory Better understand the applied procedures Determine validation parameters .98 Quality Assurance and Quality Control PROJECT Defining an aim Choosing an organizer Choosing a sample Selecting participants Choosing analysis/study type IMPLEMENTATION Sample preparation Sending samples to participants Analysis of samples Sending the analysis results EVALUATION Analysis of results Sending evaluation results to participants REPORT FIGURE 6. LLC .© 2009 by Taylor & Francis Group.

and the statistical analysis of the obtained sets of results is conducted separately for each of the procedures. with regard to the composition of the matrix. The obtained results are applied in estimating the characteristic parameters of the procedure: • • • • • • • Intra. Competence study is a research in which one or more analyses are carried out by a group of laboratories using one or more homogenous and stable test samples and using a selected or routinely used procedure by each of the laboratories participating in the interlaboratory comparison. analyte concentration. • The number of participants. This may include the following: • • • • Method performance study Competence study Certification study Proficiency testing Method performance study is an interlaboratory comparison in which all participants act according to the same protocol and use the same test procedures to determine the characteristic features (specified in the protocol) in a batch of identical test samples. This research may be conducted among laboratories that are . The obtained sample results are compared with the results obtained by other laboratories or with a known or determined (guaranteed) reference value. test samples. • By using the same materials or test samples. both on a national and international scale. Interlaboratory comparisons may also be classified according to the aims and range of studies. and the presence of interferents (research participants are usually informed about the composition of the matrix for the examined samples). it is necessary to conform to the following requirements: • The composition of the applied material or sample is usually similar to that of the materials or samples subjected to routine studies. Accredited laboratories are obliged to provide certificates of participation in such programs. and determinations as well as other details of the study are presented in the research protocol prepared by the organizer of the study.Interlaboratory Comparisons 99 Laboratories that wish to confirm their competence should participate in at least one program of interlaboratory research. all participating laboratories apply the same set of guidelines for each procedure. it is possible to compare a few procedures.© 2009 by Taylor & Francis Group. LLC .and interlaboratory precision Systematic error Recovery value Internal parameters of quality assurance Sensitivity Limit of detection Applicability limit In this type of research.

usually with a determined uncertainty. under which the participant remains anonymous to the rest of the group. Certification study is a study that assigns a reference value to a given parameter (e. usually subjected to analysis with regard to the matrix composition and the level of analyte concentration. The obtained results are compared with the previously determined guaranteed (reference) value. physical property) in a tested material or a given sample. using a procedure that ensures the estimation of the concentration (or any other parameter) with the smallest error and the lowest uncertainty value. Proficiency research is a tremendous challenge for laboratories that need to apply for accreditation based on the presentation of confirmation of their own competence. In the latter case. some problems with the stability and homogeneity of samples may occur due to the spread of the studies over a longer time. These studies are conducted to test the achievements and competence of both the individual analysts using a given analytical procedure or measurement. the applied analytical procedure may be a top-down decision or the organizer may limit the choice to a prepared list. It is a significant element in achieving and maintaining a suitable quality of results. which is why it is important to pay it a little more attention. and a specific analytical procedure. with regard to the mean level of analyte concentration and the homogeneity degree. analyte concentration. The choice of test material should be influenced by the maximum degree of similarity of the composition of samples. Proficiency testing may be conducted on the basis of the same material analysis. sample of the material being provided to all the participants at the same time for a simultaneous study or a round-robin test.© 2009 by Taylor & Francis Group. Such a material must be tested before it is distributed to the participants. This research is usually carried out by laboratories with a confirmed competence (reference laboratories) to test the material.. Proficiency testing may be conducted as an open (public) or a closed study (not public). the participants do not know that these are proficiency studies and that the obtained samples are to be analyzed in a routine fashion [6].g. the competence of participating laboratories is verified based on the determination of results of specified components in distributed samples (materials). Each laboratory is assigned an identification number. LLC . There are six ways to enable the determination of the reference value [7]: • • • • • Measurement by a reference laboratory Certified value for CRM used as a test material Direct comparison of the proficiency test (PT) test material with CRM Consensus value from expert laboratories Formulation value assignment on the basis of proportions used in a solution or other mixture of ingredients with known analyte contents • Consensus value from participating laboratories . Proficiency testing is the most frequently conducted type of interlaboratory research. In proficiency testing.100 Quality Assurance and Quality Control accredited or applying for accreditation in order to control the quality of determinations and the proficiency of researchers. In this case. which is a candidate for the reference material. In the case of closed research.

It gives laboratories a chance to improve their competence. Moreover. It generally takes a long time before the participants get to know the obtained results. each of the aforementioned types may be divided into three further categories: • Samples circulate successively from one laboratory to another. LLC . and improve their performance in the next proficiency test. the participants try to find the causes of such discrepancies. to check if the sample has not changed in an undesirable fashion. Moreover. produces a precise localization of sources and causes of errors and hence an improvement in the quality of analytical results. correct the hitherto existing mistakes. interlaboratory comparisons are retrospective studies. In this case. In reality. There are certain limitations associated with performance and participation in proficiency testing. proficiency testing accounts for only a small percentage of analyses conducted by laboratories and therefore does not reflect the full picture of routinely performed studies. After the initial research. 6.© 2009 by Taylor & Francis Group. the so-called key comparisons. proficiency testing is unusually time consuming. which is why proficiency testing may not affect any decision on quality management. it is necessary to check the work of individual laboratories because it gives them a chance to estimate the reliability of the analytical results of a given research team. a thorough analysis of an analytical process. In the case of results distinctly deviating from the assumed range of acceptable results. . there are two main types of proficiency studies: • Those examining the competence of the group of laboratories using the results from specifically defined types of analyses. • Those examining the competence of laboratories during the performance of various types of analyses. pilot studies are implemented to select the participants with suitable qualifications to participate in the actual proficiency studies.4 Characteristics and Organization of Interlaboratory Comparisons As one can gather from this current discussion. Taking into consideration the sample preparation used by participating laboratories. with the cooperation of a control center. all participants gather to discuss the obtained results. • Product or material samples are divided into several parts and each participant receives one part of each sample (this type is called the split sample study).Interlaboratory Comparisons 101 Sometimes. • Subsamples randomly selected from a large batch of homogeneous material or test samples are simultaneously distributed to participating laboratories (the most popular type of proficiency testing). With regard to conditions. The achievement of these aims requires a painstaking and reliable organization of this research. First of all. a sample may be taken back to the coordinating laboratory before a test by a subsequent participant.

6. Diagrams of this type are usually presented in final reports by the organizers of interlaboratory comparisons and proficiency tests.© 2009 by Taylor & Francis Group. Data: results: Data lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 123 111 128 138 121 123 188 114 u 11 9. assigning each result a code corresponding to the code number of the laboratory. Detailed information on the characteristics. in the case of comparative methods. LLC . laboratory codes are marked. Due to economic reasons in interlaboratory comparisons. Example 6. Their production and certification is usually very expensive. homogeneity. therefore the use of certified reference materials (CRM) should be limited to the verification of analytical procedures and. a graph may be constructed where the results are marked from the lowest to the highest. make a diagram showing the distribution of individual determination results. the individual results obtained by the laboratories and the uncertain values.8 14 16 10 11 14 18 .102 Quality Assurance and Quality Control Reference materials are a necessary tool to conduct interlaboratory comparisons.1 Problem: For a given series of measurement results obtained by various laboratories and a given reference value and its uncertainty. and stability over a sufficiently long time. They are also a precious source of information for a potential customer or the accreditation office. it should be limited to the calibration of the control and measuring instruments. On the X axis. production. the general mean (or reference value) is marked along with the determined uncertainty value. To this end. and implementation of the reference materials is presented in Chapter 5. On the Y axis. All reference materials should fulfill basic requirements with regard to similarity. The diagrams make it possible for participants to see how their results relate to the results provided by other participants.5 Presentation of Interlaboratory Comparison Results: Statistical Analysis in Interlaboratory Comparisons The first stage of interlaboratory research result processing is the graphical presentation of results [8–11]. and (optionally) the number of performed independent determinations. and/or the applied procedures. one may effectively use laboratory reference materials (LRM).

The ultimate aim of all types of studies is to determine. and the selection of suitable tests and solutions depend on the type of research.xls The manner of conducting a statistical analysis of results obtained in interlaboratory comparisons. based on experimentally obtained numerical data.Interlaboratory Comparisons lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 Solution: xref uref 250 230 210 190 xLab 170 150 103 188 122 121 142 125 132 129 121 198 131 158 193 122 111 23 15 11 13 12 17 19 21 28 14 18 13 14 17 140 11 130 110 90 70 Lab 2 Lab 22 Lab 8 Lab 5 Lab 11 Lab 16 Lab 10 Lab 21 Lab 1 Lab 6 Lab 13 Lab 3 Lab 15 Lab 18 Lab 14 Lab 4 Lab12 Lab 19 Lab 7 Lab 9 Lab 20 Lab 17 Lab Code 50 Excel file: exampl_PT01. Respective documents define the precise manner of conduct for a specified type of research. the accuracy (and/or precision) of . LLC .© 2009 by Taylor & Francis Group.

or lastly may question the suitability of the selected measurement procedure. The next step in statistical analysis is to eliminate any deviating results. A large number of doubtful and/or deviating values (outliers) may suggest a significant discrepancy of the variance values or significant differences in the competence between individual laboratories participating in the project. using the Hampel test. 12]. At the initial processing of data provided by the participants of interlaboratory comparisons. for example. the distribution type is examined. one may draw conclusions on the applied procedure and on the characteristics of the analyst. compare various procedures. The normality of the distribution may be examined using.18). The accuracy of a given measurement procedure may be determined by comparing the assumed reference value with the mean value of results obtained using the said procedure. Depending on the type of measurements and the requirements for the results. for example.© 2009 by Taylor & Francis Group. One checks if the occurrence of doubtful or deviating values may be explained by technical errors. when it is a certification study. there are two useful methods of describing precision: repeatability and reproducibility of results obtained using the specified analytical procedures.2 Problem: Find outliers in a given series of measurement results obtained by various laboratories. Elimination of outliers is especially crucial in a situation where the material used in the interlaboratory research is a material for which the reference value is determined based on the results of the very research.8. or the Hampel test. analyzed. To this end. On this basis. The choice of a suitable test is conditioned by many factors. also called the Huber test [10. LLC . a Kolmogorov-Smirnov test (Section 1. one may use the statistical tests of Cochran and Grubbs [12]. Precision is associated with the conformity of the series of results. median. and conduct certification of the material or validation of a specified procedure. and compared various tests used for outlier rejection. There are many reports in which authors critically examined. In recording the variability of the results obtained using a given procedure. Example 6. the first of which is the number of results. or the Winsorized mean (parameters presented and defined in Chapter 1). one may use the arithmetical mean. Data: results: Data lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 123 111 128 138 121 123 188 . or when the subject of the comparisons is not the reference material.104 Quality Assurance and Quality Control measurement procedures.

5 61.5 4.5 15.Interlaboratory Comparisons lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 Solution: |ri| lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 Excel file: exampl_PT02.5 66.5 4.5 5.5 15.5 3.5 15.5 71.5 61.© 2009 by Taylor & Francis Group.5 11.5 1.5 12.5 2.5 4.5 Data 123 111 128 138 121 123 188 114 188 122 121 142 125 132 129 121 198 131 158 193 122 111 Outlier or Not OK OK OK OK OK OK Outlier OK Outlier OK OK OK OK OK OK OK Outlier OK Outlier Outlier OK OK 114 188 122 121 142 125 132 129 121 198 131 158 193 122 111 105 .5 31.5 5.xls 3. LLC .5 1.5 5.5 5.

xls Example 6.4 11. Use the Cochran test to examine the intralaboratory variability.615 SD2 0. Excel file: exampl_PT03.6 12.82 0.7 Conclusion: Result obtained by laboratory “lab 6” is correct.43 13.463 0.443 12.106 Quality Assurance and Quality Control Example 6.3 Problem: Find outliers in the given sets of measurement results obtained in interlaboratory comparisons.093 0.070 0.70 11.9 11.1 12.65 0.3.05 C0.1 11.6 13.8 11.5 14.8 12.4 12.4 Problem: Find outliers in the given sets of results obtained in interlaboratory comparisons from Example 6.67 3 8 0.1 12.47 12.583 0.66 0.0 14.73 13.1 10.01 SD 0.© 2009 by Taylor & Francis Group.0 13.211 0.1 12.8 13.423 0.4 13.6 14.8 11.670 0.6 13.10 12.40 13.516 0. .33 11. LLC .5 13.31 0.430 0.1 11.76 0. Apply the Grubbs’ test for one outlier to examine the interlaboratory variability.26 0.68 0.53 n p C C0.1 12. Data: results: lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 Solution: Mean lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 12.

8 11.1 12.70 11.01 G0.4 13.688 min 2.881 1.0 13. This analysis serves to verify the hypothesis that the means in the groups are identical against the alternative hypothesis (at least two means are different).0 14.274 2.6 14.43 13.7 107 Conclusion: Result obtained by laboratory “lab 5” is correct.53 3 8 12.126 12.8 11.4 12.1 10.xls To simultaneously determine the standard deviation as the measures of repeatability and reproducibility.588 0.73 13.9 11.05 Mean 12.40 13. .1 11.Interlaboratory Comparisons Data: results: lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 Solution:   lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 n p xm s Gp min/max G0. LLC .6 13.33 11.4 11.8 12. Excel file: exampl_PT04.© 2009 by Taylor & Francis Group.1 12.5 13.6 12.5 14.10 12.1 12.47 12.1 12. When significant differences are found between the values of random errors (statistically significant differences in the variance values). The obtained numerical data are divided into m groups according to their origin (m is the number of laboratories). one may perform a one-factor (one-dimensional) variance analysis (ANOVA).1 11.8 13.6 13.

find which results are satisfactory. A total error. LLC 123 111 128 138 121 . Data: results: Data lab 1 lab 2 lab 3 lab 4 lab 5 .5 Problem: In the series of measurement results given in Example 6.8. with the identical value of the variance SD2. Use Z score. characterizing the results obtained by using an analytical procedure. Example 6. to a great extent. the value of the parameter t(α. The reliability of conclusions depends. rejecting the outliers. The essence of variance analysis is the division of the total variability. the reference values and the standard deviation sample are calculated according to all the results as the mean value and standard deviations after. of course. Situations in which a single factor completely explains a given phenomenon are rare. The manner of calculating this parameter has been described in detail in Chapter 1 (Section 1. f ) increases considerably and the precision of the evaluation decreases. The number of parallel determinations that is greater than 5 occurs only in special cases. the standard deviation being the measure of the respective variances. which are questionable. and which are unsatisfactory.15). Then one should determine the total intra.1. It shows that the interlaboratory studies should involve at least five laboratories. or when for some reason one expects deviation of the obtained measurement results from the normal distribution. the total sum of the squared deviations from all measurements from the mean. Below 4 degrees of freedom. The numerical value of the Z score parameter depends on the number and the type of data available to an analyst: • When only the mean values obtained from participating laboratories are known. consists of a few errors that are summed up according to the law of error propagation. The parameter that is most often used to evaluate the obtained results in interlaboratory comparisons is the Z score parameter. Draw a graph with Z score values for each laboratory.© 2009 by Taylor & Francis Group. and then the variance analysis is conducted for each group. by the sum of squares describing the variability within groups and the sum of squares describing the variability among groups.108 Quality Assurance and Quality Control the data are joined into groups for which the variance values are not statistically significantly different. on the number of laboratories participating in the research. The lower influence on the size of the certainty range is exerted by the number of parallel analyses conducted at a given laboratory.and intergroup degrees of freedom and calculate the standard deviation within individual groups and among the groups. An essential condition for conducting a correct interpretation of results for this analysis is the normal distribution of the population from which the samples were taken. that is.

40 8. LLC 109 123 188 114 188 122 121 142 125 132 129 121 198 131 158 193 122 111 Conclusion Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Unsatisfactory Satisfactory Unsatisfactory Satisfactory Satisfactory Questionable Satisfactory Satisfactory Satisfactory Satisfactory Unsatisfactory Satisfactory Unsatisfactory Unsatisfactory Satisfactory Satisfactory 124.51 −0.58 xm SD .08 0.97 8.Interlaboratory Comparisons lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 Solution: Z lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 −0.40 −0.78 3.16 7.61 −0.28 −0.43 1.5 .08 0.28 −1.55 −0.40 2.90 0.58 0.51 −1.69 0.16 −1.© 2009 by Taylor & Francis Group.10 −0.22 7.4 8.

LLC Lab 17 Lab 2 Lab 8 Lab 5 Lab 1 Lab 6 Lab 3 Lab 4 Lab 7 Lab 9 –4. Data: results: Data lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 123 111 128 138 121 123 188 114 188 122 .00 Z-score 4.00 .110 Graph: 10.© 2009 by Taylor & Francis Group.00 6.1.6 Problem: In the series of measurement results given in Example 6. Example 6. find for a given reference value which results are satisfactory. and which are unsatisfactory.00 –2. Use Z score. Draw a graph with Z score values for each of laboratory.00 0. which are questionable.00 8.00 2.00 Lab 22 Lab 11 Lab 16 Lab 10 Lab 21 Quality Assurance and Quality Control Lab 13 Lab 15 Lab 18 Lab 14 Lab12 Lab 19 Lab 20 Lab Code Excel file: exampl_PT05.xls • Known mean values obtained by participating laboratories and known reference value — the value of standard deviation is calculated according to the total set of measurement results — obviously after outliers are rejected.

LLC .42 −1.26 −2.94 −1.13 −2.01 −3.24 −2.01 5.85 −1.24 0.42 8.67 −2.13 −3.13 6.24 −1.24 6.67 −3.30 −2.24 −2.06 2.07 5.Interlaboratory Comparisons lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 xref Solution: SD Z lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 −2.42 −0.© 2009 by Taylor & Francis Group.77 −0.5 Conclusion Questionable Unsatisfactory Satisfactory Satisfactory Questionable Questionable Unsatisfactory Unsatisfactory Unsatisfactory Questionable Questionable Satisfactory Satisfactory Satisfactory Satisfactory Questionable Unsatisfactory Satisfactory Questionable Unsatisfactory Questionable Unsatisfactory 121 142 125 132 129 121 198 131 158 193 122 111 140 111 .

00 –4. or unsatisfactory.00 4.7 Problem: In the series of measurement results given in Example 6.1. for the given reference value and the combined uncertainty reference value.© 2009 by Taylor & Francis Group.00 Z-score 2. and the value of the reference combined uncertainty for a given material. Draw a graph with the Z score values for each laboratory.00 Quality Assurance and Quality Control Lab12 Lab 19 Lab 7 Lab 9 Lab 20 Lab 22 Lab 11 Lab 16 Lab 10 Lab 21 Lab 13 Lab 15 Lab 18 Lab 14 Lab Code Excel file: exampl_PT06.xls • Known mean values obtained by participating laboratories and known reference values. Data: results: Data lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 123 111 128 138 121 123 188 114 188 . questionable. find which of the results are satisfactory.112 Graph: 8. Use Z score. LLC Lab 17 Lab 2 Lab 8 Lab 5 Lab 1 Lab 6 Lab 3 Lab 4 .00 6. Example 6.00 0.00 –2.

00 −1.64 −2.55 −2.36 −0.Interlaboratory Comparisons lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 xref uref Solution: Z lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 −1.36 −1.09 −0.27 −0.18 −1.64 −1.73 −1.82 −1.36 4.82 1.64 4.64 −1.55 4.73 0.18 −1.73 −1.36 −2.© 2009 by Taylor & Francis Group.73 5.64 Conclusion Satisfactory Questionable Satisfactory Satisfactory Satisfactory Satisfactory Unsatisfactory Questionable Unsatisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Unsatisfactory Satisfactory Satisfactory Unsatisfactory Satisfactory Questionable 122 121 142 125 132 129 121 198 131 158 193 122 111 140 11 113 . LLC .

00 3.1.00 Quality Assurance and Quality Control Lab 6 Lab 13 Lab 3 Lab 15 Lab 18 Lab 14 Lab 4 Lab12 Lab 19 Lab 7 Lab 9 Lab 20 Lab Code Excel file: exampl_PT07. Data: results: Data lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 123 111 128 138 121 123 188 114 188 122 u 11 9.114 Graph: 6. LLC Lab 17 . Example 6.© 2009 by Taylor & Francis Group.00 1.00 Lab 2 Lab 22 Lab 8 Lab 5 Lab 11 Lab 16 Lab 10 Lab 21 Lab 1 –4.xls • Known mean values obtained in participating laboratories and known value of the reference combined uncertainty for a given material. use Z score.00 4.00 –3. taking into consideration the combined uncertainty reference value.8 14 16 10 11 14 18 23 15 .00 –1.00 5.00 Z-score 2. Draw a graph with the Z score values for each laboratory.8 Problem: In a series of measurement results given in the Example 6.00 –2.00 0.

93 −0.92 −0.97 −0.51 0.01 −1.70 −1.© 2009 by Taylor & Francis Group.09 2.23 1.43 Conclusion Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Questionable Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Unsatisfactory Satisfactory Satisfactory 121 142 125 132 129 121 198 131 158 193 122 111 140 11 11 13 12 17 19 21 28 14 18 13 14 17 115 .22 0.12 −0.50 −0.11 −1. LLC .Interlaboratory Comparisons lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 xref uref Solution: Z lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 −1.28 −1.85 3.10 −1.67 −0.80 1.09 −1.88 −0.40 −0.97 −1.

00 3.116 Graph: 4.00 Z-score 1.1) .00 2. assumed as a limit (permissible).00 –1. It is calculated in instances when participants of a given study use various methods to evaluate the obtained results. It is assumed that: • If ε ≤ x . the evaluation is satisfactory • If ε > x .xls Another parameter of the individual examination of measurement results is relative error.00 0. and therefore there is no ground to assume a common value of the sample. It is calculated using the formula: where ε = relative error (%) x lab = value of the result obtained by a given laboratory xref = reference value Evaluation of the obtained results is obvious in this case and depends on the range of analyte concentrations in a given sample.© 2009 by Taylor & Francis Group.00 –3.00 –2. LLC Lab 2 Lab 22 Lab 5 Lab 8 Lab 11 Lab 1 Lab 6 Lab 21 Lab 10 Lab 13 Lab 16 Lab 3 Lab 18 Lab 15 Lab 14 Lab 4 Lab12 Lab 19 Lab 9 Lab 17 Lab 7 Lab 20 Lab Code –4. ε= ( xlab − xref ) 100% xref (6.00 . the evaluation is not satisfactory where x denotes the relative systematic error (relative deviation).00 Quality Assurance and Quality Control Excel file: exampl_PT08.

7 −8. calculate the values of the relative errors and make an evaluation for the permissible error value ±20%.9 Problem: From the data given in Example 6.6 −1.3 Conclusion Satisfactory Unsatisfactory Satisfactory Satisfactory Satisfactory Satisfactory Unsatisfactory 123 111 128 138 121 123 188 114 188 122 121 142 125 132 129 121 198 131 158 193 122 111 140 20. Data: results: Data lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 xref x Solution: ε.1.1 34. % lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 −12.6 −12.1 −20. LLC .0% .4 −13.Interlaboratory Comparisons 117 Example 6.© 2009 by Taylor & Francis Group.

9 −13.6 34.6 1.16). but within the accepted interval.4 −6.4 −10.10 Problem: For the data given in Example 6. the standardized Z coefficient. The method of its determination is described in detail in Chapter 1 (Section 1. solely attributable to the high value of the extended uncertainty. Example 6.118 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 Excel file: exampl_PT09. En is a parameter that is decidedly less restrictive than.7 −5. apply En Score.7 Quality Assurance and Quality Control Satisfactory Unsatisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Unsatisfactory Satisfactory Satisfactory Unsatisfactory Satisfactory Unsatisfactory The next parameter for the individual evaluation (for each of the laboratories) of obtained results is En.9 −20.8 14 16 10 11 14 18 . because of the inclusion of the uncertainty value.1.9 −13. may be considered an outlier.7 −7.9 −12.3 −12.6 41. Data: results: Data lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 123 111 128 138 121 123 188 114 u 11 9. An opposite situation is possible — a result closer to the mean (compared with another result from a given series) but with the smaller value of extended uncertainty.4 12.© 2009 by Taylor & Francis Group.xls −18.8. for example. LLC . Results that are deemed satisfactory may include values significantly deviating from the mean.9 37.

01 −1.Interlaboratory Comparisons lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 xref uref Solution:   lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 Excel file: exampl_PT10.43 Conclusion Unsatisfactory Unsatisfactory Satisfactory Satisfactory Unsatisfactory Unsatisfactory Unsatisfactory Unsatisfactory Unsatisfactory Satisfactory Unsatisfactory Satisfactory Satisfactory Satisfactory Satisfactory Satisfactory Unsatisfactory Satisfactory Satisfactory Unsatisfactory Unsatisfactory Unsatisfactory 188 122 121 142 125 132 129 121 198 131 158 193 122 111 140 11 23 15 11 13 12 17 19 21 28 14 18 13 14 17 119 .40 −0.50 −0.10 −1.22 0.88 −0.92 −0.09 2.97 −1.xls En −1.67 −0.23 1.80 1.51 0.11 −1.09 −1.12 −0.© 2009 by Taylor & Francis Group.28 −1.97 −0.70 −1.93 −0. LLC .85 3.

one should divide all the measurement results obtained for a given sample into subsets. after which they are all put into one diagram. if the calculated value is equal to q1. box plots may be used. as 1. for each subset.5. one may examine if the results obtained using various analytical procedures differ among themselves in a statistically significant way. then whiskermax is not marked on the diagram. the values of median and quartiles (q1 and q3) are marked — it is a so-called box area representing the middle 50% of the data. not smaller than the limit equal q3 + 1. not smaller than the limit equal q1 – 1. 2.120 Quality Assurance and Quality Control 6. and which procedure yields more accurate data. it is possible to conclude which of the analytical procedures were used more often. Due to the type of construction of the graph. Then. separate plots are drawn. then whiskermin is not marked on the diagram.5 · IQR. On the 0Y axis. Results out of this range (lower than whiskermin or higher than whiskermax) are marked as outliers. the difference between q3 and q1 • Determination of maximum values.5 times the IQR Based on calculated values. In drawing such a plot. the maximum value in the set of results. 3. in the following manner: 1. whiskers are marked as: a. Whiskermin. whiskers.5 · IQR. one calculates the essential values based on the following reasoning: • Ordering the result in a nondecreasing sequence • Determination of median and quartiles: first (q1) and third (q3) • Determination of the interquartile value (IQR). a diagram (plot) is drawn (separately for a given set of results). b. for a given series marked by one point on the 0X axis. if the calculated value is equal to q1.© 2009 by Taylor & Francis Group. On the same plot.1  Comparisons of Results Obtained Using Various Procedures In this type of comparison. In the graphical presentation of results. LLC . the minimum value in the set of results. Based on data for which the diagrams (plots) are drawn. . Whiskermax. each containing results obtained using a specific analytical procedure.

5 × IQR q1 − 1.6 170.6 91.5 121.5 × IQR min max whiskermin whiskermax 126.5 × IQR q3 + 1.3 141. LLC . construct a box plot graph.Interlaboratory Comparisons 121 Example 6.0 19.1.6 111 198 111 158 123 111 128 138 121 123 188 114 188 122 121 142 125 132 129 121 198 131 158 193 122 111 u 11 9.8 29.© 2009 by Taylor & Francis Group.11 Problem: For the data given in Example 6. Data: results: Data lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 Solution: Median q1 q3 IQR 1.8 14 16 10 11 14 18 23 15 11 13 12 17 19 21 28 14 18 13 14 17 .

0 80.0 Quality Assurance and Quality Control Outlier or Not OK OK OK OK OK OK Outlier OK Outlier OK OK OK OK OK OK OK Outlier OK OK Outlier OK OK 123 111 128 138 121 123 188 114 188 122 121 142 125 132 129 121 198 131 158 193 122 111 WhiskerMAX Q3 Median Q1 WhiskerMIN xLab .0 40. LLC .0 20.0 60.0 0.0 100.© 2009 by Taylor & Francis Group.0 160.122 xlab lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 lab 14 lab 15 lab 16 lab 17 lab 18 lab 19 lab 20 lab 21 lab 22 Graph: 180.0 120.0 140.

0 100.2  Comparison of Measurement Results Obtained in a Two-Level Study (for Two Samples with Various Analyte Concentrations) A two-level study is a study where each of the participating laboratories has performed the series of determinations: • Either two series per one sample. a graphical method — also called the Youden diagram [8] — may be used. • Dotted lines are drawn (also vertical and horizontal) where distances from the solid lines represent values of the standard deviation from the values of main distribution estimators (arithmetic mean or median). Application of this graph shows which of the participating laboratories achieved comparable results and which laboratory obtained deviating results.and interlaboratory variability. or • Determinations for two different samples In this case. .0 20.0 160. It is an easy and very effective method of comparing both intra. • Solid lines are drawn (both vertical and horizontal) that reflect the values of main distribution estimators (arithmetic mean or median). The graph is constructed as follows: • Measurement results for both the obtained series are marked on the X and Y axes. to determine the presence of systematic errors. LLC .0 80.0 120.0 0.5.Interlaboratory Comparisons Graph — modified (box plot): 180.0 140.0 60.0 40.xls 6.0 xLab 123 Excel file: exampl_PT11.© 2009 by Taylor & Francis Group.

When the main cause of the deviations from the mean or median are random errors.4 Series 2 12. produce a Youden graph.2 11.7 10.© 2009 by Taylor & Francis Group.8 14.2 15.4 11 12.1 12. It may indicate a positive or negative bias in the analytical procedure applied in a given laboratory.6 11.7 10.7 11. If a systematic error is the main cause of differences between the values of the measurement results obtained by the compared laboratories and the mean (median).3 11.7 11.12 Problem: For the two given series of measurement results for two examined samples obtained in the examining laboratories.8 11.3 11.8 12.7 13. Example 6.8 11 10.124 Quality Assurance and Quality Control The distribution of points on such a constructed diagram is a source of information about what type of error has a dominant impact on the obtained measurement results.8 . then the majority of points are in the upper right or bottom left quarter of the graph. the results are distributed in a random manner around the mean (median).9 10.2 10. Data: results: Data Series 1 lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 lab 9 lab 10 lab 11 lab 12 lab 13 Solution:   Series 1 Series 2 11.5 10.9 11.2 11.9 11.5 Median 11. LLC .8 12.

17). The manner of conducting Mandel h and k tests is described in Chapter 1 (Section 1. quite common method of graphical presentation of the measurement results obtained by comparing laboratories is the application of Mandel h and k tests.Interlaboratory Comparisons Graph: 16 15 14 Serie 2 13 12 11 10 9 8 8 9 10 11 12 Serie 1 13 14 15 16 125 Graph — modified (with 95% limit circle): 16 15 14 Serie 2 13 12 11 10 9 8 8 9 10 11 12 Serie 1 13 14 15 16 Excel file: exampl_PT12.© 2009 by Taylor & Francis Group. The application of these tests enables the presentation of the variability of results obtained by using a given analytical procedure and enables an evaluation of a given laboratory.8. LLC . All laboratories may obtain on different levels of a study (for different analytes or for different concentrations of a single analyte) both positive and negative values of parameter h. .xls Another.

2 14 13.126 Quality Assurance and Quality Control The number of laboratories characterized with positive values of parameter h should approximate the number of laboratories characterized with negative values. Draw a graph showing the respective values of the calculated h parameters characterizing the sets of results obtained in individual laboratories.03 2. When a laboratory tends to obtain only negative values for h.86 2.01 2.23 4. When graphs for h and k connected in groups corresponding to the individual laboratories show that the values of these parameters are close to the lines of critical values.11 11.21 6.67 7. it achieved an unusually high number of large values for h.5 14.78 1.11 2.02 Level 2 7.8 12.7 13. one should pay attention to the problem of systematic errors and the small repeatability of results (great variance value).11 4. the situation should be adequately explained. Example 6.9 14.22 2.8 9.8 14.23 8.76 4.22 5.3 11 12.77 7.4 15.54 5.03 5.83 7. and at the same time different from the sign (plus or minus) of parameter h obtained in other laboratories. when a laboratory yields h values in the extreme range.56 4.44 4.98 4.55 2.34 4. one should pay attention to a situation where all values of parameter h for a given laboratory are characterized with a positive or negative value.5 11.8 12.44 1.98 1.34 7.7 lab 2 lab 3 lab 4 lab 5 lab 6 .11 Level 3 2. Similarly.3 13.6 10 14.67 9.22 2.8 11.89 1.15 2.3 6.4 15. Moreover.54 7.21 7.08 2.34 2. Data: results: Level 1 lab 1 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 4.92 8. When the graph of the statistical parameter k indicates that a given laboratory deviates from the rest due to numerous high values.73 5.54 7.22 4.2 12 11 11.02 8.77 7.34 6. one may suppose that there is a source of bias for the results obtained by that laboratory.1 Level 5 11.11 5.© 2009 by Taylor & Francis Group.32 4.73 2.4 11. LLC .13 Problem: For a given set of results obtained in interlaboratory comparison.2 14.12 2.65 2.96 10. for example.56 4.45 1.48 2.4 10. it shows a smaller repeatability of results obtained by the laboratory compared with the rest of the participants.5 10.2 13.01 8.7 11 10.23 4.23 5.45 8.12 1.8 16. calculate the values of Mandel’s h parameter.56 Level 4 14.

77645 1.02508 0.08 2.5 1 0.18460 −0. Excel file: exampl_PT13.11 2.11 9 2.04319 −1.22 1.25261 Level 5 0.76 8.15118 1.67 8.72498 −0.32 3.2 13.45 8.5 –1 –1.85611 Conclusion: Results obtained by “lab 4” for all analytes are much lower compared to those obtained by the rest — three of five analytes have exceeded the critical value for the 1% level of significance. which indicates the occurrence of a systematic error source for the results obtained by this laboratory.33181 −0.26898 0.45 1.35552 −0.84978 0.5 –2 –2.2 11 12 12.6 11.23756 −0.88 8.© 2009 by Taylor & Francis Group.xls .66888 1.55 4.71269 0.10230 −2.46858 −1. Results obtained by the other laboratories are within the permissible range of changes for all the determined analytes.51103 0. LLC .3 8.13474 Level 3 h 0.98 9.83572 0.99211 0.8 11.02 7.5 –3 Lab 1 Lab 2 Lab 3 Lab 4 Lab 5 Lab 6 Lab 7 Lab 8 Level 2 −0.57 13.5 0 –0.21330 0.32112 0.22323 −0.5 2 1.1 13.88856 0.82 127 lab 8 Solution: Level 1 lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 Graph: 2.45130 0.21 3.15017 −2.51103 −0.49321 0.11 8.19003 −1.21952 1.35 3.24180 0.38729 1.56 1.Interlaboratory Comparisons lab 7 1 2 3 1 2 3 4.40861 −1.25142 Level 4 0.26092 0.56151 0.3 11.06872 −2.44 4.15886 0.63583 0.

11 2.2 14.55 2.74083 0.34 6.3 11.3 11 12.1 13.3 6.03 5.2 13.34 7.92 8.8 11.98 4.26287 0.23 5.56 4.5 10.34 2.77 7.11 4.89 1.74083 0.26287 0.44 4.45 1.74823 1. LLC .37413 0.12845 0.56 4.8 11.45 8.11 5.11 8.98 9.3 8.© 2009 by Taylor & Francis Group.98 1.22 1.3 13.77 7.32 3.15 2.01 8. calculate the values of Mandel’s k parameter.48 2.11 9 Level 3 2.128 Quality Assurance and Quality Control Example 6.44274 0.2 Level 5 11.7 11 10.68527 1.76 4.8 14.02 7.5 11.8 16.21 7.01 2.7 13.44274 Level 4 Level 5 .11 11. Draw a graph showing the respective values of the calculated k parameters characterizing the sets of results obtained in individual laboratories.65 2.71646 1.55 4.23 4.26287 0.74083 0.68527 1.74083 0.57 Level 4 14.68527 1.54 7.37413 0.67 7.12 1. Data: results: Level 1 lab 1 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 4.44274 Level 3 k 0.26287 0.02 4.67 8.9 14.45 Level 2 7.22 2.22 4.12845 0.1 13.2 13.03 2.2 14 13.22 2.74823 1.86 2.67 9.26287 0.21 3.54 5.37413 0.08 2.78 1.44 4.68527 1.2 12 11 11.14 Problem: For a given set of results obtained in an interlaboratory comparison.6 11.11 2.8 9.12 2.74083 0.02 8.74823 1.4 15.6 10 14.68527 1.12845 0.8 12.56 1.4 15.7 11 12 12.71646 1.4 10.21 6.73 2.44274 0.5 14.45 1.73 5.76 8.88 8.44274 Level 2 0.74823 1.37413 0.34 4.12845 0.71646 1.23 4.8 12.54 7.32 4.4 11.74823 1.11 8.83 7.44 1.56 2.12845 0.82 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 Solution: Level 1 lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 0.23 8.35 3.22 5.96 10.71646 1.37413 0.71646 1.08 2.

and enables issuance of opinions on organizational procedures. It is hence obvious that laboratories that do not participate in these comparisons should be deemed unreliable. it is a system of mutual aid where a participant obtains information whether and how they should modify the applied measurement procedure to increase the reliability of obtained results.Interlaboratory Comparisons Graph: 2. • Results of interlaboratory studies enable the detection and definition of current problems in a given laboratory. both on national and international scale. High marks/grades obtained in interlaboratory proficiency studies indicate a high quality of analyses performed by the participating laboratory.xls 6. while interpreting the results of the interlaboratory studies. Bearing this in mind.5 1 0.5 2 1. one should remember that: • Participation in interlaboratory studies must not serve as a substitute for routine intralaboratory control of the results’ quality. lab 5.6 Conclusions The ultimate and most reliable manner of estimating the quality of measurement results obtained by a given laboratory is the comparison of their results with those obtained in other laboratories. and lab 6).” In the case of individual results (lab 1.© 2009 by Taylor & Francis Group. A major task in interlaboratory comparisons is the help offered to a laboratory in detecting all types of irregularities during a given analytical procedure that may affect the reliability of the obtained results. and not those that may occur. laboratories for many years have participated in various interlaboratory comparisons. the obtained values of repeatability exceed the critical value for the 5% level of significance.5 0 Lab 1 Lab 2 Lab 3 Lab 4 Lab 5 Lab 6 Lab 7 Lab 8 129 Conclusion: The greatest repeatability for results obtained is achieved by “lab 8. The test of interlaboratory proficiency is used to estimate the reliability of determination results and is the basis for the validation of analytical procedures according to EN 17025. Excel file: exampl_PT14. . In other words. LLC . However.

A.. Vander Heyden. P. Edition 1. ISO/IEC Guide 43-1.. Davies... 11. Juniper. Konieczka. 3. Assur. References 1.. 2007. Assur. and Smeyers-Verbeke. 1158. 8. R. Kandel. Qual. Qual. 1997. 173–190. “The role of and place of method validation in the quality assurance and quality control (QA/QC) system.” Analyst.R. 1998. 11. Y. 7. Use... T. P.” Fresenius Z. 1997. D. Rev. 9..com). 4. Anal. M. 331. “Statistical treatment of proficiency testing data... 158–167. Chem.” Accred Qual. ISO/IEC Guide 43-1. Assur. 2000. and Ellison. Thompson.J.P. LLC .pdf.W. 513–519. the major task of interlaboratory studies is to obtain an explicit answer to the question: “Are the measurement results obtained in a given laboratory as good as we think they are?” (http://www. 4. Analytical Methods Committee.L. 97–104. 6. Proficiency testing by interlaboratory comparisons: Part 1. “Set-up and evaluation of interlaboratory studies. 322–327. 117. 1988. 373–378. To sum up. the same applies to an analytical method. Proficiency testing by interlaboratory comparisons: Part 2. 10. Chem.© 2009 by Taylor & Francis Group. . J. 37.130 Quality Assurance and Quality Control • A successful outcome in interlaboratory studies obtained during the determination of a given analyte or a group of analytes may not be automatically related to another analyte or group of analytes.. 2007.” J. 362–366. Chromatogr. ProficiencyTesting. 1999.. Selection and use of proficiency testing schemes by laboratory accreditation bodies. 12. Available at http://www. Qual. M. “The influence of different evaluation techniques on the results of interlaboratory comparisons. “Proficiency Testing of Analytical Laboratories: Organization and statistical assessment. 1992. I.HN-Proficiency.. 1998. 3. 2006.” Accred. “Quality issues in proficiency testing. 336–341. Assur.l‑a‑b.” Accred.0. and Grasserbauer.” Accred.L.com/webshared/administration/ Hows%20And%20Whys%20of%20of%20. “Statistical evaluation of interlaboratory tests. Development and operation of proficiency testing schemes. Linsinger. S... Krska.. “Fitness for purpose — the integrating theme of the revised harmonized protocol for proficiency testing in analytical chemistry laboratories. Tholen. W..” Crit. Anal. 3. 5. 2..R. Eurachem Guide on Selection. and Interpretation of PT Schemes.

7 Method Validation
7.1  Introduction
Considerations concerning the determination of validation parameters should begin with an explanation and description of the nature of an analytical measurement. The key interests of analysts worldwide are the signals following and resulting from a conducted measurement. The goal of an analyst’s work is to obtain analytical information about an investigated object based on a received output signal, a result of a suitable measurement method. This signal reveals information about the investigated sample. The analyst’s role is to “decode” the obtained signal and do it in a manner such that the obtained information is as reliable as possible [1]. A tool that decodes information is an analytical process, including analytical methods applied in the process. Each signal is characterized by a particular quantity. In some measurements, a signal may also be assigned a position (location). Validation parameters are determined based on analysis of the obtained signal values, and one should be aware of this in the validation of any analytical method. Validation of an analytical method includes testing of its important characteristics. The final aim is to be certain that the analysis process is reliable and precise, remains under total control of the operator, and leads to reliable results. First of all, validation allows definition of a given analytical method. Using the determined parameters, in the validation process there exists the possibility of estimating the usefulness (range of use) for a given method and then choosing the optimal method. As previously stated, for the measurement results to be traceable and have an uncertainty value provided, they must be obtained using an analytical method that is subjected to a prior validation process. Most often, a validation study is carried out when [2, 3]: • A new analytical method is being developed. • Tests for the extension of the applicability of a known analytical method are being conducted, for example, determinations of a given analyte, but in samples characterized by a different matrix composition. • Quality control of the applied method showed variability of its parameters over time. • A given analytical method has to be used in another laboratory (different from the one in which it has already been subjected to the validation process) or using different instruments, or determinations are to be performed by another analyst. • A comparison of a new analytical method with another, known reference method is being performed.
131
.© 2009 by Taylor & Francis Group, LLC

132

Quality Assurance and Quality Control

The parameter range, the determination of which should underlie the validation process for a given analytical method, depends on the following factors [4]: • The character of an analytical study to be carried out using a given analytical method (qualitative or quantitative analysis, analysis of a single sample, or a routine analytical investigation). • Requirements for a given analytical method. • Time and costs, which need to be spent in the validation process. The parameters considered necessary for the validation of different types of analytical procedures are presented in Table 7.1 [2, 5]. TABLE 7.1 Parameters Whose Determination is Necessary for Different Types of Analytical Procedures [2, 5]
Impurity Test Parameter Precision Correctness Specificity Limit of detection Limit of quantitation Linearity Measuring range Ruggedness
a

Qualitative Analysis −a − + −a −a −a −a +

Limit Impurity Test − −a + + − − −a +

Quantitative Impurity Test + + + – + + + +

Assay Test + + + – + + +

It might be determined.

The more parameters included in the validation process, the more time one should spend on the process. In addition, the more restrictive the assumptions for the limit values (expected) of the respective parameters, the more often one should test, calibrate, or “revalidate” a given analytical method. It is not always necessary to conduct a full analytical method validation. Therefore, one should determine which parameters should be included in the process. Table  7.2 contains the parameters which, according to the recommendations of the International Conference on Harmonization (ICH) [6, 7] and United States Pharmacopeia (USP) [8], should be included in the validation process.

.© 2009 by Taylor & Francis Group, LLC

Method Validation

133

TABLE 7.2 List of Analytical Procedure Parameters That Should Be Validated According to Recommendations of ICH [6, 7] and USP [8]
Parameter Precision Repeatability Intermediate precision Reproducibility Accuracy Limit of detection Limit of quantification Specificity/selectivity Linearity Measuring range Robustness Ruggedness ICH + + + + + + + + + USP +

+ + + + + + + +

Apart from determining validation parameters, before commencing validation one should determine the basic features of an analytical method, namely [2]: • • • • • • • • • • • Type of the determined component (analyte) Analyte concentration Concentration range Type of matrix and its composition Presence of interferents Existence of top-down regulations and requirements for the examined analytical method Type of the expected information (quantitative or qualitative analysis) Required limits of detection and quantitation Expected and required precision and accuracy of the entire method Required robustness of the method Required instruments; whether the determinations using a given method have to be carried out using a strictly defined measuring instrument or instruments of a similar type Possibility of using a method already validated in another laboratory(ies)

.© 2009 by Taylor & Francis Group, LLC

134

Quality Assurance and Quality Control

A validation process may be conducted in any order; however, it seems most logical to proceed in the following manner [2, 4]: • Determine the selectivity in the analysis of standard solution samples (optimization of the separation conditions and determination of analytes present in the standard solution samples). • Determine the linearity, limits of detection and quantitation, and the measuring range. • Determine the repeatability (short-term precision), for example, based on deviations of the obtained retention times and/or chromatographic peak areas. • Determine the intermediate precision. • Determine the selectivity based on results obtained in the analyses of real samples. • Determine the accuracy/trueness based on the analysis of reference material samples containing an analyte at different concentration levels. • Determine the tolerance of a method, for example, based on the results obtained in interlaboratory comparisons. The validation process requires the use of various tools such as [9]: • • • • • • Blank samples (including so-called reagent blanks) Standard solutions (calibration solutions, test samples) Samples with a known quantity of added analyte (spiked with the analyte) (Certified) reference materials Repetitions Statistical processing of the results

In this work, we need to stress that the method can be subjected to the validation process only when a suitable optimization study has been conducted. The process of analytical method validation should be completed with the final report, which includes all information concerning the analytical method. Validation parameter definitions and the manner of their determination are described below.

7.2 Characterization of Validation Parameters 7.2.1 Selectivity
Usually, the first determined validation parameter is selectivity. Using basic logic, before one commences determination of the properties of an analyte based on measurement of the obtained analytical signal, one should make sure that a given signal is due only to the occurrence of an analyte in an investigated sample. A very frequent problem is the interchangeable use of the terms selectivity and specificity, although they differ in their essential meaning. According to the International Union of Pure and Applied Chemistry (IUPAC) nomenclature [10], selectivity is defined as “the extent to which it can determine particular analyte(s) in a complex mixture without interference from other components

.© 2009 by Taylor & Francis Group, LLC

Method Validation

135

in the mixture.” Specificity is described by the IUPAC as the “highest selectivity” and recommends not using the term specificity. Selectivity is thus the ability of a method to differentiate the examined analyte from other substances. This characteristic is mostly a function of the described measurement technique, but can fluctuate depending on the class or group of compounds to which the analyte belongs, or the sample matrix. A specific method is one which shows the highest selectivity. Selectivity can be defined as [11] “the ability of an analytical process to receive signals whose size depends almost entirely on the concentration of the examined analyte present in the sample.” One can also propose a practical definition [9]: “selectivity is the potential for an accurate and precise determination of the occurrence and/or concentrations of an analyte or groups of analytes in the presence of other components in a real sample under given measurement conditions.” Selectivity is therefore one of the main parameters characterizing and describing an analytical method, especially a trace analysis [12]. From a practical point of view, an analytical measurement is selective when it is possible to differentiate measurement signals and assign to them respective properties for a given analyte. This undoubtedly depends on the parameters of the obtained signal. If the signal is characterized only by its intensity, one should prove that its size depends only on the investigated properties of a given object. For example if the mass of a sample is being determined using an analytical balance, then an analyst must be certain that the measured value is due to the real mass of a sample and not, for example, refuse on the balance’s tray. This example shows that problems related to selectivity are also linked with direct measurements. A different situation is observed concerning selectivity when signals are characterized by an additional parameter — position (place). Such a situation takes place in chromatography for example, where retention time additionally characterizes the output signal and assigns it to a specific analyte. In such a case, it becomes necessary to determine the smallest differences between the positions for each analyte, for which the distinction between the obtained signals is possible. The requirement of selectivity for a measurement process depends first of all on the composition of an analyzed sample [11]. Selectivity is more difficult to obtain: • • • • • • The more unknown the sample composition is The more complex the sample’s matrix composition is The more similar the properties of the matrix components The greater the number of analytes The smaller the analyte concentration The greater the resemblance between analytes

An increase in the selectivity of an analysis may be obtained by: • Use of selective analytical methods • Elimination of the influence of interferents by removing or concealing them • Isolation of the analyte from the matrix

.© 2009 by Taylor & Francis Group, LLC

136

Quality Assurance and Quality Control

Depending on the type of analytical technique, the various ways of expressing selectivity are different.

7.2.2 Linearity
When an investigated property is certain to be associated with a given signal, one should determine the dependence between these quantities. A linear dependence is the most frequently occurring in analytical chemistry. The vast majority of analytical measurements use the calibration step, when the output signals are assigned to corresponding analyte concentrations [13]. To determine the functional dependency associating the output signal with analyte concentration, the linear regression method is commonly used. It is also applied in the determination of some validation parameters, such as: • Linearity • Trueness (based on the value of biases) • Limits of detection and quantitation It is also widely used in the calibration of measuring instruments. Linearity is defined as an interval in the measurement range of an analytical method in which an output signal correlates linearly with the determined analyte concentration. The most frequently used method of determining linearity is by using a graph of measuring instrument calibration. To this end, measurements of standard solution samples are conducted on at least six levels of concentrations (most often three parallel measurements for each level). Naturally, the selection of analyte concentrations in standard solution samples should be such that their range should include the expected analyte concentration in an investigated sample (the concentration range usually covers values from 50% to 150% in relation with the expected results of an analysis) [14]. Then, using the linear regression method, one determines the regression parameters. According to some recommendations [15], it is sufficient to calculate the coefficient of regression. Then, if this value is at levels equal to at least 0.999, we may talk about the linearity of the method within the range of concentrations for which standard solutions were prepared to determine the calibration graph. Unfortunately, this manner of documenting linearity does not always lead to correct conclusions. It can happen that the high value obtained for the coefficient of regression r (or the coefficient of determination r 2) does not necessarily prove the linearity of a method. The coefficient of regression may be used to infer the linearity of an analytical method only when standard solutions, based on which the calibration curve is determined, fulfill the following requirements [14, 16, 17]: • They include the expected analyte concentration in the investigated sample(s) within their own range of concentrations. • They include no more than three orders of magnitude of analyte concentrations within their own range. • They evenly “cover” the whole range of concentrations.

.© 2009 by Taylor & Francis Group, LLC

Method Validation

137

In addition, it is very important to determine a suitable dependence and the “visual” analysis of the obtained graph. Because of the ambiguity in usage of the coefficient r as a measure of linearity, additional methods for proving linearity have been proposed. In addition, the significance of the calibration graph coefficients needs to be determined. The coefficient of direction should differ statistically and significantly from 0, and in the case of an absolute term, its value should not differ in a statistically significant way from 0. To ascertain this, one should calculate the values of Student’s t (Section 1.8.9). Another approach is to draw a so-called graph of constant response described by the following dependence [2]: y = f (x) x (7.1)

where y = signal of a measuring instrument x = analyte concentration in a standard sample corresponding to a given signal When the range of concentrations is sufficiently large (including three or more orders of magnitude), the concentrations may be marked on the graph in a logarithmical scale. On such a graph, the sustained response is marked (usually calculated as an arithmetical mean of individual values y/x) in the form of a line parallel to the X axis, along with the admissible deviations from this value (most often ±5%). Values (points) lying outside the determined range correspond to analyte concentrations that lie outside the linear range of the measuring instrument. Naturally, this process can only be used when an absolute term of the determined simple dependence y = f(x) does not differ in a statistically significant manner from zero, which is not always the case. In some studies, one can find unambiguous and categorical statements that the value of coefficient r cannot serve to determine the degree of dependence between variables, and should be replaced by another statistical tool or specific tests for proving linearity [18]. One of the recommended tools is variance analysis. One can also use other methods and statistical tools such as [19–22]: • • • • Test of adequacy Mandel’s test Quality factor Student’s t test (Section 1.8.9)

When proving linearity is based on analysis results of the standard solution series with the simultaneous drawing of a calibration graph, it is logical to prove to what extent the calibration curve reflects the signals for standard solution samples. One can ascertain this through the calculation of relative errors for each concentration, with the reference value being the analyte concentration in the standard sample,

.© 2009 by Taylor & Francis Group, LLC

138

Quality Assurance and Quality Control

and the experimental value being that calculated from the equation of a straight calibration line [23]. Linearity by no means signifies that, within the entire range of concentrations, the function describing the dependence of the output signal on the analyte concentration assumes one form (the same calibration curve coefficients). Linearity is a characteristic showing the proportional dependence of a signal on the determined quantity, and can be described, for a given range, by several equations depending on the level of analyte concentrations [24, 25]. It is also necessary to explain the difference between correlation and regression. Correlation describes the degree of connection between two variables, and regression describes the manner of their dependence [18]. Example 7.1
Problem: Draw the calibration curve based on the results of the analyte concentration determination results in six standard solution samples (three independent measurements per each of the solutions). Calculate the regression parameters of the calibration curve. Make an appropriate graph. Data: results: Data x 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 2 2 2 4 4 4 6 6 6 8 8 8 10 10 10 12 12 12 y 1.12 1.20 1.08 2.11 2.32 2.23 3.33 3.54 3.41 4.12 4.32 4.44 5.67 5.76 5.51 6.97 6.78 6.66

.© 2009 by Taylor & Francis Group, LLC

Data: results: Data x 1 2 3 4 5 2 2 2 4 4 y 1. SDb Standard deviation of the intercept.32 .11 2. LLC .© 2009 by Taylor & Francis Group.07702 0. SDa Regression coefficient.5642x – 0.08 2. r Graph: 8 7 6 Signal 5 4 3 2 1 0 0 2 4 6 8 10 12 14 y = 0. a Residual standard deviation.1433 0.9976 Content Excel file: exampl_valid01. b Intercept.9951 139 18 0. SDxy Standard deviation of the slope.1.5642 −0. Calculations should be performed for the significance level α = 0.05.0291 R2 = 0.00989 0.12 1. examine the significance of the differences in the slope and the intercept of a calibration line and the value 0.Method Validation Solution: n Slope.xls Example 7. Apply Student’s t test.2 Problem: Using the data from Example 7.20 1.0291 0.

00989 0.0291 0.32 4.062 0. b Intercept. SDxy Standard deviation of the slope. marking the lines of the interval for the values deviating ±5% from the mean.12 4.23 3.44 5. Excel file: exampl_valid02.76 5.1433 0.97 6.9976 57.378 2.20 .54 3. draw a graph of sustained response.3 Problem: Using the data from Example 7.51 6.41 4.07702 0.66 n Slope. SDb Standard deviation of the intercept. Data: results: Data x 1 2 2 2 y 1. SDa Regression coefficient.67 5. a Residual standard deviation. r tb ta tcrit 18 0.5642 −0.33 3.xls Example 7. No statistically significant difference between the intercept and 0.140 6 7 8 9 10 11 12 13 14 15 16 17 18 Solution: 4 6 6 6 8 8 8 10 10 10 12 12 12 Quality Assurance and Quality Control 2.120 Conclusions: Statistically significant difference between the slope and 0.© 2009 by Taylor & Francis Group.78 6.12 1. LLC .1.

Method Validation 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 2 4 4 4 6 6 6 8 8 8 10 10 10 12 12 12 1.55 0.56 0.33 3.44 5.52 0.59 .58 0.54 0.60 0.© 2009 by Taylor & Francis Group.58 0.57 0.51 6.23 3.54 0.32 4.57 0.56 xm − Interval % 0.57 0.53 0.66 141 Solution: y/x 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0.08 2.67 5.41 4.97 6.53 xm + Interval % 0.56 0.59 0.12 4.32 2.78 6.56 xm 0. LLC .58 0.56 0.76 5.54 3.56 0.11 2.

40 0 2 4 6 Quality Assurance and Quality Control 8 10 12 14 16 Excel file: exampl_valid03.54 3.08 2.60 0. and the experimental value to be the value calculated from the calibration curve equation.4 Problem: Using the data from Example 7.41 4.44 5.142 Graph: 0.23 3.xls Example 7.76 5.65 0.33 3. assuming the reference value to be x.67 5. Assume an appropriate limit for the relative error and draw conclusions. Data: results: Data x 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 2 2 2 4 4 4 6 6 6 8 8 8 10 10 10 y 1.1.12 1.20 1.32 2.12 4. calculate the values of the relative errors for individual values x.32 4.70 0.75 0.11 2.45 0.50 0.© 2009 by Taylor & Francis Group.55 0. LLC .51 .

Method Validation 16 17 18 12 12 12 6.xls 1. b Intercept.60 −1.65 −0. n Slope.99 1.59 −8.56 −1. The value of this parameter can serve to determine the influence of noise level on the . The values of these parameters are closely related to the magnitude of noises in the measurement system. % 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Excel file: exampl_valid04. Signal-to-noise ratio (S/N) is an undimensional quantity that describes the relationship of an analytical signal to the mean noise levels for a specific sample.92 −1.83 3.97 6.08 0.01 2.3 Limit of Detection and Limit of Quantitation The next validation parameters that need to be determined are the LOD and the LOQ.78 6.43 1.5642 −0.00 143 Relative error — ε.0291 Conclusion OK !!! OK !!! OK OK OK !!! OK !!! OK OK OK OK OK OK OK OK 7. LLC .66 5.78 5.2.37 0. % Solution: Number of results.10 −0. a ε.22 4.08 −3.72 −5.83 8.© 2009 by Taylor & Francis Group.21 18 0.

This can be attributed to several reasons: • A large number of definitions describing the notions of both the LOD and the LOQ. LLC .g. but the most common method is the relationship of the arithmetical mean of the results in a measurement series for blind samples (or samples containing analyte in a very low level) to the standard deviation obtained for this series. • Characteristics of the applied instrumental technique. for which it is not possible to determine the noise level of the applied measuring instrument. Depending on these parameters. Instrumental detection limit (IDL) (e.144 Quality Assurance and Quality Control relative measurement deviation.2. This method can also be used for instrumental techniques. one estimates this concentration level at which detection is possible. This value should be estimated using a suitable standard sample and should not be determined through extrapolation [27].3. It can be calculated in different ways. detector) is the lowest concentration (smallest quantity) of an analyte that can be detected (without quantitative determination) using a given measuring instrument. precision. one estimates the LOD based on one’s own experiment. • Possibilities of obtaining (producing) so-called blind samples. Limit of detection (LOD) is the lowest concentration (smallest quantity) of an analyte that can be detected with statistically significant certainty [26]. LOD and LOQ are parameters that play an unusually significant role in the validation of analytical procedures. there exist several ways of determining (estimating) the LOD. this value is n times the noise level — it is most often three times as high. and uncertainty. 7. • Practical difficulties in univocally determining the basic parameter deciding the LOD — namely. The manner of determining an LOD depends on the following factors: • Nature of the analytical method (the manual method and the method based on utilization of a suitable gauge as well). Based on the results of sample analysis with the known analyte concentration (standard solutions). the magnitude of the noise level in a given measuring instrument.1  Visual Estimation For a classical method (noninstrumental). Limit of quantitation (LOQ) is the quantity or the smallest concentration of a substance that can be determined using a given analytical procedure with an assumed accuracy. Although the meaning of these parameters and their understanding do not raise questions. the determination of their values itself is sometimes problematic. .© 2009 by Taylor & Francis Group. Method detection limit (MDL) is the lowest concentration (smallest quantity) of an analyte that can be detected using a given analytical procedure..

2) The modification is the preparation of n samples with analyte concentrations on a level close to the expected LOD. LOD is equal to the mean value magnified by three times the standard deviation in this instance.3) LOD = x m + 3 ⋅ SD (7. and then to directly apply the principle that LOD is three times the noise level for an applied analytical method. LLC . it would be most convenient to prepare . This method can be applied only when it is possible to obtain the baseline of noises.2.2. The method would only have some application when the analyte concentration is measurable for a blank sample.2 Calculation of LOD Based on the Numerical Value of the S/N Ratio When calculating the LOD.Method Validation 145 7. 10 independent determinations are performed for samples in which the analyte concentration is close to the expected LOD. For the thusly obtained 10 results. it is seldom possible to obtain a numerical value for the mean. where xm = mean value SD = standard deviation In practice. with the one difference being that LOD is calculated according to the formula: LOD = 0 + 3 ⋅ SD (7.3. such samples are prepared through spike in the blank samples with quantifiable amounts of the analyte. the analyte concentration in a blank sample is at least equal to the LOQ for the applied detector).3 Calculation of LOD Based on Determinations for Blank Samples A more labor-consuming method. In this instance. however. one describes the noise level — measuring range signal changes close to the retention time for an analyte on a chromatogram (one can assume the retention time range as t Ran ± 0. it seems paradoxical to obtain a result for a value that by definition should be a submarginal quantitation.3. Otherwise. one calculates the mean value and the standard deviation. the simplest and most commonly applied way of calculating the LOD is to determine the S/N ratio for a blank sample (if it is possible) or for a sample with a very low analyte concentration. In the case of chromatography. one can determine the LOD value by using the obtained chromatogram for a blank sample. but one that is also metrologically more correct. it is possible to use the described method with a certain modification [29. one uses the determined S/N ratio for the investigated analytical procedure [2]. obtained when a blank sample is subjected to final determination. of course. Of course. 30] — namely. the so-called background level is above the LOD for the applied detector (that is to say. is using a measurement for a series of blank samples. 7.© 2009 by Taylor & Francis Group. It involves 10 independent measurements for 10 independently prepared blank samples [28]. This quantity is then multiplied by 3 and the obtained signal value is converted into a concentration.5 min). To this end. that is. The manner of conduct is then similar to the previously described one.

LOD is calculated using a dependence described by an equation for the number of degrees of freedom f = n − 1.3. one determines the absolute term SDo.2. In this case. receiving a series of n results for which one calculates the mean value and standard deviation. then IDL is determined in this manner.4) where t = parameter of Student’s t test SD = standard deviation If the prepared standard solution samples are subjected to analysis using a given analytical procedure. calculate the standard deviations.3 ⋅ SD b (7. If determinations are instead performed directly on the prepared standard solution samples. A linear dependence is determined that associates the calculated standard deviations with the respective concentrations: SD = f (c) (7. then the determined LOD is also the MDL.7) where b is the slope of calibration curve.5) Next. 7.© 2009 by Taylor & Francis Group. one should perform at least six parallel determinations. For each level of analyte concentration. 7]. One then performs an analysis on such prepared samples. and then for each series of measurements obtained in this way. LOD is calculated using the following dependence: LOD = 3. LLC .6) 7.3. .5 Calculating LOD Based on the Standard Deviation of Signals and the Slope of the Calibration Curve One most often applies analytical methods in which the final determination is based on the indirect measurement principle. where n is the number of independent samples and the accepted level of significance α.2.146 Quality Assurance and Quality Control standard solutions in which matrix compositions correspond to the matrix composition of real samples. after which one determines the LOD according to the following dependence: LOD = 3 ⋅ SDo (7. it is indispensable to perform calibration that will influence the LOD [2. LOD = t ⋅ SD (7.4 Graphical Method This method involves analyses of measurement series for three standard solution samples containing an analyte at three levels of concentration (close to the expected LOD for the samples). 6. In this case.

20 18 16 14 12 10 8 6 4 2 0 Required precision. When the LOD is calculated based on parameters of the determined calibration graph (residual standard deviation or standard deviation of the intercept). • As a residual standard deviation of the calibration curve — SDxy. LLC . Figure 7.6 Calculation of LOD Based on a Given LOQ LOQ is the lowest analyte concentration that can be determined with a suitable precision and accuracy. 7.1 presents the construction of the graph and the calculation of the LOQ [28]. The required precision for the LOQ is determined (usually = 10%). . the MDL is the determined quantity. the calculated value is the LOD for the measuring instrument. 28].Method Validation 147 Standard deviation can be determined in three different ways: • As a standard deviation of results obtained for the series of blank samples — SDbl. one performs six parallel measurements. the coefficient of variation (CV) is calculated and the graph of the f(c) dependence is drawn. One performs measurements for standard solutions (matrix standards) on at least five levels of concentrations [2.© 2009 by Taylor & Francis Group.g. and for this value the concentration equal to the LOQ is read on the graph. For each level of concentrations. described by the dependence (1. 10% CV.1  Construction of the graph and calculation of the limit of quantitation [28]. the LOD (for the analytical method or the applied detector) will be calculated depending on which parameters were used to calculate the standard deviation. e. For each solution. Hence..3. The LOD is calculated as: LOD = LOQ/3.68).2. % LOQ 0 1 2 3 4 5 6 Analyte Content 7 8 9 FIGURE 7. • As a standard deviation of the intercept of the obtained calibration curve — SDa. if measurements are conducted based on analyses of blank samples subjected to the whole analytical procedure. Of course. It is also important to appropriately select concentrations of standard solutions to draw the calibration graph (it is known that the calibration graph has a straight-line range in a strictly specific interval of concentrations and that its plot most likely has different concentration level characteristics close to the LOD).

3.and time-consuming method that does not consider the influence of calibration on LOD Probability is used for estimating LOD Relatively quick method It includes the influence of calibration procedure on LOD value Labor.2. LOQ Table 7. The . and Advantages [27] Methods for Calculating LOD Visual check Requirements Sample with known analyte content (standard solution or matrix standard) Disadvantages/Advantages Quick method Estimation Mostly used in case of classical analysis (noninstrumental) Requires vast analytical experience Quick method Used only for measuring equipment It is possible to determine the S/N ratio Labor. together with their short characterizations [27].and time-consuming method It includes the influence of calibration procedure on LOD value Method “motivated” by metrology Indirect method LOD calculated based on the determined LOQ LOD value (LOQ) depends on the assumed measurement precision Calculations based on the S/N ratio Sample with known analyte content (standard solution or matrix standard) Calculations based on the measurements for sample blanks Series of blanks or samples with known analyte content (standard solution or matrix standard) Calculations based on graphical method Calculations based on standard deviations of signals and slope of calibration curve Series of standard samples at three concentration levels. 7.3 Methods for Determining Detection Limits: Requirements.7 Testing the Correctness of the Determined LOD Many of the aforementioned ways of calculating the LOD are based on the determination of analyte concentration in the prepared standard solution samples. LLC . at least six measurements for each standard sample Series of blanks or samples with known analyte content (standard solution or matrix standard) Standard solutions for calibration curve preparation Series of standard solutions Assumed relative standard deviation for LOQ Calculations based on limit of quantification.© 2009 by Taylor & Francis Group.3 compares all the described methods of calculating the LOD. Disadvantages.148 Quality Assurance and Quality Control TABLE 7.

Recovery can be calculated using the dependence [29]: %R = xm ⋅ 100% c (7.8) (7.Method Validation 149 solutions should be characterized. the analyte concentration in the prepared standard samples is too low. one should fulfill the following conditions [29]: 10 ⋅ LOD > cmin LOD < cmin (7. A recovery being too low results in an undervaluation of the calculated LOD. with two basic features: • Matrix composition should be as close to the matrix composition of real samples as possible.10) According to the definition of LOD. • Analyte concentration should be on a level close to the expected LOD. It is known that the standard deviation for the set of measurement results determining the analyte concentrations in standard solution samples strictly depends on the concentration levels of a determined component. the numerical value of this ratio should be between 3 and 10.9).8) is not fulfilled. LLC . It can happen that the concentrations in standard samples are considerably higher than the calculated LOD. To check the calculated LOD.© 2009 by Taylor & Francis Group. One should then calculate the limits of detection for newly prepared standard solutions with a lower analyte concentration. Inversely. when the condition is not fulfilled (7.11) where %R denotes the recovery of an analyte for a given analytical procedure. the determined LOD is greater than the numerical value and one should conduct remeasurement for smaller concentrations of the analyte in standard solution samples.9) where cmin is the analyte concentration in a standard solution sample with the lowest concentration. while calculating the LOD of an analytical procedure. If condition (7. In this case. it will signify that the concentration in the prepared standard samples is too high. one should remeasure and recalculate using standard solutions in which the analyte occurs in higher concentration levels. One should also pay attention to the recovery of the analytical method in measurements conducted for standard solutions. When it is higher. . one can also estimate the S/N ratio based on the following dependence [29]: S/N= xm SD (7. To test the trueness of the calculated LOD.

A correct method for recording a determination result depending on the quantity of an analytical signal is presented in Table 7. the described methods of testing the correctness of the calculated LOD can only be applied when the measurements are performed using prepared standard solutions. For validation of an analytical method. x x < LOD LOD ≤ x < LOQ x ≥ LOQ Recording of Result Not determined Not quantified Value of concentration Example 7. Using the calculated S/N ratio. TABLE 7. ng/g: Data 1 2 0. Data: results.4. the value of the determined LOD is associated with statistical parameters such as: • Level of probability • Number of degrees of freedom For individual measurements. The choice of a suitable means for determining the LOD depends on the purpose of the limit and the requirements of a given analytical method.5 Problem: Using the given analyte concentration determinations for blank samples. it is recommended to use a way the assumptions of which are based on chemical metrology. It must be stated that the determined LOD should always be given the description and parameters of the method applied in its calculation.150 Quality Assurance and Quality Control As previously stated.132 . examine the correctness of the determined LOD. 32]. The described ways of determining the LOD and/or quantitation permit the determination of both the MDL and the measuring instrument detection limit. estimate LOD and LOQ for the validated analytical method. The determined limits of detection and quantitation also show the quality of measurements conducted using a given analytical method [31. it is recommended to apply a less time-consuming method.© 2009 by Taylor & Francis Group.4 Correct Method for Recording a Determination Result Result. LLC .155 0. Determining limits of detection and quantitation allows the unequivocal determination and presentation of results in the proximity of these values.

014 ng/g 0.18 ng/g 0. Excel file: exampl_valid05. ng/g: Data 1 2 3 4 5 6 c 0.235 0. the standard solutions were made with concentrations near the expected LOD.244 0.xls Example 7.© 2009 by Taylor & Francis Group. it was noticed that the obtained values of signals cannot be measured. and based on measurements for these solutions the estimation was made for LOD and LOQ.253 0. Check the trueness of the LOD determination through the comparison with the standard solution concentration.135 ng/g 0.143 0. Hence.6 Problem: During the measurements performed on blank samples.145 0. calculated LOD is correct.113 0. Data: results. LLC .254 0.137 151 LOD = xm + 3 ⋅ SD S/N= xm SD S/N 9 Conclusion: S/N ratio is in the range 3–10.258 0.258 0.121 0.Method Validation 3 4 5 6 7 Solution: xm SD LOD LOQ 0.250 .54 ng/g 0.

6 9.5 6. estimate the LOD and LOQ of the validated analytical method. mg/dm3: Data 1 2 3 4 5 6 7 α Solution: xm SD t LOD LOQ 8. LLC .41 mg/dm3 1.0091 ng/g 0.4 9.152 Solution: SD LOD LOQ Quality Assurance and Quality Control 0.8 7.8 mg/dm3 8.082 ng/g LOD = 0 + 3 ⋅ SD 10 ⋅ LOD > cmin LOD < cmin Conclusion: Calculated LOD is lower than the standard solution concentration used for its determination and 10 times LOD is higher than standard solution concentration. Excel file: exampl_valid06.2 9. Using the data calculated S/N ratio.447 mg/dm3 2.© 2009 by Taylor & Francis Group. Data: results.6 0.8 7.05 . calculated LOD is correct.7 Problem: Using the data given in determinations of the analyte concentrations for blind samples. check correctness of the determined LOD. using Student’s t test.xls Example 7.3 mg/dm3 8.027 ng/g 0.13 mg/dm3 2.

Method Validation LOD = t ⋅ SD S/N= S/N xm SD 7 153 Conclusion: S/N ratio is in the range 3–10. check correctness of the LOD determination through a comparison with the standard solution with the lowest concentration.11 0.8 Problem: Using the given analyte concentration determinations for standard solution samples. Draw an appropriate graph.15 0.9 257. Present LOD in units of the analyte concentration in standard solutions applied for LOD estimation. estimate the LOD and LOQ using a graphical method. In addition.11 1 2 3 4 5 6 7 Solution: Concentration 0. ppm 0.040 signal ppm ppm SD (Signal) 26. Data: results: Concentration.© 2009 by Taylor & Francis Group. Excel file: exampl_valid07. LLC .013 0. calculated LOD is correct.2 0.xls Example 7.3 170.15 Signals 198 177 132 156 205 193 135 298 237 222 257 243 313 235 0.9 19.1 34.1 30.4 101 144 124 174 102 111 121 0.23 .23 SDo LOD LOQ Signal 125.

11 0. Excel file: exampl_valid08.15 Concentration 0. Data: results: Concentration.11 0.0 10. calculated LOD is correct.2 0.154 Graph: 40.5x + 19.© 2009 by Taylor & Francis Group. applied in LOD estimation.0 0. LLC .9 Problem: Using the data from Example 7.15 Signal 101 144 124 174 102 111 121 198 177 .2 0.11 0. comparing the calculated value with the value of the analyte concentration in the standard solution with the lowest concentration.25 10 ⋅ LOD > cmin LOD < cmin Conclusion: Calculated LOD is lower than the lower concentrated standard solution used for its determination and 10 times LOD is higher than the lower concentrated standard solution.0 SD (signal) 25.0 35. Also check the correctness of LOD determination.05 Quality Assurance and Quality Control y = 67.11 0.0 0 0.11 0.0 20.0 15.1 0.8. estimate the LOD and LOQ via the method using parameters of the calibration curve. Present the value of LOD in units of standard solution concentration.0 30. ppm 1 2 3 4 5 6 7 8 9 0.11 0.15 0.xls Example 7.0 5.11 0.

23 0.1 0.25 .23 0.6 129 22. r LOD (SDx.15 0.15 0.63 29. ppm 0.23 0.Method Validation 10 11 12 13 14 15 16 17 18 19 20 21 Solution: Number of results.15 Concentration. LLC .077 ppm 3.23 132 156 205 193 135 298 237 222 257 243 313 235 155 Graph: 350 300 250 Signal 200 150 100 50 0 0 0.05 y = 1102x + 4.8901 0.y) LOD (SDa) LOD (mean) LOD = 0.15 0.6 0.23 0. SDb Standard deviation.1 0.3 ⋅ SD b 21 1102 4.© 2009 by Taylor & Francis Group. n Slope. SDa Regression coefficient.089 ppm 0.2 0.066 ppm 0. b Intercept.15 0. a Residual standard deviation.23 0. SDxy Standard deviation.15 0.23 0.

© 2009 by Taylor & Francis Group.5 2.5 2.3 3. by a method using the parameters of the calibration curve.5 3.5 2.3 Signal 1460 1725 1150 1025 1825 1310 1950 1630 2200 1650 2000 1980 2900 3200 3245 2850 3500 3890 .3 3.2 2. Also check the correctness of LOD determination comparing the calculated value with the analyte concentration in a standard solution with the lowest concentration.5 2.3 3. LLC .2 1. Excel file: exampl_valid09.5 2. Data: results:   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Concentration.xls Example 7.2 1.3 3.2 1.2 1.156 Quality Assurance and Quality Control 10 ⋅ LOD > cmin LOD < cmin Conclusion: Calculated LOD is lower than the lower concentrated standard solution used for its determination and 10 times LOD is higher than the lower concentrated standard solution. calculated LOD is correct.10 Problem: Using the analyte concentration determinations for standard solution samples.3 3. Present the values of LOD in units of standard solution concentrations applied for LOD determination.2 1. ppm 1. estimate the LOD and LOQ.

standard solutions with a higher concentration were made and new calculations were made for the new series of data (without measurements for the solution with the lowest concentration). a Residual standard deviation. . r LOD (SDxy) LOD (SDa) LOD (mean) LOD = 1.5 ppm 3. SDxy Standard deviation.8627 157 Graph: 4500 4000 3500 3000 Signal 2500 2000 1500 1000 500 0 0 0.5 10 ⋅ LOD > cmin LOD < cmin Conclusion: Because the concentration of a solution with the lowest concentration is lower than the calculated LOD. n Slope. ppm 3 3.5 2 2.5 Concentration.© 2009 by Taylor & Francis Group.2 ppm 1. SDa Regression coefficient. LLC .8 ppm 1.Method Validation Solution: Number of results. SDb Standard deviation.5 1 y = 831x + 254 1. b Intercept.3 ⋅ SD b 18 831 254 447 122 303 0.

5 2. ppm 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Solution (2): Number of results. n Slope. r LOD (SDxy) LOD (SDa) LOD (mean) 1. SDa Regression coefficient.158 Data (2): results: Quality Assurance and Quality Control Concentration.4 ppm 2.3 3.7 4.5 2.7 4.5 2.3 3.7 4.7 4. LLC .3 3. SDxy Standard deviation. b Intercept.7 4.5 3.3 ppm 1.5 2. SDb Standard deviation.5 2.3 4.© 2009 by Taylor & Francis Group. a Residual standard deviation.3 3.4 ppm 1.3 3.9132 .7 Signal 1950 1630 2200 1650 2000 1980 2900 3200 3245 2850 3500 3890 3640 4650 3860 4750 4450 4025 18 1016 −425 437 113 410 0.

0 444 450 470 400 445 450 470 30.0 635 650 660 620 610 625 615 40. Assume the maximum value of the coefficient of variation to be CV = 5%. ppm 4 4. calculated LOD is correct. ppm 5.© 2009 by Taylor & Francis Group.0 1000 990 995 1010 1005 1015 995 Signals . Data: results: Concentration. Present LOD in units of the standard solution concentration applied for LOD determinations.5 2 2. Draw an appropriate graph.0 800 810 805 825 820 840 830 50. Excel file: exampl_valid10.11 Problem: Using the values of the analyte concentration determinations for standard solution samples.5 Concentration. LLC .0 198 177 232 200 205 193 235 20. estimate the LOQ and then the LOD using an LOQ determination method based on the assumed value of determination precision.5 3 3.Method Validation Graph (2): 5000 4500 4000 3500 Signal 3000 2500 2000 1500 1000 500 0 0 0.0 1 2 3 4 5 6 7 104 144 124 124 102 111 121 10.5 1 1.5 5 y = 1016x – 425 159 Conclusion (2): Calculated LOD is lower than the lower concentrated standard solution used for its determination and 10 times LOD is higher than lower concentrated standard solution.xls Example 7.

0 10.0% CV (signal) 8.xls 7.12 Problem: Determine the calibration curve based on analyte concentration determinations in eight standard solutions samples (five independent measurements for each solution).2. Prepare an appropriate graph.0% 12.0% 0 10 CV. % 12. ppm 5.0% 0. it is described as an interval between the LOQ and the highest analyte concentration for which a measuring system shows an increase in the output signal.4 Range Determination of linearity and the LOQ enables the determination of a measuring range for an analytical method.0 20. LLC .© 2009 by Taylor & Francis Group.2 5.0 40.3 ppm 20 22 30 40 Concentration 50 60 Excel file: exampl_valid11.160 Solution: Quality Assurance and Quality Control Concentration.9 1.2 10.0% 10.8 0.0% 5% 4.0 LOQ LOD Graph: 14.0 50. . Calculate the regression parameters of the calibration curve.9 22 ppm 7.0 30.0% 2. A measuring range is a range of values (analyte concentrations) in which the error of a measuring instrument is below the assumed value.0% 6. In practice. Example 7.2 2.

75 3. ppb 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 0.Method Validation 161 Using the determinations for standard solution samples for three lowest concentration levels.12 1.65 0.44 2.12 1.8 7.12 2.75 3.12 1.44 2.4 10.25 5.44 2.25 5. Present the measuring range of the analytical method.65 1.8 7.8 7.4 Signal 780 745 756 770 735 1420 1450 1425 1350 1411 3100 3005 3000 3100 3105 4700 4650 4850 4760 4690 6750 6800 7100 6690 6990 10100 10000 9900 10350 10150 13400 13200 . LLC .75 3. Present the LOD in units of standard solution concentration applied in LOD estimation. comparing the calculated value with the analyte concentration in the standard solution with the lowest concentration.65 0.65 0.12 1. estimate the LOD and LOQ using a technique based on using parameters of the calibration curve.25 5.65 0. Also check the correctness of LOD determination.25 5.25 7.75 5.8 7.© 2009 by Taylor & Francis Group.44 3. Data: results: Concentration.8 10.75 3.44 2.

3 ⋅ SD b LOD = . SDxy Standard deviation. r LOD (SDxy) LOD (SDa) LOD (mean) LOQ Range 40 1256 59. SDa Regression coefficient. a Residual standard deviation.27 ppb 0. SDb Standard deviation. SDa Regression coefficient.3 Quality Assurance and Quality Control 13300 13000 12950 16600 16745 16600 16200 16500 Number of results.8 15 24.089 ppb 0.3 13.3 13.2 44.27–13.3 13. SDxy Standard deviation.3 200 7 52.© 2009 by Taylor & Francis Group. r Solution (LOD): Number of results. n Slope.162 33 34 35 36 37 38 39 40 Solution (calibration): 10. SDb Standard deviation.9993 15 1280 −52.3 0.3 ppb 3.4 10.063 ppb 0. LLC . b Intercept.4 13.4 10. b Intercept.3 13. n Slope.1 0. a Residual standard deviation.115 ppb 0.9991 0.

Method Validation Graph (calibration):
18000 16000 14000 12000 Signal 10000 8000 6000 4000 2000 0 0 2 4 6 8 10 12 14 y = 1256x + 59.3

163

Concentration, ppb

Graph (LOD):
3500 3000 2500 Signal 2000 1500 1000 500 0 0 0.5 1 1.5 Concentration, ppb 2 2.5 3 y = 1280x – 52.2

10 ⋅ LOD > cmin LOD < cmin

Conclusion: Calculated LOD is lower than the lower concentrated standard solution used for its determination and 10 times LOD is higher than the lower concentrated standard solution; calculated LOD is correct. Excel file: exampl_valid12.xls

.© 2009 by Taylor & Francis Group, LLC

164

Quality Assurance and Quality Control

7.2.5 Sensitivity
Sensitivity is a parameter that is not a necessary parameter in the validation of an analytical method. One can determine its value based simply on the parameters of the calibration curve. Sensitivity is the relationship of change in the output signal of a measuring instrument to the change in the analyte concentration that induces it. Thus, sensitivity shows the smallest difference in the analyte concentration that can be ascertained using a specific method (it is a slope of a calibration graph: signal in the concentration function). As a recapitulation, Figure 7.2 presents the interpretation of linearity, measuring range, LOD, LOQ, and sensitivity [28].

7.2.6 Precision
Each of the parameters below is determined based on the calculated standard deviation for the series of measurements, and therefore the manner of conduct in their determination will be described together. Repeatability, intermediate precision, and reproducibility can be determined based on the determined standard deviation, relative standard deviation, or the socalled coefficient of variation. Precision is the closeness of agreement between indications or measured quantity values obtained by replicate measurements on the same or similar objects under specified conditions [26]. It is associated with random errors and is a measure of dispersion or scattering around the mean value, usually expressed by a standard deviation. Repeatability is the measurement precision under a set of repeatability conditions of measurement [26]. The precision of results obtained under the same measurement conditions (a given laboratory, analyst, measuring instrument, reagents, etc.). It is usually expressed by a repeatability standard deviation, variance, relative standard deviation, or coefficient of variation.
Range Linearity

Signal

Slope sensitivity

Intercept LOD LOQ Analyte content

FIGURE 7.2  Interpretation of linearity, measuring range, limit of detection, limit of quantitation, and sensitivity [28].

.© 2009 by Taylor & Francis Group, LLC

Method Validation

165

Intermediate precision is the precision of results obtained in a given laboratory over a long-term process of measuring. Intermediate precision is a more general notion (due to the possibility of changes in the greater number of determination parameters) compared to repeatability. Reproducibility is the precision of results obtained by different analysts in different laboratories using a given measurement method. In determining repeatability, it is recommended for an analysis to be conducted on samples characterized with different analyte concentrations and differing in matrix composition. According to recommendations by the ICH [6, 7], standard deviation can be calculated in one of the following ways: • At least nine independent determinations in the whole measuring range (e.g., three independent determinations for three concentration levels). • Six independent determinations of an analyte in standard samples for the concentration level corresponding to the concentration of a real sample. • Six independent determinations of analytes occurring in three different matrices and for two or three concentration levels. According to EURACHEM recommendations [28], one should perform 10 independent determinations and calculate the standard deviation based on these. The determined method’s repeatability can refer both to (1) a very specific analytical method in which matrix composition is specific and defined (e.g., the method of determining analyte X concentration in matrix Y) and (2) determination methods for a given analyte without specifying matrix composition. In the former case, the standard deviation is calculated based on measurements performed for samples characterized by the same matrix composition. In the latter case, one needs to calculate the standard deviation using the measurements conducted for samples differing in matrix composition. Intermediate precision is a notion with a wider scope than repeatability because its value is influenced by additional parameters such as [2, 3]: • Personal factors — different analysts conducting determinations and instability in the work of a given analyst over a specified period. • Instrumental factors — due to the fact that measurements can be carried out using: • Different measuring instruments from a given laboratory • Standard solutions and reagents coming from different producers, or from different batches • Different accessories, for example, different GC columns, with the same characteristics but from different producers, or from different batches If determining precision uses samples in which analyte concentration is stable, the standard deviation is a sufficient parameter that one may determine precision with. However, in the analysis of samples characterized by different levels of analyte concentration, one should use the relative standard deviation or coefficient of variation.

.© 2009 by Taylor & Francis Group, LLC

166

Quality Assurance and Quality Control

Each of these of two quantities is used to compare repeatability, intermediate precision, or reproducibility. 7.2.6.1 Manners of Estimating the Standard Deviation Determining intermediate precision, repeatability, and reproducibility is based on calculating the standard deviation for the series of obtained measurement results [33–26]. The simplest means of estimating this parameter is by calculating the relative standard deviation or coefficient of variation and comparing (assessment) the obtained values. Frequently, one can find the statement that if a relative standard deviation (RSD) is smaller than a certain determined limit, then using a given method can yield precise results. An estimation of standard deviation can be performed using suitable statistical tests: • With a set point of this parameter — chi square (χ2) test (Section 1.8.4) • With the value obtained from a statistical assessment of the set of results obtained using a reference method — Snedecor’s F test (Section 1.8.5) Sometimes it is necessary to compare the standard deviation for sets of measurement results obtained using more than two methods. If the number of measurements on which the calculation of standard deviations is based is similar for all methods (equinumerous series of measuring), then one can apply the Hartley Fmax test (Section 1.8.6). When the number of results obtained using the compared methods are different, one should compare the calculated standard deviations using Bartlett’s test (Section 1.8.7). If the standard deviations are to be compared for two sets of correlated results, one should use Morgan’s test (Section 1.8.8). Example 7.13
Problem: For the given measurement result series, check (at the significance level of α  =  0.05) if the calculated standard deviation differs statistically significantly from the set value of the standard deviation. Apply the χ2 test. Data: results: 11.0 12.0 12.9 12.0 12.5 12.1 14.2 12.1 17.1 12.1 12.4 15.1 12.3 12.0 10.2 Solution: Number of results, n Standard deviation, SD χ2 χ2 crit (f = 14, α = 0.05) 15 1.677 27.88 23.68 SDo = 1.23

.© 2009 by Taylor & Francis Group, LLC

111 7.05) 7 2.05) if the standard deviation values for both series are statistically significantly different.© 2009 by Taylor & Francis Group.160 10 12 13 14 18 15 17 Series 2 11 11 13 11 13 12 Conclusion: Because F > Fcrit . α = 0. SD n/(n − 1) · SD2 F Fcrit (f1 = 6. there is a statistically significant difference in variance values for the compared series. check (at the significance level of α = 0. Equinumerous series — apply Hartley’s Fmax test.983 1.14 Problem: For the given series of measurement results.xls Example 7. f2 = 5. check (at the significance level of α = 0.85 4. LLC . Excel file: exampl_valid13. Excel file: exampl_valid14. Data: result series: Series 1 1 2 3 4 5 6 7 Solution: Series 1 Number of results. the series differ in precision.95 Series 2 6 0.15 Problem: For the given series of measurement results. Apply Snedecor’s F test.Method Validation 167 Conclusion: Because χ2 > χ2 crit . n Standard deviation.795 9.xls Example 7. there is a statistically significant difference in variance value. .05) if the values of the standard deviation for the given series of results are statistically significantly different.

check (at significance level of α = 0.05) 15 1. Excel file: exampl_valid15.16 Problem: For the given series of measurement results.699 Conclusion: Because Fmax < Fmaxo.017 3.356 2 15 1.506 3 15 2.76 4 15 2. .168 Data: result series: Series 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Solution: 11 12 13 12 13 12 14 12 15 12 12 15 12 12 10 Series 2 13 12 12 15 11 10 13 11 12 14 15 12 14 12 11 Quality Assurance and Quality Control Series 3 10 13 14 12 13 14 11 12 17 14 17 12 11 12 14 Series 4 10 12 16 18 13 14 14 12 17 14 10 12 11 13 15 Series 5 17 11 13 14 13 12 13 11 13 14 15 11 11 12 12 Series 1 Number of results. α = 0.05) if the values of the standard deviation for a given series of results are statistically significantly different.09 4. Not equinumerous — apply the Bartlett test. there is no statistically significant difference in variance values for the compared series.384 5 15 1.© 2009 by Taylor & Francis Group. LLC . f = 14. n Standard deviation.xls Example 7. SD Fmax Fmaxo (k = 5.

257 25.356 1.701 4.22 11. n Standard deviation − SD 1/(n − 1) (n − 1) · log(SD2) (n − 1) · SD2 c SDo2 Q χ2 crit (f = k − 1 = 5.xls .05) 2 3 4 5 6 15 11 14 13 9 12 1.04 3.245 9.Method Validation Data: result series: Series 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 11 12 13 12 13 12 14 12 15 12 12 15 12 12 10 Series 2 13 12 12 15 11 10 13 11 12 14 15 Series 3 10 13 14 12 13 14 11 12 17 14 17 12 11 12 Series 4 10 12 16 18 13 14 14 12 17 14 10 12 11 Series 5 17 11 13 14 13 12 13 11 13 Series 6 10 13 14 12 13 14 11 12 17 14 17 12 169 Solution: Series 1 Number of results.000 76.529 1.© 2009 by Taylor & Francis Group.075 2.250 1.091 3.635 2.077 0.083 0.07 Conclusion: Because Q > χcrit 2. LLC .071 0. α = 0.803 2.095 7.769 26. there is a statistically significant difference in variance values for the compared series.845 51.125 0.137 0.100 0.727 56. Excel file: exampl_valid16.000 50.270 8.733 26.672 4.

Excel file: exampl_valid17.1 9.6 10.4 9. In each series. Apply the Morgan test.440 0.18 Problem: To determine the values of repeatability. .7 8.17 Problem: For the given series of measurement results–dependent variables. LLC .1 10.9 9.1 9. Using the obtained measurement results.1 8.7 8.1 Series 2 9.6 8.201 8. six independent series of measurements were performed for six standard solution samples.2 9.05) if the values of the standard deviation for the given series of results are statistically significantly different.816 1.2 9.9 9.2 9.3 8.xls Example 7.8090 0.7 9.6 Conclusion: Because t < tcrit.9 8.576 2. five repetitions were made.1 8.8 9. check (at the significance level of α = 0.1 9. calculate value repeatability for the analytical method.580 0.8 9.3 9.© 2009 by Taylor & Francis Group. Data: result series: Series 1 1 2 3 4 5 6 7 8 9 10 11 12 13 Solution: r SD1 SD2 L t tcrit 0.170 Quality Assurance and Quality Control Example 7.9 9. there is no statistically significant difference in variance values for the compared series.2 9.

Using the obtained measurement results.9 14.02 Series 3 7.67 2.121 2.8 13.12 5. repeatability should be calculated as a mean value CV for the given series. Series 1 Number of results.532 5. α = 0.© 2009 by Taylor & Francis Group.xls 3.4 Series 6 17. the calculations should use the values of CV and not SD. The first step is to check the homogeneity of variances for individual series of results. It is possible to calculate repeatability as a mean value CV for the given series.Method Validation Data: result series: Series 1 1 2 3 4 5 2.327 2.14 7.3 10.43 2.24 5. LLC . Because series are equinumerous. CV (%) Fmax Fmaxo (k = 6.34 5.19 Problem: To determine the values of repeatability and the intermediate precision of the analytical method. f = 4. six independent series of measurements for the samples were performed for one standard solution.65 2.15 7.8 17. CV repeatability Excel file: exampl_valid18.16 5.9 11.11% Example 7. If variances are not homogeneous.05 5 5 0.3 14.2 17. calculate the values of repeatability and intermediate precision for the analytical method.50 Conclusion: Because Fmax < Fmaxo. If variances are homogeneous.65 4 5 0.54 2.3 17.52 29. n Standard deviation.05) 5 0.0 17.119 1.34 7.34 Series 4 10. there is no statistically significant difference in variance values for the compared series.2 10.76 11.5 171 Solution: Because the levels of analyte concentrations in the investigated standard solutions samples are different. six repetitions were performed.305 1.1 Series 5 14.34 3 5 0. .34 Series 2 5.2 14.28 6 5 0.2 10. SD Coefficient of variation.60 2 5 0. In each series. one should apply the Hartley Fmax test.142 5.09 7. one should reject the deviating value (series) and perform the calculations again.

Because series are equinumerous. α = 0.172 Data: result series: Series 1 1 2 3 4 5 6 101 104 103 101 100 102 Series 2 103 106 102 105 109 104 Series 3 111 107 104 102 110 105 Quality Assurance and Quality Control Series 4 100 102 101 117 115 103 Series 5 103 102 106 103 107 104 Series 6 103 108 102 107 105 103 Solution: The first step is to check the homogeneity of the variances for individual series of results. LLC . n Standard deviation.05) 6 1. repeatability should be calculated as a mean value CV for the given series.472 2 6 2.70 Conclusion: Because Fmax > Fmaxo.© 2009 by Taylor & Francis Group.xls Data: result series: Series 1 1 2 3 4 5 6 101 104 103 101 100 102 Series 2 103 106 102 105 109 104 Series 3 111 107 104 102 110 105 Series 4 – – – – – – Series 5 103 102 106 103 107 104 Series 6 103 108 102 107 105 103 .422 6 6 3. If variances are homogeneous.52 18.507 7. f = 5. SD Fmax Fmaxo (k = 6. one should apply the Hartley Fmax test. one should reject the deviating value (series) and perform the calculations again. If variances are not homogeneous. Excel file: exampl_valid19a. Results from series 4 should be rejected due to lack of homogeneity of variances and calculations should be performed again. there is a statistically significant difference in variance values for the compared series.483 3 4 5 6 1. Series 1 Number of results.581 26.941 6 6 2.

Relationships between trueness. Analysis of these definitions shows that the hitherto existing notion of “accuracy” was replaced by the term “trueness.30 173 Conclusion: Because Fmax < Fmaxo.3 [4. 37–39]: • Sample analysis of suitable certified reference materials • Comparison of the obtained result with a result obtained using a reference (original. final) method [40–42] • Standard addition method . α = 0.422 6 0 3.75 7.941 6 6 2. It is influenced mostly by the bias of the analytical method. Repeatability was calculated as a mean of SD values for individual series. f = 5.483 3 4 5 6 1.xls 2.37 2. precision.472 2 6 2. n Standard deviation.” and the previously applied notion of “accuracy of a single measurement” is now simply “accuracy. the more accurate the result of a single measurement is. LLC . Of course.2. SD Fmax Fmaxo (k = 5. there is no statistically significant difference in variance values for the compared series. Trueness is the closeness of agreement between the average of an infinite number of replicate measured quantity values and a reference quantity value [26]. and accuracy are presented schematically in Figure 7. Intermediate precision is SD.Method Validation Series 1 Number of results. other parameters such as linearity and sensitivity also influence the accuracy of an analytical method. Trueness and accuracy can be determined using different approaches [33. Accuracy is a combination of trueness and precision.© 2009 by Taylor & Francis Group.7 Accuracy and Trueness Accuracy is defined as closeness of agreement between a measured quantity value and a true quantity value of a measurand [26]. The truer and more precise the results obtained using a given method. calculated using all the 30 results. SD repeatability SD intermediate precision Excel file: exampl_valid19b.05) 6 1. 9].68 16.507 – 5.” It is trueness that describes the conformity of results obtained using a given analytical method to real (expected) results.

174 Quality Assurance and Quality Control INCREASING TRUENESS µ µ INCREASING ACCURACY µ µ INCREASING PRECISION FIGURE 7. their influence on measurements varies.© 2009 by Taylor & Francis Group.4 [9]. and accuracy [4. The difference is due to the occurrence of different errors [44].12) ε xi = d xi µx (7. With regard to the manner of presenting a determination result.1 Measurement Errors The notion of accuracy is closely connected with the notion of errors [43]. 7. There are three basic types of errors: • Gross errors • Biases • Random errors The influence of individual types of errors on a measurement result is presented schematically in Figure 7.2. which can be described by the dependence: d xi = x i − µ x • Relative error εx.3  Relationships between trueness.7. precision. 9]. one can distinguish: • Absolute error dx. described by the equation: (7. Depending on the type of errors. LLC . The value of a single measurement result may differ (and actually always differs) from the expected (real) value.13) .

• It appears only in some measurements. one can distinguish: • Methodological errors • Instrumental errors • Human errors The total error of a single measurement result may be divided into three components.4  Influence of individual types of errors on a measurement result [9].Method Validation ∆xsys δxj 175 x2 x1 x3 x4 x5 x6 µx xm xj ∆x1 ∆xm FIGURE 7. • It is a random variable — however. and therefore to eliminate. Gross error is characterized by the following properties: • It is the result of a single influence of a cause acting temporarily. With regard to the source of errors. there is a high probability of detecting a result(s) with a gross error. one with unknown distribution and an unknown expected value. as described by the following equation [45]: d xi = xi − µ x = ∆x sys + ∆xi + δxi (7. .© 2009 by Taylor & Francis Group.14) where dxi = total error of a measurement result xi = value of a measurement result µx = expected value ∆ xsys = bias ∆ xi = random error δxi = gross error For measurement series (at least three parallel analyte determinations in the same sample). • It is the easiest to detect. LLC .

μ2x are the expected values for two standard samples. • The cause of its occurrence can be. Figure 7. often described as “outliers” [9]. There are many known ways of detecting results with gross errors.5 presents specific methods of bias determination [9]. a mistake in instrument reading or a mistake in calculations. x2m are the mean values determined for standard samples. often described as “outliers” [9]. After eliminating results with gross errors. Constant bias asys is determined according to the formula: asys = µ1x x1m − µ 2 x x 2 m µ1 x − µ 2 x (7.5  Selection criteria for a suitable manner of action in detecting and rejecting results with gross errors. Methods of gross error determination are described in Chapter 1. and x1m. SERIES OF RESULTS OF INDEPENDENT DETERMINATIONS Known SDg value Unknown SDg value Known Rm value Rcrit Not numerous series Rcrit Numerous series kα Not numerous series Numerous series kα Unbiased series t Biased series wα Q-Dixon FIGURE  7.5 schematically presents the selection criteria for a suitable manner of action in detecting and rejecting results with gross errors. LLC . The determination of biases is one way to determine the trueness of an analytical method. TABLE 7.176 Quality Assurance and Quality Control • It assumes both positive and negative values (unlike bias). for example. Each of them is applied in specific conditions.© 2009 by Taylor & Francis Group. Table 7. . the trueness of the obtained final determination (most often the mean value of the measurement series) is influenced by biases and/or random errors.5 Basic Information Concerning Methods of Bias Determination [9] Bias Type Constant Requirements Samples of two standards (reference materials) with different analyte content Course of Action Series determinations for two standard samples (reference material samples) with different analyte content. using the developed method.15) where μ1x.

The correction multiplier value is determined according to the formula: B= x 2 m (ref ) − x1m (ref ) x 2 m − x1m (7. x mCst are the mean values determined for sample and sample with standard addition. x2m are the mean values determined for the first and second standard with using the developed method. The value of variable bias is determined according to Equation (7. x2m(ref) are the mean values determined for the first and second standard when using the reference method. and xm. and x1m. The value of variable bias is determined according to the equation: bsys = 1− B B (7. The values of constant bias and variable bias are determined according to the formulas:   (a) constant bias a = − asys b (7.18) Constant and variable Series samples with different analyte content Reference method where x1m(ref)1. Series determination for samples with different analyte content with the use of the reference method and the developed method.64).17).63) and (1.19) (continued) .Method Validation Bias Type Requirements Sample and sample with standard addition Course of Action 177 Series determination with the use of the developed method for sample and sample with standard addition. Regression parameters of the regression line Y = b · X + a are determined according to Equations (1. LLC . The relationship between results obtained by the reference method (0Y axis) and results obtained by the developed method (0X axis) is determined.16) Variable where Ct is the expected value increase of analyte concentration due to standard addition.© 2009 by Taylor & Francis Group.17) Variable Samples of two standards (reference materials) Reference method Two series of determination for two standard samples with the use of the reference method and the developed method. The correction multiplier value is determined according to the formula: B= Cst x mCst − x m (7.

© 2009 by Taylor & Francis Group. than s → 0.24) where dxmet = total error of a determination result for the applied analytical method E(xmet) = value of a determination obtained as a result of a given analytical method used (expected value for a given analytical method) µx = expected value (real) Δxsys = bias .20) A determination result (arithmetical mean of a series of parallel measurements) can only have a bias and random error according to the following dependence [45]: d xm = x m − µ x = ∆x sys + ∆x m (7.178 Quality Assurance and Quality Control TABLE 7.23) where dxm = total error of a determination result (arithmetical mean of the series of measurements) xm = mean value of the series of measurement results µx = expected value Δxsys = bias Δxm = random error If the determined bias refers to an analytical method. the following dependence is true [45]: d xmet = E ( x met ) − µ x = ∆x sys (7. LLC .21) asys = − a b (7. the random error is negligibly small with relation to the bias when n → ∞. In this case. then with a large number of conducted measurements.5 Basic Information Concerning Methods of Bias Determination [9] (continued) Bias Type Requirements Course of Action therefore   (b) variable bias   therefore bsys = 1 −1 b (7.22) b = B (7.

it is not always possible to use reference material samples precisely satisfying given needs. In each case.9) for the significance of differences between two results. LLC .8. when the result of the Snedecor F test application is negative (standard deviations for the series of measurements obtained by the compared analytical methods differ statistically and significantly). samples of the certified reference material) using the investigated analytical method. one should prepare a standard solution by adding a strictly specific quantity of analyte into the investigated sample and subject it to determination.8. According to the general definition. One may differentiate between two types of bias: • A constant bias. In case of its inaccessibility.25) Assuming that the value of a random error is negligibly small compared to the bias value.© 2009 by Taylor & Francis Group. Another manner (most often applied) to determine the trueness or accuracy is the analysis of a reference material sample (or better still. one can present the following dependence: x m = µ x + ∆x sys = µ x + asys + bsys µ x = asys + (1 + bsys )µ x (7. One of them is comparing the obtained measurement value with the value obtained resulting from a method of reference for which the obtained results are treated as accurate. can a result have a random error. whose value depends (most often linearly) on analyte concentration levels — bsysµx Bias is described by the dependence: ∆x sys = asys + bsys µ x (7. In this case. whose value is not relative to analyte concentration levels — asys • A variable bias. this test can only be applied when the compared methods do not differ in a statistically significant manner with respect to precision (Snedecor’s F test.11). Its value influences the precision of the obtained results. one should perform independent determinations for a blind sample and correct the . However.8. Of course. the bias of an analytical method is determined. one can compare both results visually.8. The occurrence of bias makes a given series of measurement (analytical method) results differ from the expected value by a constant value — hence they are either overstated or understated.10) or Aspin and Welch test (Section 1.Method Validation 179 In this manner. but it is more metrologically correct to use Student’s t test (Section 1. one may use for “poor” (small) result series the “approximate test” of Cochran’s C and Cox test (Section 1.5). however. Of course. Trueness or accuracy can be determined using different techniques [37–39].26) Only after rejecting results with a gross error and determining biases (regarding their values and correcting the determination result). reference material is characterized by a constant and strictly defined analyte concentration and with a known concentration determination uncertainty [26]. Section 1.

8. The inference is as follows: if the interval of a determined ratio ± the uncertainty of its determination (R ± U ) includes 1.29) 2 u(2xi ) + u(2xref ) (7. To test if the obtained measurement value does not differ in a statistically significant manner from the certified value (expected value). One determines the ratio of the obtained means (if the values did not differ between themselves. Using obtained values.27) and then the uncertainty U. one should apply Student’s t test (Section 1. LLC . one should calculate the value of the R relation according to the formula: R= x1m x2m (7. An insignificant difference between two obtained results may also be tested by using the method of calculating the ratio between the obtained results and uncertainties of their determination.180 Quality Assurance and Quality Control result for the sample with the known analyte concentration by the obtained measurement result.28) where U = expanded uncertainty for determined relation k = coverage factor whose value depends on the accepted level of probability (most often 95% for which k = 2) There is also another approach based on the comparison of values calculated from the dependence that can be presented using the following expressions: xi − xref (7. using a dependence described by the equation: U=k  x1m + x 2 m    2   ( SD 2 1 2 + SD2 ) (7.9). one should infer that the compared mean values do not differ in a statistically significant manner. the ratio should be 1) and values of uncertainty for such a determined quantity.30) where xi = value of a determination result xref = reference value u( xi ) = uncertainty of a determination result u( xref ) = uncertainty of a reference value .© 2009 by Taylor & Francis Group.

• When. Data: result series. Apply the confidence interval method. The mean obtained for blank samples is then deduced from the mean obtained for the reference material.20 Problem: In the given series of measurement results. This manner of inference is based on comparing differences between two results with the expanded uncertainty (for k = 2) calculated using the uncertainty for the compared values. the corrected mean obtained for the investigated method is compared with the one obtained by the primary method. after the initial outlier rejection.05. The calculated trueness should be presented as the percentage of recovery of the expected value or as a difference between the mean and the expected value together with the given confidence interval. however. determining trueness should be carried out using at least nine parallel determinations at three different analyte concentration levels (at least three determinations per each level of concentration). in this instance. check if there is a result with a gross error. characterized by a null value of bias. EURACHEM [28] recommends 10 parallel determinations for a blank sample and the same number of determinations for reference material samples. According to recommendations by ICH [6. Assume the value α = 0. LLC . mg/dm3: Data 1 2 3 4 5 8. In this case.8 9. and so the corrected value is compared against the certified value. Example 7.2 9. is as follows: • If the inequality occurs: xi − xref < 2 u(2xi ) + u(2xref ) then the result is deemed to be in conformity with the reference value.5 6. 7]. the following dependence is true: xi − xref ≥ 2 u(2xi ) + u(2xref ) then the result is acknowledged to not be in conformity with the reference value. It is also recommended to perform a series of measurements for the reference material using a so-called primary method.8 7.© 2009 by Taylor & Francis Group.3 .Method Validation 181 Inference.

Excel file: exampl_valid20. without the initial outlier rejection.8 0.77 mg/dm3 0.3 8.8 7. apply the confidence interval method.xls Example 7.2 .5 9.44) mg/dm3 Conclusion: The value xmin lies outside the determined confidence interval — hence it has a gross error.2 2.182 6 7 8 α Solution: xmin xmin + 1 xmax xmax − 1 tcrit Initially.8 9.05.20.77 ± 1.21 Problem: Using the data given in Example 7.5 6.© 2009 by Taylor & Francis Group. mg/dm3: Data 1 2 3 4 5 6 8.8 9. the result xmin was rejected. LLC .3 7.67 mg/dm3 (7. Data: result series. xm SD Quality Assurance and Quality Control 8.2 9.447 8. Assume the value α = 0.1 8.05 6.10 ÷ 10.2 9.59 mg/dm3 n SD n− 2 g = xm ± tcrit g 8.

2 9. apply the Dixon Q test.3 8.59 mg/dm3 Example 7.2 9.05.xls 8.77 mg/dm3 0. mg/dm3: Data 1 2 3 4 5 6 7 8 α 8.8 0.39) mg/dm3 Conclusion: The value xmin lies outside the determined confidence interval — hence it has a gross error.Method Validation 7 8 α Solution: xmin xmax wα xm SD 6. xm SD Excel file: exampl_valid21. It should be rejected and the values of xm and SD should be calculated for the new series of data.93 mg/dm3 (6.22 Problem: Using the data given in Example 7.5 6.8 0.05 183 8.© 2009 by Taylor & Francis Group.46 ± 1.1 8.3 9.20. LLC .8 9.8 7.05 .03 mg/dm3 g = xm ± w α ⋅ SD g 8.1 8.5 1.53 ÷ 10. Data: result series. Assume the value α = 0.87 9.46 mg/dm3 1.

check if there are any results with a gross error.8 .2 13.2 13.23 Problem: In a given series of measurement results.4 13.4 13.3 13.77 mg/dm3 0.9 14. It should be rejected and the values of xm and SD should be calculated for the new series of data.59 mg/dm3 Example 7. R Q1 Qn Qcrit R = xn − x1 Q1 = Qn = x2 − x1 R xn − xn−1 R 8 3.7 14.469 0.20 0. Data: result series.05.7 13.094 0.468 Conclusion: Because Q1 > Qcrit. Assume the value α = 0.© 2009 by Taylor & Francis Group. the value xmin has a gross error. ppm: Data 1 2 3 4 5 6 7 8 9 10 11 12 13 13.184 Solution: Quality Assurance and Quality Control Number of results Range. LLC .4 13.1 13.2 11.xls 8.2 13. Apply the confidence interval method. xm SD Excel file: exampl_valid22.

8 13.7 13. LLC .2 13.6 13.4 .7 13.1 14.9 13.73 ppm 0.4 13.73 ± 1.2 13.8 0.2 13.2 13.19 ppm (12.7 13.1 13.92) ppm Data 1 2 3 4 5 6 7 13.4 13.7 14.65 14.9 14.2 13.05 185 g = xm ± kα ⋅ SD g 13.8 15.Method Validation 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 α Solution: xm SD kα 13.53 ÷ 14.1 14.2 14.2 15.3 13.0 13.2 14.4 13.72 ppm 1.© 2009 by Taylor & Francis Group.

05.8 14.7 14. xm SD Excel file: exampl_valid23. LLC .7 13.24 Problem: Check if there are results with a gross error in a given series of measurement results.4 13.2 13. .2 Outlier Outlier 13. Results of measurements were obtained using a method for which the standard deviation method had been determined.7 14.2 13.7 13. Apply the critical range method.4 13. Assume the value α = 0. After their rejection the values of xm and SD were calculated again.8 Conclusion: Results 10.2 13.xls 13.2 14.1 14.186 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Quality Assurance and Quality Control 13.8 13.© 2009 by Taylor & Francis Group.2 Outlier 13.9 13.0 13.2 13.68 ppm 0.6 13.3 13. 16.1 14. and 17 lie outside the determined confidence interval — hence they have a gross error.36 ppm Example 7.2 14.

39 21.05 4.Method Validation Data: result series. new calculations for the new series should be done. ppb: Data 1 2 3 4 5 6 7 8 9 α SDg Solution: xmin xmin + 1 xmax xmax − 1 z R Rcrit 113 115 134 127 4. ppb: Data 1 2 3 4 5 6 7 8 9 113 125 120 127 115 118 117 – 124 .© 2009 by Taylor & Francis Group.5 187 Rcrit = z ⋅ SDg Conclusion: Because R > Rcrit.8 ppb 113 125 120 127 115 118 117 134 124 0.0 ppb 19. LLC . Data (2): result series. a result xmax is considered to be an outlier.

9 52. Results were obtained using a method for which a standard deviation had been determined before.3 51. Apply the confidence interval method.1 51.0 56. the values of xm and SD could be calculated.3 ppb Conclusion: Because R < Rcrit.7 .xls 121.25 Problem: Check if there is a result with a gross error in a given series of measurement results.05.65 ppb Example 7.1 57.2 55. Data: result series.5 57.2 54.29 14.00 ppb 6.0 ppb 19. xm SD Excel file: exampl_valid24.188 Solution (2): xmin xmin + 1 xmax xmax – 1 z R Rcrit Quality Assurance and Quality Control 113 115 127 125 4.© 2009 by Taylor & Francis Group.1 56. Assume the value α = 0.1 54. LLC .8 53.8 56. ng/g: Data 1 2 3 4 5 6 7 8 9 10 11 12 13 14 55. there are no more outliers in the series.7 53.

Apply the critical range method.0 1.9 189 Result xmin was initially rejected.7 ng/g 1. .© 2009 by Taylor & Francis Group.xls 54.9 ± 3. and the values of xm i SD were calculated again.1 57.2 ng/g (51.5 0.26 Problem: Determinations were made for 25 samples. It has been included in the series. LLC . xm SD Excel file: exampl_valid25.7 51. xm 54.1) ng/g Conclusion: An initially rejected result xmin lies in the determined confidence interval. The confidence interval value was calculated for the new series. Using the data obtained measurement results.6 ÷ 58.3 55.8 ng/g Example 7. check them for the occurrence of outliers.65 54.2 54.Method Validation 15 16 17 α SDg Solution: xmin xmin + 1 xmax xmax – 1 ka 51. performing three parallel determinations per each sample.05 1.9 ng/g n n −1 g = xm ± kα ⋅ SDg g 54.9 57.05. Assume the value α = 0.

56 3.96 Result 3 3.01 3.48 3.23 3.01 3.11 3. ppm: Sample 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Result 1 3.88 3.23 3.01 3.35 3. LLC .45 3.33 0.65 3.11 3.56 3.20 0.41 3.23 3.08 3.33 3.23 3.12 3.44 3.49 3.22 3.© 2009 by Taylor & Francis Group.98 3.45 3.190 Data: result series.41 0.65 3.65 3.32 3.04 3.04 3.12 3.72 3.05 0.33 3.13 3.49 3.82 3.07 3.49 1.22 3.34 0.20 3.74 3.58 Quality Assurance and Quality Control Result 2 3.14 3.33 3.21 3.62 α Rm zα Solution: Sample 1 2 3 4 5 6 7 8 Ri 0.51 3.45 3.11 3.67 3.09 0.67 3.98 3.24 0.12 3.13 3.07 3.01 3.08 0.62 3.41 3.11 3.41 3.04 3.82 3.37 3.60 3.34 0.45 3.33 3.62 Conclusion OK OK OK OK OK OK OK OK .28 3.41 3.62 3.71 3.99 3.52 3.

87 0.97 0.98 0. Excel file: exampl_valid26.59 0.01 10.47 0. Ri > Rcrit results should be rejected as an outliers.07 10.29 0.23 10. A second standard solution was obtained by double dilution of the first standard solution.60 0.13 5.23 Series 2 5.22 10.74 0.28 0.66 0. determine the value of the constant bias asys.11 10.11 5.Method Validation 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 0. LLC .49 0.28 10.04 5.33 5.© 2009 by Taylor & Francis Group.45 5. Using the obtained result series. with seven parallel determinations performed per sample.96 191 Ri = xni − x1i Rcrit = zα ⋅ Rm Conclusion: For series 9 and 21. Data: result series.12 .67 0.41 5.27 Problem: Analyte concentrations were determined in two standard solution samples. ppm: Results Series 1 1 2 3 4 5 6 7 10.61 0.19 0.xls Example 7.54 Rcrit Outlier OK OK OK OK OK OK OK OK OK OK OK Outlier OK OK OK 0.15 0.

0 10.0 Series 2 57. LLC .2 57.9 58.9 33.5 . six parallel measurements were made.0 5. Using the data obtained result series.290 ppm Example 7. determine the value of the variable bias bsys.16 5.8 58.8 34.72 57. Using the calculated value of the correction multiplier.2 33.192 x1st x2st Solution: x1m x 2m k k= asys = asys Excel file: exampl_valid27.85 25.28 Problem: Analyte concentrations were determined in a real sample and in a real sample with the addition of the standard.1 33. Data: result series.4 33.9 Cst Solution: xm xmCst 33.23 2 x1st x2st kx1m − x2m k −1 0.5 58. correct the values obtained for the real sample. ppm: Results Series 1 1 2 3 4 5 6 33.xls Quality Assurance and Quality Control 10.2 56. For each of the samples.© 2009 by Taylor & Francis Group.

Using the calculated value of the correction multiplier.29 Problem: Analyte concentrations were determined in two real samples.75 746 740 753 758 743 750 746 755 Sample 2 x2ref 945 947 956 960 948 955 960 966 Validated Method Sample 1 x1 765 772 758 768 783 749 777 769 Sample 2 x2 967 980 978 984 974 984 975 988 . ppb: Reference Method Sample 1 x1ref 1 2 3 4 5 6 7 8 Solution: x1m(ref) x2m(ref) x1m x 2m 748. For each of the samples.035 34.63 978. LLC . using both methods.xls 1. using an investigated method and the reference method. determine the value of the variable bias bsys.Method Validation Cst xmCst − xm 1− B B 193 B= bsys = xm( corr ) = B ⋅ x B bsys xm(corr) Excel file: exampl_valid28.© 2009 by Taylor & Francis Group.93 Example 7. eight parallel measurements were made. Data: result series. Using the obtained result series. correct the values obtained using the validated method.036 −0.62 767.88 954.

2 35.2 19.2 12.7 86.026 748.9 88.3 109 59.8 56.8 77.83 Excel file: exampl_valid29.3 38.3 47.5 101 79. Data: result series.2 111 Reference Method xref 45. LLC . determine the variable bias bsys and the constant bias asys.4 21.8 89.3 57. and the mean values were presented.9 97. Using the obtained data. ppb: Validated Method x 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 46. For each of the samples three parallel measurements were made using each of the methods.2 26.194 Quality Assurance and Quality Control x2m( ref ) − x1m( ref ) x2m − x1m bsys = 1− B B B= xm( corr ) = B ⋅ x B bsys x1m(corr) x2m(corr) 0.08 953.2 37.6 10.3 90.2 39.© 2009 by Taylor & Francis Group.3 106 .975 0.1 27.9 103 56. Apply the linear regression method.30 Problem: Analyte concentrations were determined in 15 real samples using the validated method and the reference method.2 44.xls Example 7.

However.589 0. The greater the influence of slight changes in parameters of the measurement process on final determination results. the greater the attention one should pay to maintaining these parameters at a stable level. LLC .589 195 a b 1 −1 b –0. and can be estimated based on reproducibility [46.8 Robustness and Ruggedness The robustness of a method is determined to find the influence of slight fluctuations of conditions in a given analytical method on the result of final determination.972 0. 47]. .Method Validation Solution: xref = b ⋅ x + a asys = − bsys = a b (B) asys bsys Graph: 120 100 80 60 40 20 0 0 20 40 60 x 80 100 120 xref xref = 0.0292 Excel file: exampl_valid30. Robustness influences the manner of conducting measurements using a given analytical method. ruggedness (flexibility) is a parameter describing the usefulness of a given analytical method in different conditions.607 0.© 2009 by Taylor & Francis Group.972x – 0. 47].xls 7. It is a parameter concerning changes in internal conditions [46.2.

Problem 1: Determine the selectivity of the CV-AAS method. mercury is released from the analyzed sample. the amalgam is heated to 600°C and the released atomic mercury is directed through the air stream to the absorption cell. Solution: In the case of the cold vapor technique. After this step. .31 General problem: An analytical procedure was developed.7 nm. sent by hollow mercury cathode lamp. and then (after an eventual reduction to atomic mercury). The amalgamation reaction is a selectivity reaction for mercury. The absorption measurement is realized using a characteristic wavelength for mercury. Problem 2: Based on measurement results for the series of standard solutions. LLC . together with a description of its determination. 7. The exact characterization of this parameter. determine the linearity of the method. pH fluctuations. Example 7. one can determine the usefulness of a given analytical method for a given determination. it is trapped on the gold bed as an amalgam.196 Quality Assurance and Quality Control Similarly to the reproducibility of an analytical method. is presented in Chapter 4. but it should be presented in the final method validation report. Conclusion: Such a measurement method guarantees high selectivity for indicating mercury for two reasons: 1. the influence of fluctuations in temperature.g. determining the appropriate validation parameters. 47].© 2009 by Taylor & Francis Group. with a wavelength of 253. 2. in which an absorption measurement is conducted. conditions of chromatographic isolations) [46. indicating the total mercury content in samples of muscle tissue of great cormorant (Phalacroxorac carbo) with the use of atomic absorption spectroscopy (cold vapor technique). its robustness and ruggedness are also determined in interlaboratory studies. These parameters can be calculated based on a study of changes in the standard deviation of the measurement series using a given analytical method.. and slightly fluctuating the parameters of the applied analytical method. Based on the estimated uncertainty value. although the influence of fluctuations from some measurement conditions (in a method subjected to validation) may be conducted in one laboratory (e. Determination of a combined uncertainty for an investigated analytical method (most often expressed as a percentage of the determined value) makes it possible to know the quality of results obtained with a given method.2. changes in purity and types of reagents.9 Uncertainty Uncertainty is not considered a basic validation parameter. The validation process method was conducted.

7 0.9986 .2 169.90 2.6 175.730 −2.3 171. r 25 1.3 66.12 1.1 137.79% Conclusion: There are no statistically significant differences in variation values.61% 60 5 92.03 1. SDxy Standard deviation. LLC .1 66.6 Hg.8 33. % Fmax Fmaxo 20 5 31.620 2. n Signal. with a significance level of α = 0. b Intercept. a Residual standard deviation. Hartley’s Fmax test was applied. the homogeneity of variation for the results of the series being analyzed should be checked. n Slope.20 80 5 129.1 136.40 25.2 138 100 167.5 34. For this.03 2.2 32.79% 40 5 62.781 1.019 1. ng 197 60 99.880 2.Method Validation Data: results: Unit Content of Hg Signal 20 33.1 100.6 80 142. mean Standard deviation.3 0.8 68.© 2009 by Taylor & Francis Group. SD CV.2 170.93% 10.2 95.58 0.xls Due to no statistically significant differences in variation for the compared series.8 140.77% 100 5 158. a calibration curve was constructed and their regression parameters were determined.4 63. SDa Regression coefficient.3 99. Excel file: exampl_valid31_1. SDb Standard deviation.5 98.838 1.290 1.11 2.1 35.1 Solution: Before constructing the calibration curve.9 40 67.05. Content of Hg No results.

and the range.198 Graph: 200 180 160 140 Signal 120 100 80 60 40 20 0 0 20 40 Quality Assurance and Quality Control y = 1. SDa Regression coefficient.73x – 2. r The LOD value was determined using the equation: LOD = LOD (SDxy) LOD (SDa) LOD (mean) 3.023 0.9986 60 80 Hg Content. check the correctness of the LOD determination. In addition. SDxy Standard deviation. requires a high linearity procedure.9987 .62 1.3SD b 2. with the fulfillment of the equal distribution of the standard in the range of the calibration line. Problem 3: Based on the series of measurement results for the standard solutions with the three lowest mercury content levels (20. 40.974 0. r.99 ng 2. determine the LOD values. a Residual standard deviation.11 r = 0. b Intercept. SDb Standard deviation. the LOQ value.45 ng 15 1. Solution: n Slope. LLC .© 2009 by Taylor & Francis Group. ng 100 120 Excel file: exampl_valid31_2.91 ng 1.65 1. and 60 ng).4 0.xls Conclusion: A high value of the regression coefficient.

62x + 1.© 2009 by Taylor & Francis Group. calculate the repeatability.4 ng Whereas the range was presented as: 7. LLC . Based on the relationship: LOQ = 3 · LOD The LOQ value was calculated to be: LOQ = 7.4 ÷ 100 ng Excel file: exampl_valid31_2.65 199 The correctness of LOD determination was made according to the equations: 10 ⋅ LOD > cmin where cmin = 20 ng.Method Validation Graph: 120 100 80 Signal 60 40 20 0 0 10 20 30 40 Hg Content. LOD < cmin Conclusion: The determined LOD value is correct. .xls Problem 4: Based on the series of results for the three real samples (lyophilized muscle tissue of great cormorant). ng 50 60 70 y = 1.

7 22. ppm 3.2 27.82 78.21 2.9 24.3 22. the Dixon Q test was applied (with a significance level α = 0. LLC . mg 1 2 3 4 5 6 7 30.22 Hg Concentration.1 Hg Content.42 66.11 78.05).26 3.25 Solution: Before performing the calculation. ng 64.94 92.20 91.3 24.37 85.79 3. in order to indicate precision.17 2.© 2009 by Taylor & Francis Group.7 20. For this.4 20.9 25.68 Mean Sample 2 Sample Mass.93 84.11 3.21 3.77 Hg Concentration. ng 83.3 35. one should check whether there are no outliers in the measurement results series.8 28.2 37.00 Mean Hg Concentration.77 80.01 79. .62 70.82 72.69 3.25 2.11 2.54 3.4 32.93 72.69 74.33 3.91 81.5 33.87 3.84 71.8 21.9 19.27 2. ppm 2.90 90.50 Mean Sample 3 Sample Mass.41 3.92 3. ppm 3.97 3.15 3.44 2.0 Hg Content. ng 78. mg 1 2 3 4 5 6 7 25.48 81.14 71.200 Quality Assurance and Quality Control Data: Measurement results for individual samples: Sample 1 Sample Mass.61 3. mg 1 2 3 4 5 6 7 21.5 Hg Content.7 31.1 20.09 3.32 2.8 30.

30% No results.xls Determinations were conducted for three different real samples.753 2.678 2. The calculated repeatability value. n Standard deviation.250 0.063 0. Data: results are given as (ng/mg): Data 1 2 3 4 5 2. SD CV. of results.507 Sample 3 7 0. the homogeneity of the variation should be checked for the series of results to be analyzed.116 201 Conclusion: In the series of measurement results.24% Excel file: exampl_valid31_4.05 was chosen).918 .33 0. n Range.162 4.516 2.107 4.xls exampl_valid31_3c. R Q1 Qn Qcrit 7 0.xls Problem 5: Based on the results determined for certified reference material samples (BCR-463 — Tuna fish: total Hg and methylmercury).364 Sample 2 7 0.Method Validation Sample 1 No.32 0. there are no outliers. % Fmax Fmaxo Conclusion: There are no statistically significant differences in variation values. can be calculated as a mean value from the coefficient of variation counted for three series: CVrepeatability = 4. therefore.38 Sample 3 7 0. Excel files: exampl_valid31_3a. LLC .© 2009 by Taylor & Francis Group.118 3.68 8. however.43 0. Sample 1 7 0. before calculating the repeatability value (as the mean of the coefficient variation for the results of the three series results).xls exampl_valid31_3b.182 0.163 0.970 2. determine the trueness value (as a recovery value).67% 1. The Hartley Fmax test was applied with this aim (a significance level of α = 0.76% Sample 2 7 0.

Solution: As the main components of the uncertainty budget.2% Determined Conclusion: Results obtained with the use of the developed method are correct. as well as the uncertainty value from the indication of trueness.202 Value CRM Solution: Mean SD U R U (k = 2) 2. Excel file: exampl_valid31_5.5 0 CRM Trueness = 97.1% 8.5 3 2.164 97.xls Problem 6: Estimate an uncertainty value for the determination results of the total mercury content in real samples.1 ± 8. LLC .767 0.85 Quality Assurance and Quality Control U 0.© 2009 by Taylor & Francis Group.16 k 2 2.184 0. .2% where the expanded uncertainty of the recovery value is calculated in accordance with the equation: U=k (u 2 CRM 2 + udet )  xCRM + xdet    2   Graph: 3. obtained with the use of the elaborated method. the following were recognized: the uncertainty value resulting from the calibration curve. the uncertainty value related to the unrepeatability of the measurement results.5 2 1.5 1 0.

LLC .96 1.96 1. concentration.77 0. ng Hg.0 7 61. Sample 1 No results.10 0.11 1. % utrue.82 3.34 9.29 8.10 4. conducting measurements for the series of standard solutions.80 4. ppm ucal.69 3.73x – 2.9986 Sample 2 7 71.25 1.39 4.22 0.61 0. an approximation of measurement points of the calibration line using line regression) was conducted on the basis of the calibration parameters.21 9.43 0.14 2. % usmpl. Calculations were conducted for minimal weighted masses for each of the analyzed real samples. ppm Usmpl (k = 2). % urep.17 0.9 Sample 3 7 72. % usmpl.11 r = 0.2 .14 0.© 2009 by Taylor & Francis Group. n Minimum Hg content.Method Validation 203 The estimation of the combined uncertainty value was conducted using the relationship: 2 2 2 usmpl = ucal + urep + utrue where usmpl = combined relative standard uncertainty for determined results for the real sample ucal = relative standard uncertainty related to the calibration step urep = relative standard uncertainty related to repeatability of measurement results utrue = relative standard uncertainty related to indicating trueness The determination of the standard uncertainty value related to the calibration step (preparation of the series of standard solutions. ppm Usmpl (k = 2).51 0. ng 80 100 120 y = 1.63 4. % Graph: 200 180 160 140 Signal 120 100 80 60 40 20 0 0 20 40 60 Content.

29 ppm sample 3: 3.21 ppm sample 2: 3.25 ± 0. LLC .).xls exampl_valid31_7. together with precise specification (purity.© 2009 by Taylor & Francis Group. standards. and. in case of laboratory synthesis. quality. 9]: • • • • • • • • • • • • • • • • • • • Subject matter and the purpose of the analytical method (applicability range) Metrological principles Type of the applied analyte(s) and matrix composition List of all reagents. chromatograms and calibration curves Conformity of the determined validation parameters with the assumed limits The uncertainty of a measurement result Criteria that one should fulfill in revalidation Full name of the person who conducted the validation process List of literature used . for example. etc.4 Conclusions Validation of an analytical method should be finished with a final report containing [2.22 ± 0.77 ± 0.xls exampl_valid31_6c. a detailed description of this synthesis) Description of the methods used for testing the purity of the substances used and the quality of standards Safety requirements A plan describing the means of transferring the method from laboratory conditions to routine measurements Parameters of the method A list of critical parameters whose slight fluctuations can significantly influence a final determination result — parameters resulting from determination of the analytical method’s ruggedness List of all types of laboratory instrumentation together with their characteristic features (dimensions. precision class. and reference materials used. producer. block schemes in case of complicated instrument kits Detailed description of the conditions for conducting the analytical method Description of statistical conduct together with the enclosed suitable equations and calculations Description of the method in order to inspect its quality in routine analyses Suitable figures and graphs.34 ppm Excel files: exampl_valid31_6a.xls 7.204 Quality Assurance and Quality Control Conclusions: The estimated expanded uncertainty value for measurement results for real samples does not exceed 10% and allows for the notation of measurement results as follows: sample 1: 2.xls exampl_valid31_6b.

Inc. Therefore.22 μg/ml in 3...85 ± 0. Great cormorants (Phalacrocorax carbo) were used as bioindicators for mercury contamination. Solution: Seabirds are useful bioindicators of coastal and marine pollution. The analytical procedure pertains to the indication of total mercury content (after converting the total mercury content into an atomic form). in this case. 98% (Nacalai Tesque. Belgium) • Deionized water Preparation of Standard Solutions There are various methods available for preparing standard solutions. making them susceptible to bioaccumulation of pollutants. Example 7. The released mercury is detected using the cold atomic absorption method at a wavelength of 253.. The analytical procedure is intended for determining whole mercury content in muscle tissue samples from great cormorants. . Poland) • Nitric acid — suprapure (Merck. and free mercury vapor in the generated gas is collected by a mercury collection agent (gold-coated diatomite particle support) in the form of a gold amalgam. However. IRMM (Geel.05 (POCh. due to their specific feeding habits.3% HCl (Inorganic Ventures. Poland) • CRM: BCR-463: Total and methyl mercury in tuna fish. Marine birds spend a significant portion of their lives in coastal or marine environments and are exposed to a wide range of chemicals. Mercury content is determined in lyophilized muscle tissue of great cormorants. Germany) • Buffer solution. A sample is thermally decomposed.48 ± 0. and long life span. Nippon Instrument Corporation obtained good results using l-cysteine. the following reagents are used: • Mercury standard — MSHG.Method Validation 205 • Recapitulation and conclusions • Confirmation and signature of the person responsible for the test and confirmation of the validation.© 2009 by Taylor & Francis Group. 2.32 Problem: Based on the validation parameters indicated for the analytical procedure in Example 7. During the analytical procedure. Ltd. The mercury collection agent is then heated up to 600°C to release atomic mercury. Measurements of the content of total mercury will be performed using the cold vapor AAS technique.7 nm in the detector’s absorption cell. 100-ppm. mercury is further atomized.31. solution stability degrades with age or due to long storage in a warm place. Japan) • Additive B (Wako Pure Chemical Industries. LLC . pH 7.16 μg/g. create a validation report. wide geographical ranges. standard solutions should be kept in a cool. concentration 100. Inc. Japan) • Additive M (POCh.00 ± 0. Kyoto. dark place. because most occupy higher trophic levels. USA) • l-Cysteine.

bring the total volume to 1. For determining total mercury content in analyzed samples. In particular. using pipettes during the preparation of standard solutions. laboratory coat. Now. wash it with acid. keep in a cool. ensure that any mercury contained is in the form of HgCl2.32-1. It should be noted that any mercury present in reagents or redistilled water should also be taken into consideration when a very dilute solution is prepared. the sample changer (BC-1). carefully wash the flask with acid and ensure that its tap is thoroughly washed. solid. Once samples are in position in BC-1. Some products contain Hg(NO3)2 as a mercury component. LLC . and a personal computer (PC). While ensuring uniformity of the contents in the flask by shaking it well. Before using a new volumetric flask. during the preparation of standard solutions. therefore.206 Quality Assurance and Quality Control Preparation of 0. when any solution of 1-ppm or less is prepared. and gas (optional parts required) samples. respectively. Because Hg(NO3)2 may react with l-cysteine and lose its function as a fixing agent. The Mercury/MA-2000 is a mercury analysis system that can measure mercury in liquid. a standard solution of 10-ppm has been prepared. and 10-ppm or less standard solution should be reprepared after 1 year or 6 months have elapsed. then add water and 2 ml of guaranteed reagent-grade concentrated nitric acid. an automatic mercury analyzer is used. MA-2000 from NIC (Japan). By diluting in a similar manner. a standard solution of any concentration may be prepared.32-1  Mercury MA-2000 analysis system. For storage. Standard Solution Preparation Take 1 ml of 100-ppm solution and dilute it to 10 ml with 0. Protective attire should be worn: safety glasses. 100-ppm standard solution. However. It is acceptable to use commercially available undiluted standard stock solutions (100-ppm or 1000-ppm) of mercury intended for atomic absorptiometry as HgCl2.© 2009 by Taylor & Francis Group.001% l-cysteine solution. do not use standard undiluted Hg(NO3)2 solutions. because of the high temperatures of some of its components. As shown in Figure 7. Any diluted solution. The work should be conducted under a fume hood. .000 ml by adding deionized water. rubber gloves. Care should also be taken while working with the atomic absorption analyzer. the system consists of the mercury analyzer (MA-2). such as ovens heated up to 850°C. Mercury has toxic properties. each of them in turn is automatically transferred to the analyzer FIGURE 7. it is advisable to adhere to procedure guidelines for these types of substances. dark place.001% l-Cysteine Solution Measure 10 mg of l-cysteine and place it in a 1000-ml flask.

gas washing is performed.32-2  Schematic diagram of MA-2000. Homogenized samples should be directly weighed (10–50 ± 0. LLC . A block diagram of the apparatus is presented in Figure 7. to be measured. freeze-dried (lyophilized). In addition. Homogenized samples should be stored in a refrigerator with a temperature of 0–6°C. The PC reads the resulting measurements in the order that the various analyses.1 mg) into precleaned combustion boats and automatically inserted into the Mercury/MA-2000 system (NIC. the additives should be subjected to a heat treatment in a heat treatment furnace at 750°C for at least 3 hours.Method Validation Jacket heater Sensor S2 Sensor S1 207 H2 Sample inlet Fan F1 H3 Sensor S3 VS Activated charcoal filter Absorption cell Fan F2 Fan F3 Sensor S4 Dehumidifying bottle COMMON Valve NO NC Gas washing bottle Hg Lump Pump Flow controller Activated charcoal filter FIGURE 7. Analytical Procedure Carefully separated bird tissues should be immediately deeply frozen. Before use. As a method of removing any substances that could interfere with the measurement.© 2009 by Taylor & Francis Group. can be performed. it is recommended that two types of additives be used: additive B (activated alumina) and additive M (sodium carbonate and calcium hydroxide). To remove any interfering substances that are generated when thermally decomposing a sample. and homogenized.32-2. which would adversely affect measurements. preheating the gold-coated diatomite particle support collection agent allows for the measurement to be done without the influence of any organic components. . Japan). which would be physically absorbed to a certain extent. including statistical calculations. if not done so.

However. once more cover with additive M. which corresponds to 20–100 ng Hg. The minimal mass of the lyophilized tissue samples undergoing determination is limited on the one hand by the accuracy of the weight measurement. Using an automatic pipette. The method for using the additives is presented schematically in Figure 7. Additive M Recover the sample with additive M. Calibration Determine the calibration curve as a function of the peak surface area and the mercury content (Hg). Additive B en. For each mercury mass. the maximum sample mass is restricted by the maximum . The sample boats that will be used should also be subjected to the same heat treatment. Sample M B M FIGURE 7.© 2009 by Taylor & Francis Group. cover with additive B. Additive Finally. this value should not be less than 20 mg. Taking this into account.208 Sample Quality Assurance and Quality Control Additive M Put the dispensed sample onto additive M. dose at least five different volumes of the standard solution with a concentration of 1-ppm from the 20.32-3  Method for using the additives. LLC . as well as the level of its homogeneity.32-3. repeat at least three times.to 100μL section.

For each of the solutions. Linearity A series of standard solutions was prepared with a mercury content of 20–100 ng.© 2009 by Taylor & Francis Group. which can be introduced into the ceramic boat and consequently into the furnace.1–5. The next steps of the analytical procedure are schematically presented in Figure 7. 2. substance mass. or the values corresponding to the section of values that most often appear in muscle tissue of great cormorants. the calibration curve corresponds to the range of Hg values in lyophilized tissue 0. the values for the following parameters were indicated. The amalgamation reaction is a selective reaction for mercury. LLC . and based on the obtained results. During the analytical validation procedure. The absorption radiation measurement is realized for mercury’s characteristic wavelength. regression parameters were indicated and . Selectivity Applying the measurement technique ensures high selectivity for indicating mercury for two reasons: 1. Compare these values with the determined values that are contained in the report. This value should not exceed 200 mg. ASA ANALYSIS FINAL DETERMINATION Calculation of Hg content on the basis of calibration curve parameters FIGURE 7.Method Validation 209 Lyophilisation of muscle tissue of great cormorant SAMPLE PREPARATION Weighing of sample directly in ceramic boat Addition of additives M and B Introduction of ceramic boat into MA-2000 instrument and starting of the analysis Analysis parameters: • Concentration level – mode HIGH.32-4. three independent measurements were conducted. Draw the calibration curve and indicate the value of the regression parameters. Taking this into account. • Temperature program – MODE2.32-4  A schematic presentation of the analytical procedure for the determination of total mercury content in muscle tissue of great cormorant samples.

The obtained values are presented in Table 7. and the relationship: LOD = 3. a Residual standard deviation. .11 2. ng 100 120 y = 1. n Slope.730 −2.3SD b A calibration plot is presented in Figure 7.32-5  Calibration curve for linearity determination.32-5.3 0.32-6.73x – 2. A calibration curve was outlined based on the obtained measurement results. LLC . commands a high linear procedure. parameters that determined LOD values. A high regression coefficient.210 Quality Assurance and Quality Control the calibration curve was determined. and 60 ng). r 25 1. and the calibration curve is presented in Figure 7. b Intercept.32-1. TABLE 7. SDxy Standard deviation of the slope.11 r = 0.© 2009 by Taylor & Francis Group.9986 FIGURE 7. Limit of Detection and Quantitation The LOD value is determined based on a series of measurement results for standard solution samples with the three lowest levels of mercury content (20.32-1 Calculated Regression Parameters for Linearity Determination Number of results.7 0.9986 200 180 160 140 Signal 120 100 80 60 40 20 0 0 20 40 60 80 Hg Content. SDb Standard deviation of the intercept. SDa Regression coefficient. r after fulfilling conditions for a “uniform” concentration distribution in terms of the calibration curve. 40.019 1.

Trueness The trueness value is determined based on determination results for certified reference material samples (BCR-463 — Tuna fish: total Hg and methylmercury) and is presented as a recovery value. The LOD value was deemed to be 2. Therefore. the LOQ value was determined to be LOQ = 3 · LOD. ng 50 60 70 y = 1.45 ng. assuming the mass of the sample that underwent indication is an even 20 mg. .12.0 ppm Repeatability Repeatability is determined based on a series of measurement results for three real samples (muscle tissue of great cormorant after lyophilization). The determined trueness value is equal to: 97. corresponds to the mercury concentration in tissue samples of an even 0.65 211 FIGURE 7. to a maximum standard solution concentration used for calibration. However.4 ÷ 100 ng which.24%. A series of five independent determinations are conducted.37). which. LLC .32-6  Calibration curve for LOD determination.4 ng (assuming the mass of the 20 mg sample corresponds to a concentration of 0.Method Validation 120 100 80 Signal 60 40 20 0 0 10 20 30 40 Hg Content. Range The measurement range is a concentration range from the LOQ section. assuming the sample mass that underwent indication of an even 20 mg. corresponds to a mercury concentration of: 0. The determined repeatability value is equal to: CVrepeatability — 4.1 ± 8.37 ÷ 5.32-7.2%.© 2009 by Taylor & Francis Group. This value is determined as an average CV value for three series. it is equal to: 7.62x + 1. equaling 7. The determined trueness value is graphically presented in Figure 7.

The determined value parameters for the calibration curve should not differ by more than ±5% in relation to values determined during the validation process (Table 7. An estimation of the combined uncertainty value is conducted using the calculation: 2 2 2 usmpl = ucal + urep + utrue where usmpl = the combined standard uncertainty from the determination results of real samples ucal = the standard uncertainty from results of determination results from real samples. realization of measurements for the series of standard solutions. as well as the uncertainty value indicating trueness. Uncertainty The main components of the uncertainty budget were the uncertainty value resulting from the determination of the calibration curve. as well as repeatability. . whose value should not exceed CV = 5%.5 1 0. attention should be paid to the stability of the calibration curve.5 0 CRM Quality Assurance and Quality Control Determined FIGURE 7. LLC .© 2009 by Taylor & Francis Group. The calculated uncertainty value for k = 2 equals 9. in relation to the repeatability of measurement results utrue = the standard uncertainty of results related to the determination of trueness. the uncertainty value related to the unrepeatability of measurement results.5 2 1.212 3.32-1).32-7  Comparison of the determined value with a certified Hg content value — trueness determination.5 3 2. Calculations are conducted for minimal masses for each of the analyzed real samples. related to calibration urep = the standard uncertainty for determination results for real samples. indicated based on CRM determinations.1% (as an average of the three samples). The consecutive parameter is trueness. Determination of standard uncertainty related to the calibration step (preparation of a series of standard solutions. an approximation of measurement points of the calibration line with the use of linear regression) is conducted based on calibration parameters. During the revalidation process.

108. D. “Mercury in precipitation and its relation to bioaccumulation in fish: a literature review.” Pure Appl. Chem. intestines.org/vim/VIM_final_fd_13april041. http://www. E. Fukuda. T. 2000. International Conference on Harmonization (ICH) of Technical Requirements for the Registration of Pharmaceuticals for Human Use: Text on Validation of Analytical Procedures. 1994. International Vocabulary of Basic General Terms in Metrology.” Environ. Tanaka. and Tatsukawa. Y... Houserová.L. D.” Environ. 2007.. Geneva. Pollut. Downs. 40. 2005. 2000. http://www.. US Rockville. H.. Kim. International Conference on Harmonization (ICH) of Technical Requirements for the Registration of Pharmaceuticals for Human Use: Validation of Analytical Procedures: Metrology. 1999. S. 185–194. Draft April 2004..” Water Air Soil Pollut. Y. C. “Ecological effects. Saeki.. “Determination of total mercury in muscle. and kidney tissues of cormorant (Phalacrocorax carbo). Kracmar. Pollut. S. Subramanian.. J. Geneva. S. Draft 1.pdf Thompson. 1995... 261–265. 50(2). 1996. P. Anan. liver. transport.. Ambrus. 94.. Matejcek.. J. A. J. 61–68.W. 2005.R. and Lester.. Komar. K. Sitko.. “Harmonized guidelines for single-laboratory validation of methods of analysis. and fate of Mercury: a general review. D. 1335–1351..N.. S. R.iaea. Huber. 503–514. FAO/IAEA Training and Reference Centre for Food and Pesticide Control..© 2009 by Taylor & Francis Group. Kim.. 249–255. United States Pharmacopeial Convention. Med. P. Okabe.com. 1998. and Tatsukawa. Pollut. ICH-Q2B. M. “Specific accumulation of 20 trace elements in great cormorants (Phalacrocorax carbo) from Japan.. and Sitko. and Kuban. and Tanabe. Kubán. “Specific accumulation of mercury and selenium in seabirds.. great crested grebe (Podiceps cristatus). 149–187..G. S..-Y.” Chemosphere.... 3rd edn.. – Czech. Ellison. “Mercury and cadmium in common cormorants (Phalacrocorax carbo).H. 1998. “Total mercury and mercury species in birds and fish in an aquatic ecosystem in the Czech Republic..” Vet.. Saeki. V. and Wood.. Practical Approach to Validation of Methods for Analysis of Residues.. J.. 145.labcompliance. Hedbavny. . United States Pharmacopeia 23. EURACHEM Guide: The Fitness for Purpose of Analytical Methods. Houserová.html. M. Saeki..-Y.. Ikemoto. A. R. 134. K...Y. L. 74.at/programmes/nafa/d5/trcfpc/d5-trcfpc. 1998. Okabe.” Environ. Y. www.. Tanabe.Method Validation Bibliography 213 Boening. E. First Internet version. K. 1996. S. ICH-Q2A. E.. S. Nam.” Environ. V. 108..ncsli. and Eurasian buzzard (Buteo buteo). Macleod..L.. Pollut. 835–855. Tanabe.. 2002.. R. Kim..or. Validation and Qualification in Analytical Laboratories. LLC .

R. high precision.M. Chem.” Crit. ICH-Q2B. 1998. J. and therefore. 1994. 6. Ambrus.. Geneva. Danzer. 2000. No. Vogelgesang.or. Practical Approach to Validation of Methods for Analysis of Residues.” Anal. “The role of and place of method validation in the quality assurance and quality control (QA/QC) system. 20 June 2008..1 ± 8. 2.214 Quality Assurance and Quality Control Analytical Methods Committee.. Marine Environmental Support Office. Conclusions This analytical procedure fulfils requirements for a procedure serving to determine whole mercury content in lyophilized tissue samples from muscle tissue of the great cormorant. The validation report was checked and confirmed by Prof. Traverniers. J. . identification. Johnston. corresponds to a concentration of 0.. “A closer look at analytical signals. The procedure is characterized by high selectivity. and determination: a statistical approach for practitioners. Rev. De Loose. 2004. 4. LLC .at/programmes/nafa/d5/trcfpc/d5-trcfpc.. 1996. Jacek Namies ´nik. P. San Diego.. “Trends in quality in the analytical laboratory: II. 256–259.” Analyst. 1999.K. Huber. and Valente.labcompliance.24%). Results obtained while using this method are characterized by low uncertainty (about 10%). 1996. ICH-Q2A. 6. A. International Conference on Harmonization (ICH) of Technical Requirements for the Registration of Pharmaceuticals for Human Use: Validation of Analytical Procedures: Metrology. and Hädrich. Technical Memorandum 99-01. References 1. 1998. Konieczka. International Conference on Harmonization (ICH) of Technical Requirements for the Registration of Pharmaceuticals for Human Use: Text on Validation of Analytical Procedures. and allows for the discovery of trace amounts of mercury in analyzed samples. M. Specifying and evaluating analytical chemistry quality requirements for ecological risk assessment. 2007. “Limits of detection. 23. 380.html. assuming the minimal mass is an even 20 mg.. The validation process was conducted by Dr.5 ng of total mercury) in the sample. 376–382. Validation and Qualification in Analytical Laboratories. FAO/IAEA Training and Reference Centre for Food and Pesticide Control. www. http://www.. and Van Bockstaele. Chem.. Anal. R. 3. “Mercury/MA-2000. 535–552. Geneva. 242–255. L.iaea. repeatability (CV = 4.2%). Piotr Konieczka. Bioanal.” Trends Anal. Nippon Instruments Corporation. 5. Wisconsin Department of Natural Resources. The estimated LOD value (LOD = 2.” Accred.. Analytical Detection Limit Guidance. 126. Qual. Gdan ´ sk. E.com. Analytical method validation and quality assurance. trueness (recovery = 97. Draft 1. 2004. K. “Measurement of near zero concentration: recording and reporting results that fall close to or below the detection limit.. 7. I. Chem. 2001... 37.12. Mercury Analysis System. NIC-600-2009-04. Assur. 173–190.” Instruction Manual.© 2009 by Taylor & Francis Group..

González. “On the use of the correlation coefficient r for testing the linearity of calibration functions. 2001.. “A procedure to assess linearity by ordinary least squares method. 2004. 2005. 31. and Rubio S. Qual. 64... Konieczka. “Linearity and the limitations of least squares calibration.” Fresenius J. 673–678.T. 300–301. 12. S. K. M. JCGM 200. R.. 14. . Asuero..” Pure Appl. 2004 (in Polish). EURACHEM Guide: The Fitness for Purpose of Analytical Methods.F.. 1997. “The correlation coefficient attacks again.. D. and Sayago. 25. Chem.L.. Fajgelj. 2002. Chem.” Rev.. Qual. Rev. H. Chem..” Accred. Instytut Medycyny Pracy im.. Chem.. 1995. 29. Assur.. 17.” Chem.. 953–962.” Accred. 386–393. 18. 41–59. Van Loco. 10. 23. D.. “Comparison of detection limits in environmental analysis — is it possible? An approach on quality assurance in the lower working range by verification. 46. International vocabulary of metrology — Basic and general concepts and associated terms (VIM). M. 370. “In defense of the correlation coefficient. 2006. 2003. “Harmonized guidelines for single-laboratory validation of methods of analysis. Anal. and Namieśnik. Assur. Wisconsin Department of Natural Resources. R. Bioanal. A.. 2006...© 2009 by Taylor & Francis Group. 726...” Crit. H.A.” Pure Appl. and Wardencki. Valcárcel..” Pure Appl. Nofera. Inż. Hibbert. 70. Asuero. W. 19. “The application of single drop extraction technique for chromatographic determination of solvent residues in edible oils and pharmaceutical products. A. 119–124. Roum.B. M. Chromatogr. Vesseman. Namieśnik. R. 256–258. 2001. 2002. A. A. 30. 73–82. J.G. W.. eds.V. (red).G. 7.. L.B.. 2006.R. LLC . 191–197. 10. 2001.W.. and Currie.” Accred. “Quality of analytical results.. J. Gómez-Hens. Joint Committee for Guides in Metrology. Ellison. Zapewnienie jakości analiz chemicznych. Acta. Warszawa. 2001.Á. United States Pharmacopeia 23. Assur. Geiß. 762. 36. J. R. 1381–1386.” Accred. S.. 1998.” Anal. 15. Assur. “Linearity of calibration curves: use and misuse of the correlation coefficient. A. 16. and Einax. K. A.. De Souza. 22. Lindner. 1060–1070. 21. Michulec. 2004... Łódź..” Trends Anal. Qual.. 13.. 146–152. 1996. 281–285. 26. J. 1998. “Sposoby wyznaczania granicy wykrywalności i oznaczalności. M. Chem...” J..G. Kontrola i zapewnienie jakości wyników pomiarów analitycznych.L.. 2005. Qual. Ellison. D. T. “Further comments on the (mis-)use of r for testing the linearity of calibration functions.. 11. J. US Rockville. 1071.. Burns. 835–855.” Accred.. 2003 (in Polish). Anal.. W. 24.. and Beernaert. Chem. WNT. 11. P.R. J.C.G. J. 2006. and Hibbert.G. M. Danzer. Prof. W. 74.. “Selectivity in analytical chemistry revisited. Mulholland. “Quantifying selectivity: a statistical approach for chromatography. A. Danzer. Herrador. Assur. Chim.. 73. “Selectivity in analytical chemistry (IUPAC Recommendations 2001). Michulec. 552. 10. Elskens.. and Junqueira. 20. Ekol. Dobecki. P.. “The correlation coefficient: an overview. and Wardencki.I. 9. 9.. Croux. First Internet version. S. S.. 993–1014... 639–654. 27. 20. and Górecki. and Müller.. 11.. M.Method Validation 215 8... C. Thompson... Chem. Konieczka. 25–35. Kapeller. “Guidelines for calibration in analytical chemistry. “Development of headspace solid-phase microextraction–gas chromatography method for the determination of solvent residues in edible oils and pharmaceuticals. Sayago.. 28.” Anal. Stefan. Qual. A. United States Pharmacopeial Convention. and Wood.. Chromatogr. 2007 (in Polish). M. Chim. and González.” Chromatographia... M. A. Van Staden. 2008. A. Analytical Detection Limit Guidance.” J.. 377. Huber.

38. Lett. 57–69. B. Hibbert. 2005. “An exact test for analytical bias detection. 71. A. 1158. Hulanicki. and Namieśnik. 1994. Assur. Przyk.” Talanta... Romero. Galea-Rojas. 2007. 2001. ISO 5725-2. 38. 1158. 138–157. 2005. H. Accuracy (trueness and precision) of measurement methods and results — Part 3: Intermediate measures of the precision of standard measurement method.216 Quality Assurance and Quality Control 32. Accuracy (trueness and precision) of measurement methods and results — Part 4: Basic method for the determination of the trueness of a standard measurement method.. Qual.. 47. P.© 2009 by Taylor & Francis Group. “Primary methods of measurement in chemical analysis. M.” Anal. ISO 5725-6. ISO 5725-3. 375–381. Accuracy (trueness and precision) of measurement methods and results — Part 5: Alternative methods for the determination of the precision of a standard measurement method. 35. Chim.. Danko. LLC . Z. “The role of the robustness/ruggedness and inertia studies in research and development of analytical processes. 1994.. Cuadros-Rodríguez... R. Warszawa. Świtaj-Zawadka. 1994. 43. 353–376. H. 34. “Systematic errors in analytical chemistry. Hulanicki. 529–536.. Y. 1979 (in Polish). Acta. Rev. . ISO 5725–4. 1997. W. M. “Absolute methods in analytical chemistry. ISO 5725-5. Warszawa.: Analiza ilościowa związków organicznych.. 2. 44. R. A. J. 45. Konieczka. Anal. Richter.. and Vander Heyden.. 1995. Accuracy (trueness and precision) of measurement methods and results — Part 2: Basic method for the determination of repeatability and reproducibility of a standard measurement method.B.” J. “Ruggedness and robustness testing.V. 67(11)...” Pure Appl. 25–32. 41. Dejaegher. Accuracy (trueness and precision) of measurement methods and results — Part 1: General principles and definitions. 35. 2007.. ISO 5725-1. L. Chem. Accuracy (trueness and precision) of measurement methods and results — Part 6: Use in practice accuracy values. 538.. Kozłowski.. 2001 (in Polish). A.. Anal. 40. 42. 46.. Bolfarine.. “Statystyczne kryteria oceny wyników i metod analitycznych.. Chromatogr. PWN. Polkowska-Motrenko. E. and Samczyński. and De Castilho.” Crit.” J. D. Chem. and Bosque-Sendra. 1905–1911. Współczesna chemia analityczna. E. 1994. 39. 33. A. 1994. Chromatogr.. “Calibration in metrological approach. 354–359. “RNAA in metrology: a highly accurate (definitive) method.” in Bobrański B.” Accred. De Castro. Dybczyński. B.. A. 36.. 37.. PWN. J. 1994. 2005.M. M..

Student’s t Test f 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 22 24 26 28 30 35 40 45 50 60 70 80 100 ∞ α = 0.05 12.106 3.262 2.750 2.250 3.101 2.690 2.064 2.947 2.447 2.878 2.921 2.014 2.660 2.009 2.648 2.032 3.678 2.179 2.977 2.499 3.626 2.925 5.048 2.01 63.021 2.074 2.990 1.716 2.012 2.056 2.984 1.779 2.706 4.093 2.228 2.841 4.055 3.776 2.030 2.149 2.604 4.819 2.160 2.110 2.355 3.707 3.763 2.576 217 .960 α = 0.706 2.169 3.365 2.994 1.898 2.797 2.639 2.182 2.131 2.© 2009 by Taylor & Francis Group.201 2.120 2.086 2. LLC .306 2.571 2.000 1.567 9.Appendix Table A.042 2.861 2.845 2.1 Critical Values.303 3.

944 1.945 1.928 1.492 2.903 1.715 1.955 1.440 2.870 1.208 2.848 1.645 1.941 1.542 2.447 2.948 1.956 1.142 2.294 2.409 1.920 1.923 1.931 1.399 2.895 1.01 1.933 1.05 1.432 2.916 1.576 .256 2.936 1.757 1.414 1.385 2.953 1.454 2.368 2.951 1.960 α = 0.885 1.537 2.2 Critical Values of Parameter wα f 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 22 24 26 28 30 35 40 45 50 60 70 80 100 ∞ α = 0.498 2.926 1.051 2.© 2009 by Taylor & Francis Group.487 2.412 2.423 2.470 2.910 1.937 1.940 1.524 2.324 2.935 1.918 2.348 2.479 2.547 2.814 1.553 2.529 2.943 1.950 1.949 1.218 Appendix Table A.954 1.509 2.518 2.460 2. LLC .

30 4.66 α = 0.96 1.00 4.05 n f 1 5 10 15 20 30 40 60 120 ∞ 2 18.482 0.74 4.1 6.69 3.73 4.05 2.44 4.24 4.557 0.679 0.63 5 37.468 0.0 4.555 0.4 6.560 0.16 4.46 5.90 4.23 2.698 0.30 4.399 0.64 4.67 4.91 4.370 0.65 4.96 3.03 7 43.52 4.31 5.91 4.01 2.1 6.780 0.0 7.80 2.8 5.77 3 27.36 3.12 4.48 4.39 4.33 4.46 4.56 4.60 5.6 7.988 0.65 4.40 3.72 4.60 4.79 3.36 4.08 4.63 4.64 3.17 8 45.29 9 47.03 4.3 Critical Values of z Parameter for Significance Level α = 0.434 0.20 5.60 3.44 3.83 2.637 0.40 5.37 4.76 1.4 6.98 Table A.349 α = 0.55 12 53.99 5.31 4 32.889 0. LLC .57 1.46 1.10 0.98 3.80 5.78 4.89 2.10 4.507 0.1 5.95 2.10 4.83 5.82 4.71 1.05 0.941 0.84 3.72 5.15 3.86 6 40.62 4.4 Critical Values of Parameter zα n 2 3 4 5 α = 0.67 3.32 5.31 4.60 4.10 2.86 2.49 3.45 4.5 Critical Values (Qcrit) of Dixon’s Q Test f 3 4 5 6 7 8 9 10 α = 0.39 10 49.92 3.4 6.20 5.01 4.527 .58 5.14 1.94 4.437 0.77 4.886 0.88 3.83 4.92 4.04 3.01 0.23 4.01 3.81 4.© 2009 by Taylor & Francis Group.72 4.58 3.22 4.Appendix 219 Table A.11 4.0 3.642 0.590 0.17 5.43 2.74 3.08 3.412 α = 0.06 1.62 Table A.47 11 50.765 0.55 4.33 5.50 α = 0.23 4.

377 0.374 0.483 0.647 0.627 0.502 0.462 0.611 0.502 0.6 Critical Values (Qcrit) of Dixon’s Q Test (Modification for n ≤ 40) f 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 α = 0.670 0.384 0.467 0.451 0.605 0.535 0.970 0.565 0.459 0.220 Appendix Table A.628 0.672 0.412 0.454 0.442 0.423 0.569 0.393 0.821 0.564 0.544 0.546 0.635 0.477 0.371 α = 0.926 0.680 0.740 0.478 0.608 0.407 0.489 0.436 0.468 0.529 0.446 0.517 0.555 0.594 0.501 0.417 0.479 0.586 0.388 0.580 0.829 0.489 0.579 0.510 0.567 0. LLC .514 0.717 0.526 0.458 0.994 0.710 0.01 0.397 0.402 0.472 0.© 2009 by Taylor & Francis Group.438 .05 0.697 0.381 0.495 0.429 0.443 0.610 0.530 0.450 0.

68 21.7 Critical Values χ2 Test f 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 α = 0.09 16.17 36.80 36.81 18.41 37.92 18.14 30.99 7.92 35.01 6.58 32.34 13.41 34.68 25.49 11.19 37.00 33.29 41.28 15.67 33.93 40.36 23.65 α = 0.Appendix 221 Table A.84 5.03 22.© 2009 by Taylor & Francis Group.21 24.59 28.07 15.67 23.48 20.87 30.81 9.64 9.69 29.21 11.22 27.09 21.05 3.00 26.98 44.51 16.41 32.14 31.59 14.64 42.31 19.57 38.31 . LLC .30 27.07 12.72 26.

19 3.98 4.65 4.78 4.39 8.63 6.39 5.48 5.69 5.88 10.32 6 19.41 12.05 4.86 4.38 99.46 3.50 6.95 10.54 3.57 3.74 9.21 8.60 6.33 5.88 27.39 4.07 7 19.63 3.55 30.90 4.06 2.59 16.34 5.26 8.76 9.98 5.39 3.29 5.79 7.74 3.82 3.96 14.22 4 19.15 8.19 11.84 3.10 5.01 (Bottom Row) f1 f2 2 3 4 5 6 7 8 9 10 11 2 19.40 99.98 3.18 2.01 4.79 13.46 6.44 6.12 7.84 7.37 5.03 7.59 3.02 4.67 6.01 3.01 9.17 9.37 99.27 5.45 4.55 3.23 5.40 8.00 99.91 3.13 5.39 15.8 Critical Values.18 5.24 6.28 29.31 5.74 10.53 9.06 4.15 4.25 9.22 5.99 3.67 4.52 5.09 14.07 5.78 27.88 8 19.01 28.10 7.63 6.66 4.42 3.62 3.96 4.21 3.05 (Top Row) and α = 0.93 14.58 6.35 8.26 2.27 4.85 3.14 10.64 3.45 4.37 3.33 8.87 7.98 7.92 4.20 5.94 27.16 99.25 99.35 3.87 3.69 6.33 99.13 5.34 8.85 2.97 4.81 27.10 7.62 3.00 5.30 9.47 3.© 2009 by Taylor & Francis Group.28 8.20 3 19.19 3.07 7.63 10 19.71 6.36 99.70 9.36 5.95 4.94 4.23 5.36 8.46 .80 4.54 4.30 99.00 3.09 5.56 3.97 7.68 6. LLC .80 3.21 4.46 8.41 8.26 15.76 27.71 3.97 4.00 14.26 3.10 3.222 Appendix Table A.82 10.45 4.47 3.15 4.67 5 19.84 27.71 6.49 6.78 10.06 7.14 5.06 3.38 8.12 28.74 9 19.99 3.59 6.03 3.39 99.48 6.78 2.95 2.94 18.82 4.04 14.81 6.91 6.02 4.86 6. Snedecor’s F Test for Significance Level α = 0.54 11 19.05 10.16 15.34 6.73 6.55 4.79 3.

42 4.2 20.7 10.19 4.7 9.6 13.37 3.26 1.85 3.7 13.60 7.8 15.5 37.8 9.34 4.37 3.91 2.5 8.00 5.5 13.11 6.00 .Appendix 223 TABLE A.9 33.00 10 550 104 44.40 4.49 3.00 8 403 83.0 15.10 3.6 14.17 1.01 5.11 1.78 8.46 2.30 1.3 9.92 4. Hartley’s Fmax Test for Significance Level α = 0.95 3.5 10.03 3.68 3.1 24.9 41.95 7.1 9.70 8.87 5.5 18.02 2.66 5.7 15.29 2.76 2.04 1.29 5.7 25.36 2.3 11.54 2.0 11.4 8.33 1.2 16.7 10.18 6.94 3.3 12.00 11 626 114 48.94 6.7 10.80 6.8 15.86 2.5 27.43 4.00 6 266 62.1 9.6 20.59 4.03 7.05 k f 2 3 4 5 6 7 8 9 10 15 20 30 60 ∞ 2 39.40 1. LLC .54 2.2 19.31 5.34 4.61 1.8 9.00 5 202 50.00 3 87.77 4.72 2.00 9 475 93.24 3.78 2.22 1.00 4 142 39.67 1.0 29.9 16.96 1.12 2.45 8.7 17.0 28.67 4.15 5.8 8.29 2.5 22.9 Critical Values.12 7.99 4.01 3.3 12.41 7.95 2.00 7 333 72.38 6.85 1.6 26.44 7.07 1.5 18.82 4.4 9.2 10.© 2009 by Taylor & Francis Group.91 8.21 2.5 11.1 12.

85 1.74 1.64 1.79 1.73 1.82 1.72 1.71 1.75 1.72 1.69 1.65 1.86 1.80 1.72 1.68 1.78 1.72 1.71 1.76 1.78 1.72 1.73 1.1 1.3 1.5 1.71 1.73 1.75 1.70 1.67 1.73 1.67 1.64 0.94 1.65 1.74 1.94 1.76 1.80 1.71 1.72 1.81 1.73 1.64 0.82 1.75 1.73 1.80 1.70 1.79 1.66 1.80 1.85 1.85 1.72 1.90 1.94 1.73 1.79 1.74 1.65 1.81 1.79 1.69 1.76 1.76 1.69 1.72 1.73 1.64 1.72 1.90 1.8 1.64 1.64 0.65 1.78 1.70 1.73 1.75 1.86 1.86 1.0 1.73 1.69 1.76 1.69 1.90 1.69 1.73 1.69 1.78 1.79 1.65 1.67 1.64 1.72 1.64 0.64 8 10 15 20 ∞ .66 1.78 1.78 1.72 1.70 1.71 1.75 1.72 1.69 1.69 1.73 1.76 1.86 1.76 1.69 1.65 1.73 1.70 1.69 1.76 1.72 1.75 1.71 1.90 1.70 1.85 1.82 1.71 1.73 1.65 1.94 1.75 1.72 1.76 1.72 1.70 1.66 1.67 1.76 1.76 1.76 1.71 1.86 1.65 1.82 1.70 1.64 0.75 1.82 1.82 1.73 1.82 1.90 1.64 1.76 1.79 1.64 1.81 1.71 1.70 1.76 1.70 1.79 1.70 1.73 1.72 1.65 1.76 1.68 1.72 1.85 1.75 1.76 1.82 1.94 1.65 1.65 1.81 1.78 1.81 1.71 1.78 1.70 1.65 1.80 1.70 1.64 1.70 1.76 1. LLC .79 1.78 1.76 1.73 1.65 1.94 1.67 1.90 1.64 0.71 1.81 1.73 1.71 1.72 1.69 1.66 1.73 1.67 1.76 1.73 1.81 1.72 1.94 1.76 1.70 1.65 1.70 1.67 1.72 1.86 1.70 1.73 1.73 1.68 1.65 1.90 1.72 1.05 c f1 6 f2 6 8 10 15 20 ∞ 6 8 10 15 20 ∞ 6 8 10 15 20 ∞ 6 8 10 15 20 ∞ 6 8 10 15 20 ∞ 6 8 10 15 20 ∞ 0.64 0.81 1.68 1.71 1.80 1.66 1.94 1.73 1.70 1.69 1.76 1.78 1.80 1.74 1.81 1.72 1.73 1.82 1.70 1.71 1.72 1.70 1.66 1.74 1.86 1.71 1.90 1.78 1.76 1.73 1.76 1.76 1.71 1.6 1.70 1.73 1.70 1.70 1.94 1.73 1.76 1.73 1.81 1.72 1.90 1.72 1.68 1.70 1.74 1.64 1.68 1.70 1.66 1.75 1.85 1.71 1.64 0.73 1.82 1.71 1.65 1.78 1.65 1.73 1.65 1.85 1.64 1.70 1.71 1.9 1.75 1.86 1.69 1.94 1.224 Appendix TABLE A.72 1.71 1.79 1.64 1.85 1.79 1.65 1.71 1.66 1.85 1.69 1.76 1.73 1.65 1.94 1.71 1.90 1.73 1.73 1.64 0.71 1.70 1.66 1.71 1.94 1.72 1.70 1.73 1.2 1.71 1.76 1.65 1.70 1.68 1.71 1.0 1.76 1.81 1.4 1.72 1.76 1.71 1.69 1.72 1.76 1.80 1.70 1.71 1.67 1.73 1.85 1.72 1.80 1.71 1.69 1.76 1.90 1.79 1.86 1.85 1.70 1.70 1.80 1.71 1.72 1.86 1.72 1.85 1.73 1.7 1.76 1.66 1.73 1.71 1.80 1.68 1.79 1.71 1.72 1.© 2009 by Taylor & Francis Group.66 1.72 1.80 1.86 1.76 1.73 1.65 1.68 1.74 1.66 1.73 1.75 1.75 1.10 Critical Values vo of Aspin-Welch Test for Significance Level α = 0.76 1.81 1.71 1.71 1.64 1.69 1.72 1.82 1.70 1.90 1.82 1.75 1.72 1.86 1.69 1.

111 0.212 0.05 0.131 0.514 0.387 0.463 0.794 0.191 0.588 0.906 0.097 p = number of laboratories.291 0.189 0.114 α= 0.788 0.138 0.01 – 0.119 0.202 0.228 0.251 0.164 0.274 0.182 0.159 0.220 0.05 – 0.185 0.147 0.276 0.360 0.261 0.172 0.124 0.575 0.793 0.131 0.325 0.541 0.155 0.179 0.438 0.164 0.249 0.906 0.471 0.144 0.234 0.214 0.276 0.388 0.570 0.121 0.308 0.179 0.389 0.532 0.155 0.197 0.238 0.335 0.235 0.504 0.146 0.140 0.318 0.450 0.155 0.138 0.204 0.598 0.151 0.754 0.261 0.134 0.209 0.481 0.181 0.301 0.288 0.768 0.250 0.653 0.203 0.306 0.354 0.312 0.303 0.229 0.116 0.287 0.168 0.108 0.393 0.154 0.332 0.339 0.358 0.366 0.246 0.331 0.243 0.293 0.316 0.01 0.418 0.124 0.208 0.Appendix 225 Table A. .475 0.165 0.721 0.877 0.137 0.466 0.492 0.281 0.168 0.186 0.377 0.127 0.590 0.135 0.413 0.262 0.209 0.256 0.616 0.259 0.286 0.418 0.151 0.178 0.544 0.193 0.437 0.168 0.174 0.993 0.602 0.291 0.223 0.561 0.242 0.197 0.150 0.219 0.121 0.246 0.871 0.445 0.232 0.230 0.447 0.129 0.05 0.746 0.307 0.270 0.271 0. LLC .392 0.521 0.220 0.170 0.841 0.160 0.255 0.676 0.243 0.140 0.184 0.108 n=6 α= 0.883 0.167 0.198 0.300 0.242 0.995 0.281 0.480 0.198 0.334 0.798 0.363 0.326 0.205 0.553 0.564 0.349 0.403 0.532 0.106 0.173 0.967 0.184 0.127 0.120 0.928 0.154 0.355 0.937 0.478 0.516 0.01 0.134 0.696 0.240 0.197 0.164 0.160 0.322 0.348 0.288 0.939 0.177 0.270 0.391 0.781 0.427 0.213 0.293 0.423 0.220 0.200 0.445 0.664 0.520 0.145 0.204 0.124 0.176 0.209 0.297 0.508 0.343 0.161 0.11 Critical Values of Cochran’s Test n=2 p 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 α= 0.117 0.215 0.431 0.465 0.343 0.© 2009 by Taylor & Francis Group.262 0.629 0.568 0.142 0.150 0.680 0.838 0.101 0.707 0.267 0.05 0.599 0.391 0.626 0.118 0.262 0.347 0.356 0.274 0.781 0.265 0.05 0.883 0.238 0.237 n=3 α= 0.332 0.140 0.01 0.252 0.294 α= 0.624 0.310 0.144 0.300 0.141 0.212 0.172 0.190 0.131 0.684 0.425 0.397 0.230 0.137 0.134 0.633 0.450 0.116 0.325 0.144 0.215 0.305 0.515 0.158 n=4 α= 0.975 0.192 0.262 0.099 0.127 0.638 0.230 0.942 0.235 0.159 0.278 0.332 0.343 0.452 0.164 0.111 0.208 0.159 0.204 0.425 0.248 0.151 α= 0.189 0.372 0.134 0.200 0.330 0.188 0. n = number of results for one level.221 0.496 0.506 0.382 0.218 0.192 α= 0.722 0.402 0.417 0.372 0.573 0.131 0.161 0.175 0.480 0.168 0.407 0.318 0.684 0.280 0.684 0.196 0.308 0.241 0.403 0.480 0.196 0.615 0.288 0.229 0.536 0.157 0.222 0.864 0.352 0.319 0.959 0.149 0.191 0.173 0.727 0.307 0.968 0.224 0.329 0.365 0.179 0.979 0.255 0.392 0.371 0.147 0.113 0.128 α= 0.369 0.166 0.114 0.357 0.316 0.126 n=5 α= 0.834 0.181 0.131 0.304 0.01 0.273 0.185 0.718 0.255 0.172 0.373 0.103 0.434 0.246 0.155 0.

01 1.709 2.5381 0.087 3.1738 0.3603 0.549 2.496 1.4376 0.733 2.3585 0.991 3.852 2.6247 0.2530 0.199 3.5123 0.3822 0.507 2.781 2.764 1.0000 0.236 3.2990 0.05 1.5789 0.5554 0.893 2.0851 0.482 2.699 2.5469 0.973 2.5714 0.822 2.5288 0.0563 0.0116 0.126 2.1448 0.651 2.12 Critical Values of Grubbs’ Test One Greatest and One Smallest p 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Upper α = 0.412 2. LLC .155 1.4985 0.218 3.270 3.1492 0.135 3.2537 0.286 3.4994 0.4638 0.979 2.3367 0.05 – 0.0090 0.938 2.4875 0.© 2009 by Taylor & Francis Group.6101 0.481 1.3927 0.5192 0.155 1.4556 0.031 3.715 1.330 3.5766 0.01 – 0.876 2.5672 0.215 2.343 3.301 3.755 2.4214 0.5856 0.001 3.2767 0.894 2.4391 0.462 2.178 3.4025 0.4510 0.6382 0.968 3.335 2.2280 0.387 2.841 2.356 3.3398 0.253 3.4759 0.924 2.025 3.274 2.5636 0.2213 0.0018 0.585 2.0708 0.908 2.3200 0.226 Appendix Table A.2836 0.1101 0.3761 0.965 2.4234 0.020 2.003 3.0002 0.681 2.0308 0.112 3.802 2.636 2.4857 0.316 3.6023 0.3112 0.060 3.014 3.5470 0.5360 0.6175 0.932 2.4711 0.859 2.5574 0.381 Lower α = 0.620 2.0349 0.6445 p = number of laboratories .5091 0.952 2.1864 0.139 2.036 Two Greatest and Two Smallest Upper α = 0.5941 0.1150 0.564 2.887 2.290 2.157 3.806 2.5245 0.2016 0.4085 0.5862 Lower α = 0.6316 0.369 3.758 2.

57 10 1.60 1.40 2.69 1.36 2.52 1.71 1.10 4 1.77 1.35 2.53 1.65 1.65 1.77 1.69 1.56 1.55 1.46 2.57 1.91 1.53 1.90 1.90 1.55 1.67 1.66 1.90 1.79 1.59 1.61 1.08 2.© 2009 by Taylor & Francis Group.65 1.78 1.53 1.45 1.49 1.13 2.54 1.71 1.58 1.06 2.60 1.90 1.20 2.69 1.49 1.64 1.56 1.07 2.85 1.77 1.09 2.29 2.55 1.76 1.64 1.52 1.53 1.63 1.59 1.51 1. n = number of results for one level.62 1.91 5 1.56 1.62 1.98 2.65 1.81 1.65 8 1.56 1.14 2.49 3 1.01 k n p 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 h 1.60 1.51 1.79 1.44 2.74 1.32 2.94 1.73 1.84 1.70 1.42 2.65 1.64 1.99 2.77 1.36 2.Appendix 227 TABLE A.50 1.60 1.59 1.07 2.79 1.47 2.25 2.58 1. LLC .65 1.89 1.70 1.60 1.56 1.22 2.56 1.60 1.25 2.02 2.55 1.44 2.49 1.55 1.30 2.59 1.55 1.51 1.01 2.48 2.52 1.41 2.91 2.60 1.42 2.66 1.05 2.64 1.43 2.72 1.90 1.85 1.70 1.44 2.54 1.55 1.56 1.43 1.08 2.89 1.47 2.57 1.71 1.69 1.53 1.56 1.60 1.48 1.39 2.71 1.53 1.03 2.08 2.46 1.61 9 1.53 1.42 2.65 1.49 2.52 1.55 1.53 1.50 1.18 2.53 1.65 1.38 2.15 1.79 1.70 1.09 2.56 1.73 1.53 1.46 2.27 2.85 1.88 1.04 2.33 2.08 2.57 1.09 2.57 1.60 1.37 2.09 2.68 1.59 1.53 1.63 1.71 1.48 2.64 1.53 p = number of laboratories.54 1.47 1.79 1.50 1.77 1.41 2.06 2. .43 1.60 1.52 1.90 1.59 1.71 1.60 1.05 2.44 2.44 2.06 2.78 1.71 7 1.78 1.46 1.89 1.39 1.91 1.88 1.05 2.07 2.48 1.47 2.58 1.45 2 1.52 1.63 1.59 1.34 2.74 1.45 2.68 1.71 1.58 1.97 1.13a Parameters h and k of Mandel’s Test for Significance Level α = 0.49 2.79 1.88 1.87 1.78 1.64 1.90 1.67 1.58 1.56 1.39 2.78 1.56 1.39 2.41 1.76 1.79 1.68 1.71 1.75 1.32 2.53 1.53 1.65 1.53 1.00 2.87 1.86 1.70 1.62 1.71 1.51 1.65 1.70 1.76 1.52 1.48 1.57 1.51 1.87 1.82 1.60 1.53 1.79 6 1.45 2.63 1.63 1.

37 1.53 6 1.93 1.© 2009 by Taylor & Francis Group.93 1.30 1.42 1.38 10 1.47 1.69 1.75 1.70 1.29 1.40 1.36 1.46 1.41 1.44 1.41 1.56 1.71 1.65 1.51 1.42 1.88 1.36 1.37 1.94 1.37 1.53 1.40 1.49 1.78 1.41 1.94 1.43 1.94 1.42 1.36 1.85 1.35 1.43 1.44 1.40 1.38 1.88 1.46 1.35 1.38 1.44 1.71 1.70 1.86 1.92 1.36 1.52 1.59 1.53 1.59 1.44 1. n = number of results for one level.71 1.43 1.43 1.37 1.94 3 1.40 1.40 1.43 1.38 1.38 1.37 1.39 1.93 1.53 1.94 1.47 1.40 1.44 1.46 1.94 1.47 1.66 1.52 1.34 1.38 1.32 1.43 1.60 1.47 1.48 1.58 1.60 5 1.62 1.52 1.43 1.48 1.41 1.82 1.48 1.85 1.71 1.91 1.35 1.53 1.70 1.80 1.71 1.34 1.53 1.94 1.41 1.58 1.86 1.53 1.88 1.72 1.40 1.52 1.33 1.44 1.40 1.60 1.47 1.94 1.36 1.92 1.38 1.94 1.37 1.60 1.53 1.72 4 1.54 1.35 1.36 1.81 1.41 1.36 1.87 1.52 1.89 1.59 1.91 1.51 1.36 1. LLC .38 1.93 1.15 1.38 1.42 1.90 1.37 1.38 1.36 1.38 1.38 1.50 1.36 1.47 1.90 1.51 1.35 1.35 1.71 1.59 1.93 1.50 1.53 1.45 1.38 1.53 1.46 1.71 1.71 1.36 1.36 1. .90 1.83 1.76 1.36 1.71 1.36 1.35 1.48 7 1.52 1.31 1.40 1.90 1.91 2 1.33 1.60 1.36 1.36 1.60 1.38 1.94 1.58 1.91 1.13b Parameters h and k of Mandel’s Test for Significance Level α = 0.38 1.43 1.59 1.52 1.70 1.36 1.60 1.89 1.59 1.41 1.41 1.47 1.41 9 1.69 1.47 1.71 1.34 1.45 1.36 1.87 1.41 1.40 1.90 1.60 1.36 p = number of laboratories.44 1.44 1.90 1.47 1.39 1.89 1.34 1.05 k n p 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 h 1.50 1.44 1.48 1.36 1.38 1.57 1.50 1.40 1.67 1.38 1.44 1.48 1.45 1.41 1.44 1.60 1.52 1.59 1.48 1.228 Appendix Table A.68 1.52 1.32 1.44 8 1.42 1.42 1.71 1.47 1.71 1.48 1.55 1.57 1.84 1.94 1.38 1.37 1.57 1.68 1.46 1.39 1.43 1.92 1.66 1.39 1.91 1.60 1.40 1.69 1.59 1.41 1.64 1.35 1.

63 0.83 0.05 0.60 0.22 1.99 λα 1.53 0.15 0.71 0.44 0.45 0.25 0.14 Critical Values (λ α ).07 1.49 0.10 0.67 0.50 0.83 0.38 0.Appendix 229 Table A.47 0.20 α = 0.77 0.75 0.28 0.30 0.56 0.57 0.77 0.60 0.25 .14 1.27 0.89 0.22 0.70 0.87 0.40 0.02 0.© 2009 by Taylor & Francis Group.25 0.35 0.52 1.20 0.66 0.90 0.01 0.74 0. Kolmogorov-Smirnov Test α 0.01 0.71 0.30 0.63 1.71 0.58 0.39 0.05 0.50 0.35 0.80 0.64 0.33 0.97 0.42 0.59 0. LLC .36 1.80 0.62 0.54 0.15 Critical Values of Regression Coefficient rcrit f 5 6 7 8 9 10 12 14 16 18 20 25 30 40 50 60 80 100 α = 0.02 0.44 Table A.

Sign up to vote on this title
UsefulNot useful