This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

**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

Piotr.© 2009 by Taylor & Francis Group. II.com and the CRC Press Web site at http://www. Jacek. and are used only for identification and explanation without intent to infringe. Chemical laboratories‑‑Quality control.S. Jacek Namiesnik. LLC CRC Press is an imprint of Taylor & Francis Group. Reasonable efforts have been made to publish reliable data and information. Chemistry.com . CCC is a not‑for‑profit organization that pro‑ vides licenses and registration for a variety of users. 2. For permission to photocopy or use material electronically from this work. cm.copy‑ right.crcpress. now known or hereafter invented. ISBN 978‑1‑4200‑8270‑8 (hardcover : alk. Except as permitted under U. Title. 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. Namiesnik. or other means.4. 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. For organizations that have been granted a photocopy license by the CCC. an Informa business No claim to original U.CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW. QD75. paper) 1.1‑‑dc22 Visit the Taylor & Francis Web site at http://www. including photocopying. Library of Congress Cataloging‑in‑Publication Data Konieczka. I.com (http://www. III. 222 Rosewood Drive. a separate system of payment has been arranged.com/) or contact the Copyright Clearance Center.S. and recording. Inc. ‑‑ (Analytical chemistry) Includes bibliographical references and index. p. Suite 300 Boca Raton. or utilized in any form by any electronic. without written permission from the publishers. transmitted. If any copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint. (CCC). MA 01923. Copyright Law. mechanical. Trademark Notice: Product or corporate names may be trademarks or registered trademarks.taylorandfrancis. Analytic‑‑Qualitative. or in any information storage or retrieval system. Series. FL 33487‑2742 © 2009 by Taylor & Francis Group. Danvers. Quality assurance and quality control in the analytical chemical laboratory : a practical approach / Piotr Konieczka. microfilming.Q34K66 2009 542’. no part of this book may be reprinted. please access www. 978‑750‑8400. LLC 2009000685 . reproduced. but the author and publisher can‑ not assume responsibility for the validity of all materials or the consequences of their use.copyright.

......................................3 Dixon’s Q Test [3...........................8.18 Kolmogorov-Smirnov Test [2................................8..16 En Score [10...2............ix About the Authors...... 29 1.......3 1....................................8 1..10 Linear Regression............4 Measures of Dispersion....................................................................9 Control Charts........2 Critical Range Method [3]....................................9................................. 13]....12 Cochran’s Test [6].. 13 1......................... 4........................................... 36 1.. 4]. 19 1..................................5 1...................................... .. 5]..........8............ 14]..8. 15 1............................. 4].........................15 Z Score [10.....................1 Distributions of Random Variables.......................................................xi List of Abbreviations...............7 1....... 27 1........7 Bartlett’s Test [3].............................. 29 1............. 38 References............1 Characterization of Distributions............................8............6 Hartley’s Fmax Test [3].......Contents Preface.........................11 Significant Digits: Rules of Rounding.................................................... 21 1.....................2 Shewhart Chart Preparation............................................ 16 1...............8...................................................................... 39 1.....14 Hampel’s Test.................5 Measures of Asymmetry... 11]......7 Statistical Hypothesis Testing.....1 1...................................1 Introduction..................8.................1 Shewhart Charts.............. 10 1......................................7 1.............© 2009 by Taylor & Francis Group..........4 Chi Square Test [3]......................................8.. 11]...........13 Grubbs’ Test [6..............11 Aspin-Welch Test [3]....................8.................. xiii Chapter 1 Basic Notions of Statistics............. 17 1........................ 30 1..................8..........................8.................................................................17 Mandel’s Test [6.......................8............8..5 Snedecor’s F Test [3.......... 12 1...........8......... 12.............................................................9.................................................... ..8........................................ 16 1............10 Cochran-Cox Test [3].....2 v ....................................28 1.8......... 27 1............. 23 1.....1 1..........................................................................1 Confidence Interval Method [3].... ......................3 Measures of Location.........................8 Statistical Tests........ 22 1.............. 29 1................................................................................ 7]... 18 1.............1 1....8....................8.................6 Measures of Concentration................... LLC ...............26 1..................................24 1...............................9 Student’s t Test [3.............................. 10 1..................................................................................................8 Morgan’s Test [3].........................

.................................................5 Conclusion...............................................5 Certified Value.............................................................. 57 Methods of Estimating Measurement Uncertainty........3.............................................................3.......................................46 Chapter 3 Traceability.......................................................46 References........................... . 67 4..............3 Definitions [1]........................4 Tools Used for Uncertainty Estimation............ 82 5.......................... 41 2.............................. 51 3.........................................................4 Stability.............. 42 2................................................................................................. 95 References.................................................. 77 Parameters that Characterize RMs.................................. ........................ 67 4.....1 Definitions [1–3]................. 55 Chapter 4 Uncertainty.........................................................................84 5.............7 Conclusion........................................... 41 2..................................................................3.................................................... 69 4...............................1 Procedure for Estimating the Measurement Uncertainty According to Guide to the Expression of Uncertainty in Measurement.................. 49 Introduction.......................................................4 Conclusion.. 77 Definitions [1.........1 4.................................. 57 Introduction...................................................................... 59 4.3...... 73 References..................................................vi Contents Chapter 2 Quality of Analytical Results........................... 74 Chapter 5 Reference Materials.2 Introduction...........................................80 5.................. 85 5.....................................................3...3 Homogeneity.................................3 .....................................................2 3...6 Calibration Uncertainty......................................© 2009 by Taylor & Francis Group..........................2 4........ 58 4................... 55 References. 49 3.......................... LLC ..............................4 Conclusions................. ........................................5 Uncertainty and Confidence Interval............................... 82 5............................... 49 Role of Traceability in Quality Assurance/Quality Control System................................... 57 Definitions [1–3]...............................................................3..................2 5................4 Practical Application of CRM................................................................................................................. 95 5................. 82 5...........80 5........ 41 2..............1 3...................................................... 77 Introduction.........................3 4..................................3 Quality Assurance System.........1 General Information.............. 2]..........................................................1 5..........2 Representativeness...................................................

.........2...................5........................3...................................© 2009 by Taylor & Francis Group. 123 6.....................2..................... 164 7........2..........3................2 Calculation of LOD Based on the Numerical Value of the S/N Ratio..........................2.......... 145 Determinations for Blank Samples............................6 Conclusions... 166 ........................ 129 References...... 146 7.....2 Introduction.....3..............................................1 Manners of Estimating the Standard Deviation..3 6...........................2 Comparison of Measurement Results Obtained in a Two-Level Study (for Two Samples with Various Analyte Concentrations)...................... 164 7...........................................................................Contents vii Chapter 6 Interlaboratory Comparisons..... 101 6........................................................... .......................... 134 7..... 102 6...1 Selectivity........... 130 Chapter 7 Method Validation........................1 6...... LLC ...............2..... ................5 Calculating LOD Based on the Standard Deviation of Signals and the Slope of the Calibration Curve.......... 147 7.6 Precision....................... 144 7............. 143 7....97 Classification of Interlaboratory Studies.........................................................97 Introduction..............2.97 6.................2 Linearity...3 Calculation of LOD Based on .......................... .................2......3............3... 160 7.1 Visual Estimation............. 120 6................4 Graphical Method...............................3 Limit of Detection and Limit of Quantitation........1 7...... 131 7......6...................2..................3.......2..............................3.......................4 Definitions [1.................. 136 7..........................................2 6.................................6 Calculation of LOD Based on a Given LOQ.....................2.5 Presentation of Interlaboratory Comparison Results: Statistical Analysis in Interlaboratory Comparisons... 98 Characteristics and Organization of Interlaboratory Comparisons. 146 7................................7 Testing the Correctness of the Determined LOD...................1 Comparisons of Results Obtained Using Various Procedures.........2..... .. 145 7................... 148 7..5....................................4 Range..................... 2].........................5 Sensitivity........................ 131 Characterization of Validation Parameters........................................2......... 7.........................2.....2..................................... 134 7.................

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

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

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

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 .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 .© 2009 by Taylor & Francis Group. Measurements.

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 .© 2009 by Taylor & Francis Group. LLC .

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

+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. +a〉 is constant and not equal to zero. Normal distribution. but outside the interval is equal to zero. also called the Gaussian distribution (particularly in physics and engineering). concentration. Unfortunately.g.© 2009 by Taylor & Francis Group. These parameters can be divided into four basic groups: • • • • measures of location measures of statistical dispersion measures of asymmetry measures of concentration . It is an infinite family of many distributions. 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.. physiochemical property). The distribution is determined by a pair of parameters – a and +a. 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. Then. is a very important probability distribution used in many domains. such complete information is seldom available. N (μx. one may infer a characteristic using an estimation of some of its parameters (statistical parameters) or its empirical 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. As a rule.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. defined by two parameters: mean (location) and standard deviation (scale). Because this distribution is continuous. Statistical parameters are numerical quantities used in the systematic description of a statistical population structure. LLC . content. it is not important whether the endpoints – a and +a are included in the interval.

LLC . .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.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. • The arithmetic mean from a sample is a good approximation (estimation. estimator) of the expected value.2) • The sum of deviations of individual values from the mean is equal to zero: ∑ (x − x ) = 0 i m i =1 n (1.Basic Notions of Statistics 3 1.© 2009 by Taylor & Francis Group.4) • The arithmetic mean is sensitive to extreme values of the characteristic. • The arithmetic mean fulfills the condition: x min < x m < x max (1.3 Measures of Location Measures of location use one value to characterize the general level of the value of the characteristic in a population [1]. 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.

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

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

12) where k is the number of series of parallel determinations. the formula is simplified to the following equation: SDg = 1 k ∑ SD i =1 k 2 i (1. xm ≠ 0. LLC . The standard deviation of the arithmetic mean SD is calculated according to the following equation: SD = SD n (1.© 2009 by Taylor & Francis Group. For series with equal numbers of elements. If an expected value µx is known. and it is always expressed in the same units as the results.13) The mean absolute deviation D is an arithmetic mean of absolute deviations of the values from the arithmetic mean.9) Relative standard deviation (RSD) is obtained by dividing the standard deviation by the arithmetic mean: RSD = SD xm (1.10) Obviously.14) . the standard deviation is calculated according to the following formula: SD = ∑( x − µ ) i x i =1 n 2 n (1.11) 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 (1.6 Quality Assurance and Quality Control • Standard deviation is always a denominate number. 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.

the greater asymmetry.15) The CV is the quotient of the absolute variation measure of the investigated characteristic and the mean value of that characteristic. 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.6 Measures of Concentration A concentration coefficient K is a measure of the concentration of individual observations around the mean. 1. The skewness coefficients are applied in comparisons to estimate the force and the direction of asymmetry.1 Problem: For the given series of measurement results.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. give the following values: • • • • • • • • • • mean standard deviation relative standard deviation mean absolute deviation coefficient of variation minimum maximum range median mode . The quartile skewness coefficient shows the direction and force of result asymmetry located between the first and third quartiles. It is an absolute number.© 2009 by Taylor & Francis Group. Example 1. The CV is obtained by multiplying RSD by 100%: CV = RSD ⋅ 100% (1. the greater their value. LLC . usually presented in percentage points. 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 greater the value of the coefficient.

98 mg/dm3 0. .52 mg/dm3 12.34 12. Formulating the null hypothesis and the alternative hypothesis The null hypothesis H0 is a simple form of the hypothesis that is subjected to tests.12 12. assumed in order to explain some phenomenon. law.34 12.0257 0.32 mg/dm3 0.48 mg/dm3 0. Statistical hypothesis testing means checking propositions with regard to a population that have been formulated without examining the whole population.xls Quality Assurance and Quality Control 12.57% 12.52 12. or fact.67 12.91 12. LLC .00 mg/dm3 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_stat01.79 12. The plot of the testing procedure involves: 1.98 mg/dm3 12.264 2.53 12.8 Data: result series.02 12.34 mg/dm3 1. based on probability. A hypothesis requires testing.© 2009 by Taylor & Francis Group.98 12. The alternative hypothesis is contrasted with the null hypothesis.67 12.00 12.34 12.7 Statistical Hypothesis Testing A hypothesis is a proposition concerning a population.

6. it means that there is not enough evidence to reject the null hypothesis. 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. 4. The basic classification of a statistical test divides tests into parametric and nonparametric ones. Calculation of a test’s parameter using a sample The results of the sample are processed in an appropriate manner. The choice of an appropriate test The test serves to verify the hypothesis. Otherwise. the conclusion that the null hypothesis may be true. Parametric tests serve to verify parametric hypotheses on the distribution parameters of the examined characteristic in a parent population. the null hypothesis H0 is rejected. 5. It means that the value of the calculated parameter is not greater than the critical value of the test (read from a relevant table). 3. . It means that the value of the calculated test parameter is greater than the critical value of the test (read from a relevant table). If the calculated p value is smaller than the α value ( p < α). The p value is then compared with the assumed value of the level of significance α. Type II error — accepting the null hypothesis H0 when it is false. after selecting an appropriate statistical test. the goodness of fit in two populations. LLC . then the null hypothesis should be rejected as false. −− If the value is outside the critical region. and the randomness of sampling.g. and its location is determined by the alternative hypothesis. the null hypothesis is not rejected. Statistica). Determining the critical region of a test The size of the critical region is determined by any low level of significance α.Basic Notions of Statistics 9 2. In this case. Nonparametric tests are used to test various hypotheses on the goodness of fit in one population with a given theoretical distribution. Determination of the level of significance α Errors made during verification: Type I error — incorrectly rejecting the null hypothesis H0 when it is true. determined using the sample. hence. they are used to test propositions concerning arithmetic mean and variance. Conclusion The test statistic. is compared with the critical value of the test: −− If the value falls within the critical region.. Usually. the procedure is limited to calculating the parameter p for a given set of data.© 2009 by Taylor & Francis Group. Nowadays. 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.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. application. .17) Requirements Course of action where xi denotes uncertain result. 1.8. Calculate the value of the parameter tcalc according to the following formula: tcalc = xi − x m SD (1. • 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. it is compensated for in further calculations. read for f = n – 2 degrees of freedom (Table A. SD is the standard deviation for an unbiased series. n is the entire size of a series. together with an uncertain result.10 Quality Assurance and Quality Control 1.1.8 Statistical Tests During the processing of analytical results. xm the mean value for the unbiased series. and inference based on these tests are presented below. it is rejected. and SD the standard deviation for the unbiased series. Calculate the endpoints of the confidence interval for a single result based on the following formula: g = x m ± tcrit n SD n−2 (1.16) Course of action Inference where xm is the mean for an unbiased series. Appropriate tables with critical values for individual tests are given in the attachments at the end of the book (Appendix). various statistical tests can be used. LLC . and the values of xm and SD are calculated again. Their description. If an uncertain result falls outside the limits of the confidence interval. Appendix). otherwise.© 2009 by Taylor & Francis Group.

LLC . otherwise. SD is the standard deviation for the unbiased series. and n is total number of a series. kα is the confidence coefficient for a given level of significance α. wα is the critical parameter determined for the number of degrees of freedom f = n – 2 (Table A.01. read for f = n – 2 degrees of freedom (Table A.18) Inference where n is the entire size of a series. the initially rejected result is considered to have a gross error.2. it is rejected and xm and SD are calculated again. for α = 0.33. If the uncertain result(s) falls outside the endpoints of the determined confidence interval. If tcalc ≤ tcrit(corr).© 2009 by Taylor & Francis Group. it is rejected and xm and SD are calculated again. SD is the standard deviation for the unbiased series.1.19) Requirements Course of action Inference where xm is the mean for the biased series.65. then the initially rejected result is included in further calculations and xm and s are calculated again. Appendix). If the uncertain result falls outside the endpoints of the determined confidence interval. • 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. from a normal distribution table: for α = 0. kα = 1. together with an uncertain result.20) where xm denotes the mean for the biased series.05. Requirements Course of action Inference . • 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.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. tcrit is the critical parameter of the Student’s t test. kα = 2. Appendix).

65.2 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. for α = 0.33. with n determinations in each series (most often. kα = 2. 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.01.21) Inference where xm is the mean for the unbiased series. it is included in the series. k ≥ 30) Inference Requirements .8. • known value of the mean range for the series — Rm • known results of k series of parallel determinations. it is rejected. and xm and SD are calculated again. n = 2 or 3. kα = 1. from a normal distribution table: for α = 0.3. If the uncertain result falls outside the endpoints of the determined confidence interval.© 2009 by Taylor & Francis Group. LLC .22) where SDg is the standard deviation of the method. Appendix). If R > Rcrit. SDg is the standard deviation of the method. 1. kα is the confidence coefficient for a given level of significance α. the extremum result is rejected and the procedure is conducted anew.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. otherwise.05.

If Ri > Rcrit. 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.23) Calculate the value of the critical range according to the formula: Rcrit = zα ⋅ Rm (1. . 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. 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. the authors use a certain type of Dixon’s Q test that makes it possible to test a series comprising up to 40 results. . .4.24) 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. read for the selected level of significance α and the number of degrees of freedom f = n. Calculate the value of the range R according to the formula: R = Xn − x1.3 Dixon’s Q Test [3.8. Appendix). xn. the ith series of the measurement results is rejected. H1: In the set of results there is a result with a gross error. . If one of the calculated parameters exceeds the critical value Qcrit. In some studies [1]. LLC . .© 2009 by Taylor & Francis Group.5. 1.25) Course of action Inference Compare the obtained values with the critical value Qcrit (Table A. Appendix).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. • 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.

6. . . . . xn.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. 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 the range R according to the formula: R = Xn − x1. .© 2009 by Taylor & Francis Group. read for the selected level of significance α and the number of degrees of freedom f = n. • 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. read for the selected level of significance α and the number of degrees of freedom f = n. 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 (1.6. • 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: x1.26) Requirements Course of action Compare the obtained values with the critical value Qcrit (Table A. H1: In the set of results there is a result with a gross error. . . . 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: x1.28) .27) Requirements Course of action Compare the obtained values with the critical value Qcrit (Table A. Appendix). xn. Appendix). LLC . . 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.

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.Basic Notions of Statistics 15 Inference Compare the obtained values with the critical value Qcrit (Table A.© 2009 by Taylor & Francis Group. then it may be inferred that the compared values of the standard deviation differ in a statistically significant manner — rejection of H0. LLC . If the calculated χ2 value is greater than the critical value read 2 from the tables ( χ 2 > χcrit ).8.29) Requirements Course of action where SD is the standard deviation calculated for the set of results. 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. If one of the calculated parameters exceeds the critical value Qcrit. and n is the number of results in an investigated set. Appendix). H1: The variance calculated for the series of results is different from the set value in a statistically significant manner. If the calculated χ2 value does not exceed the critical value 2 ( χ 2 ≤ χcrit ). 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. • normal distribution of results in a series Calculate the standard deviation for the series of results.7. SDo is the set value of the standard deviation. Calculate the chi square test parameter χ2 according to the formula: χ2 = n ⋅ SD 2 2 SDo (1. read for the selected level of significance α and the number of degrees of freedom f = n. 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.6. 1.

8. 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. 1. 4. then it may be inferred that the compared values of the standard deviation differ in a statistically significant manner — rejection of H0.16 Quality Assurance and Quality Control 1.5 Snedecor’s F Test [3. SD2 denote the standard deviations for the two sets of results.8. then it may be inferred that the calculated values for the standard deviation do not differ in a statistically significant manner — acceptance of H0. n2 are the number of results for two sets. • normal distributions of results in a series Calculate the standard deviations for the compared series of results. 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 1.30) Inference where SD1. and n1. 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.6 Hartley’s Fmax Test [3] Test Aim Hartley’s Fmax test Compare the standard deviations (variances) for many sets of results . If the calculated F value does not exceed the critical value (F ≤ Fcrit).© 2009 by Taylor & Francis Group.8. 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. H1: The variances calculated for the compared series of results differ in a statistically significant manner. Appendix). If the calculated F value is greater than the critical value read from the tables (F > Fcrit). LLC .

then it may be inferred that calculated standard deviations do not differ in a statistically significant manner — acceptance of H0. LLC . the calculated number of degrees of freedom f = n – 1. Calculate the value of the Fmax test parameter according to the formula: Fmax = 2 SDmax 2 SDmin (1. Compare the calculated value with the critical value of the parameter for the assumed level of significance α. then it may be inferred that the compared standard deviations differ in a statistically significant manner — rejection of H0. If the calculated Fmax value does not exceed the critical value (Fmax ≤ Fmaxo). SDmin are the greatest and smallest value from the calculated standard deviations for the sets of results. Appendix). and the number of the compared series k (Table A.9. H1: The variances calculated for the compared series of results differ in a statistically significant manner. • the number of results in each series of the sets is greater than 2 Requirements .8. 1.31) Inference where SDmax.© 2009 by Taylor & Francis Group. H1: The variances calculated for the compared series of results differ in a statistically significant manner. 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. • 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.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.

8. then it may be inferred that the calculated standard deviations do not differ in a statistically significant manner — acceptance of H0. H1: The variances calculated for the compared series of results differ in a statistically significant manner.© 2009 by Taylor & Francis Group.33) SD o = 1 n−k ∑ SD ( n − 1) 2 i i n =1 (1. LLC .7. • number of results in each series of the sets is greater than 2 Requirements .303 n − k log SD o − c ( ) ∑ ( n − 1) log ( SD ) i 2 i i =1 k (1.34) where n is the total number of parallel determinations.32) in which: c = 1+ 1 3 k −1 2 1 1 − ( ) ∑ n −1 n − k i =1 k i k (1. Appendix). If the calculated Q value does not exceed the critical value 2 (Q ≤ χcrit ). then it may be inferred that the compared standard deviations differ in a statistically significant manner — rejection of H0. Inference 1.8 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. ni is the number of parallel determinations in a given series. If the calculated Q value is greater than the critical value read 2 from the tables (Q > χcrit ). k is the number of the compared method (series). Calculate the value of a Q test parameter according to the formula: Q= 2 2. and SDi is the standard deviation for the series i. 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.18 Quality Assurance and Quality Control Course of action Calculate the standard deviation for the compared series of results.

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. 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. . If the calculated t value is greater than the critical value read from the tables (t > tcrit). If the calculated t value does not exceed the critical value tcrit. then it may be inferred that the compared standard deviations differ in a statistically significant manner — rejection of H0.Basic Notions of Statistics 19 Course of action Calculate the standard deviations for the compared series of results. x2i denote the individual values of results for the compared sets.37) Inference where k is the number of pairs of results. Compare the calculated t value with the critical value tcrit. LLC . Appendix). H1: The calculated means for the compared series of results differ in a statistically significant manner.36) Calculate the value of parameter t according to the formula: t= (1 − L )( k − 2) L (1. a parameter for the assumed level of significance α and the calculated number of degrees of freedom f = k – 2 (Table A.9 Student’s t Test [3. then it may be inferred that the calculated standard deviations do not differ in a statistically significant manner — acceptance of H0.1.© 2009 by Taylor & Francis Group. so that the relation t ≤ tcrit is satisfied. 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.8. and x1i. 1.

.© 2009 by Taylor & Francis Group. H1: The calculated mean differs in a statistically significant manner from the assumed value..38) Inference where x1m . Appendix). 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. If the t value does not exceed the critical value tcrit (t ≤ tcrit). 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. μ is the reference (e. certified value). Section 1. SD is the unit of deviation. 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. Calculate the Student’s t test parameter according to the equation: t= xm − µ SD n (1.5) Calculate the means and standard deviations for the series of results. SD2 are the standard deviations for the sets of results. LLC . • 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.1.8.20 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. x2m denote the means calculated for the two compared sets of results. and SD1. then it is inferred that the compared means differ in a statistically significant manner — rejection of H0. then it may be inferred that the obtained means do not differ in a statistically significant manner — acceptance of H0.g. If the calculated t value is greater than the critical value read from the tables (t > tcrit).39) Test Aim Hypotheses Requirements Course of action where xm is the mean calculated for the set of results.

If the calculated t value is greater than the critical value read from the tables (t > tcrit). and n is the number of results. and SD1. H1: The calculated means for the compared series of results differ in a statistically significant manner. If the t value does not exceed the critical value tcrit (t ≤ tcrit). SD2 are the standard deviations for the sets of results.Basic Notions of Statistics 21 Inference e. x2m denote the means calculated for the two compared sets of results. it is inferred that the mean is different from the set value in a statistically significant manner — rejection of H0.10 Cochran-Cox C Test [3] Test Aim Cochran-Cox C test Compare the means for the series of sets of results. Appendix).© 2009 by Taylor & Francis Group. 1.1. 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. • 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. and z2 = n1 − 1 n2 − 1 Hypotheses Requirements Course of action C= x1m − x 2 m z1 + z2 (1. the standard deviation of the set of results which the mean was calculated based on. LLC . then it may be inferred that the obtained mean is not different from the set value in a statistically significant manner — acceptance of H0. Calculate the value of parameter C according to the formula: in which: z1 = 2 SD12 SD2 . Calculate the critical value of the parameter C (Ccrit) according to the formula: Ccrit = z1t1 + z2t2 z1 + z2 (1.40) (1.41) where x1m..8. 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.g.42) .

then it may be inferred that the obtained mean values do not differ from one another in a statistically significant manner — acceptance of H0. H1: Calculated means for the compared series of results differ in a statistically significant manner.44) in which (1.1. If the calculated C value is greater than the calculated critical value (C > Ccrit).© 2009 by Taylor & Francis Group.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).8.43) SD12 n1 c= 2 SD12 SD2 + n1 n2 2 SD12 SD2 < n1 n2 (1. then it is inferred that the obtained means differ from one another in a statistically significant manner — rejection of H0. • 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. 1.45) . for f1 = n1 – 1 and f2 = n2 – 1. 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 (1.22 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. and the level of significance. Appendix). LLC . respectively. Compare the calculated C value with the calculated critical value Ccrit. the number of degrees of freedom.

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. and the calculated values of c. .Basic Notions of Statistics 23 Inference where x1m . f 2. If the calculated v value is greater than the calculated critical value (ν > νo ). SD2 are the standard deviations for the sets of results.© 2009 by Taylor & Francis Group. 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. If the value of ν does not exceed the critical value νo (ν ≤ νo ).46) 2 i where SDmax is the maximum standard deviation in the investigated set (among the investigated laboratories). Compare the calculated value ν with the critical value νo for the corresponding level of significance α. it is inferred that the obtained means differ from one another in a statistically significant manner — rejection of the hypothesis H0. f 2 = n2 – 1. and p is the number of standard deviations (the number of compared laboratories).8. 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. c) (Table A. LLC . and SD1. f1. Appendix).13) Calculate the standard deviations for each of the compared sets of results. SDi is the standard deviation for a given series (data from a laboratory).8. the number of degrees of freedom f1 = n1 – 1. Calculate the value of parameter C using the formula: C= 2 SDmax Requirements Course of action ∑ SD i =1 p (1. x2m denote the means calculated for the two compared sets of results. and thus νo (α.10. 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 = 0.01.11. it enables the detection of one outlier. .12. If the value of the test parameter is greater than the critical value corresponding to the level of significance α = 0.05.05. xm is the mean.13 Grubbs’ Test [6. the number of results in a series. thus it should be repeated until no outliers are observed in the remaining results Calculate the standard deviation for the set of results.47) Course of action Inference where xp is the value in the set of results considered to be an outlier. Appendix). then the investigated result is an uncertain value. . 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. then the investigated result is considered an outlier. . Appendix).8. the number of laboratories (Table A. . LLC . p in an increasing sequence. Order the set of data xi for i = 1. Compare the calculated value Gp with the critical value for a given p value. then the investigated result is considered correct.© 2009 by Taylor & Francis Group. then the investigated result is considered correct. Calculate the value of parameter Gp according to the relation: Gp = (X p − Xm ) SD (1. and SD is the standard deviation.24 Quality Assurance and Quality Control Inference Compare the calculated C value with the critical value for a given n value. 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. If the value of the calculated test parameter is less than or equal to the critical value corresponding to the level of significance α = 0.8. If the numerical value of a respective test parameter is greater than the critical value corresponding to the level of significance α = 0. 2. and p the number of laboratories (Table A. . 1.12) • with a single use.05 and less than or equal to the critical value corresponding to the level of significance α = 0.

and the course of action should be continued until there are no more outliers in the set of results. 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. 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. 2. . . then the investigated result is an uncertain value.12) • in a given course of action. p) when testing two of the highest results or xm(1.2) for two of the lowest results. according to the equations: x m ( p−1. then the investigated result is considered an outlier.48) Calculate the values of parameters xm( p –1.01.2 ) 2 .2) = 1 p −1 ∑x i i =3 p (1. after rejection of this value from the set of results. two (the greatest or the smallest) results may be rejected from the set of results Order the set of data xi for i = 1.8. LLC .2) according to the equations: SD(2p−1. or x m (1.05.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. or 2 (1. . p ) ) . . p in an increasing sequence. p) = 1 p −1 ∑ i =1 p− 2 xi . and less than or equal to the critical value corresponding to the level of significance α = 0. p) or SD(1. If the value of the test parameter is greater than the critical value corresponding to the level of significance α = 0.© 2009 by Taylor & Francis Group. 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. the test for the series of p – 1 results may be conducted again. p) = ∑( x − x i i =1 p i i =3 p− 2 m ( p−1.01.1) = ∑( x − x ) m (1.50) SD(22.49) Calculate the values of respective parameters: SD( p –1.

01.26 Quality Assurance and Quality Control Calculate the value of parameter G according to the equations: G= SD(2p−1. after rejection of these values from the set of results. 9]. then the investigated results are considered outliers.5 ⋅ Me ri (1. where xi includes the interval from x1 to xn.2 ) 2 SDo (1.8. Appendix).05. Compare the values of |ri| with 4. If the numerical value of a corresponding test parameter is greater than the critical value corresponding to the level of significance α = 0. 1. or G = SD(2 1.52) Inference Calculate the absolute values |ri|.51) Inference Compare the calculated value of G with the critical value for a given p value.53) then the result xi is considered an outlier.01. then the investigated results are uncertain. then the investigated results are considered correct.14 Hampel’s Test The Hampel test by some authors is called Huber’s test [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. the number of laboratories (Table A. Order the values of |ri| in an increasing sequence.5 ⋅ Me ri . If the following condition is satisfied ri ≥ 4. LLC . and the course of action should be continued until there are no more outliers in the set of results. . 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. If the value of a test parameter is greater than the critical value corresponding to the level of significance α = 0. Calculate the median deviations Me ri .12. Calculate the median Me for all the results xi.© 2009 by Taylor & Francis Group. Calculate the deviations of ri from the median for each result using the formula: ri = xi − Me ( ) (1. the test for the series of p – 2 results may conducted again.05 and less than or equal to the critical value corresponding to the level of significance α = 0. p) 2 SDo .

16 En Score [10. a result is considered unsatisfactory.8. and SD is the deviation unit: The standard deviation calculated using all the values in a set. LLC . a result is considered uncertain. u( xlab ) is the combined standard uncertainty result obtained by a given laboratory. 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. If 2 < Z < 3 . Combined standard uncertainty calculated using the relation: u = u(2xlab ) + u(2xref ) (1. Inference If Z ≤ 2. and u( xref ) the combined standard uncertainty of the reference values.54) Requirements Course of action where xlab is result obtained by a given laboratory. the estimation is unsatisfactory. a result is considered satisfactory.Basic Notions of Statistics 27 1. . 1.55) where u( xref ) is the standard uncertainty of the accepted value/ reference value. xref is the reference value. xref is the assumed value/the reference value. The modified standard deviation calculated using the relation: SDmod = SD 2 + u(2xref ) (1. the estimation is satisfactory.56) where u( xlab ) is the standard uncertainty of a value obtained by a given laboratory.57) where x lab is the value obtained by a given laboratory.© 2009 by Taylor & Francis Group.15 Z Score [10.8. Inference If En ≤ 1. If Z ≥ 3. 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. If En > 1.

Calculate the values of parameter ki for a given series and for a given laboratory.13b.17 Mandel’s Test [6. mark the horizontal lines to test the data’s configuration.28 Quality Assurance and Quality Control 1.01 or 0. 12. 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. 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. according to the formula: hi = x mi − x m 1 ( p − 1) ∑ (x i =1 p − x m )2 (1. and p is the number of laboratories. LLC .13a or Table A. Appendix).60) . On the graph of the values of parameter h.05) (Table A. according to the formula: ki = SDi p Test Aim Requirements Course of action ∑ SDi2 (1.58) i where ni is the number of results for a given series obtained by a given laboratory. The value of parameter hi greater than h needs to be checked from an analytical viewpoint. Calculate the values of parameter hi for a given series and for a given laboratory.8. corresponding to the Mandel h coefficients for a given level of significance (α = 0.© 2009 by Taylor & Francis Group.59) mi Inference Make a graph of the values of parameter hi for each series in the sequence of laboratories. 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.

then it may be inferred that there are statistically significant differences in distribution functions for both compared series.© 2009 by Taylor & Francis Group. n2 denote the number of results for a given series. If the λn value does exceed the critical value λα (λn > λα ).9 Control Charts 1. and Dn1n2 is the greatest value of differences between empirical distribution functions. 1. Calculate the values of parameter λn. then it may be inferred that there are no statistically significant differences in distribution functions for both compared series. 1.13a or Table A. Appendix). If the λn value does not exceed the critical value λα (λn ≤ λα ). Appendix). 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. On the graph of the values of parameter k. Compare the λn value with critical value — λα for a given level of significance (Table A. The value of parameter ki bigger than k value needs to be checked from an analytical viewpoint.61) Inference where n1. 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.18 Kolmogorov-Smirnov Test [2. corresponding to the Mandel k coefficients for a given level of significance (α = 0. according to the formula: λn = n1n2 Dn n n1 + n2 1 2 (1. the most frequently used charts are Shewhart charts.13b. It enables fast and simple .01 or 0.14. LLC .1 Shewhart Charts Control charts are used to test the stability of research results conducted in a given laboratory.8. In practice.9.05) (Table A.Basic Notions of Statistics 29 Inference Make a graph of the values of parameter ki for each series in the sequence of laboratories. mark the horizontal lines to test the data’s configuration.

the upper and lower control limits (UAL and LAL. and two statistically determined control limits. hence.g.. LLC . that is. provided that the process is statistically ordered. Limits of ±2SD are also marked. charts are prepared separately for each procedure. and the values of the observed characteristics (the mean) on the y-axis. or in other words the upper and lower warning limits. Mark the obtained measurement results for 20 consecutive samples. and thus fast correction and confirmation of the reliability of the research. every batch).2 Shewhart Chart Preparation Preparation of a chart depicting mean and standard deviation (xm – SD) will be described as an example. the occurrence of any value from a sample falling outside these limits is simply warning about a possible transgression of the control limits.9).© 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. start preparation of the first chart: Mark the consecutive numbers of result determinations on the x-axis of the graph. expressed either in time (e. but the results obtained for control samples should be marked parallel to the received results for the investigated samples: .8. respectively). The possibility of transgressing the control limits as a result of random incident is insignificantly small. The course of action in preparing the control chart is as follows: • Conduct 10–20 measurements for a standard sample.30 Quality Assurance and Quality Control detection of abnormalities in the configuration of the marked points. every hour) or quantity (e. The measurement is obtained from a series of measurement results obtained at approximately regular intervals. where SD is the standard deviation of the investigated characteristics. • Calculate the mean xm and the standard deviation SD. • If the hypothesis is not rejected.. both values should be determined for the unbiased series. 1. however. usually a graph with control limits depicted. it is recommended that action be taken on the chart. the values of a certain statistics measurement are registered. Both the upper and lower limits on the chart are found within ±3SD from the central line.g. therefore. when a point appears outside the control limits ±3SD. one line on either side of the central line. the limits of ±2SD are called warning limits (UWL and LWL). On such a graph.7% of the values fall in the area bounded by the control lines. Limits of ±3SD (so-called action limits) show that approximately 99. Mark a central line CL on the graph corresponding to the reference values of the presented characteristic. The main role is played by an appropriate control chart.9. after the initial rejection of outliers.

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

1005 4.20 0.23 4.32 4. Mark the central line.27 4.07 4.© 2009 by Taylor & Francis Group.17 4. 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 4.04 4.50 4.12 4.34 4.90 Data 4. LLC .14 4.22 4. and the warning and action lines.03 4.2 Problem: Draw a Shewhart chart for the 20 given measurement results obtained for the test samples.36 4.40 4.22 .23 4.30 4.21 4.11 4.20 4.32 4.32 Quality Assurance and Quality Control Example 1.10 4.00 3.

00 3.Basic Notions of Statistics Graph: 4.32 4.44 4.17 4.12 4.40 4.15 4.11 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 4.33 4.xls Example 1.18 4.20 4.35 4.11 Conclusion ! ok ok ok ok ok ok ! ok ok ok ok ok ok ok ok ok ok ok ok .23 4.3 Problem: Mark the following data from the previous example on the chart.90 3.01 4.12 4.32 4.08 4.60 4.11 4.10 4.00 4.80 0 5 10 15 LWL LAL 20 CL UAL UWL 33 Excel file: exampl_stat02.30 4.20 4.41 4.34 4.50 4. LLC .© 2009 by Taylor & Francis Group.

34 Solution: Mean1 SD1 Mean2 SD2 UAL UWL LWL LAL Graph: 4.10 4.30 4.60 4.© 2009 by Taylor & Francis Group.40 4.80 0 5 10 Quality Assurance and Quality Control 4.21 0.90 UAL UWL CL LWL LAL 15 20 Excel file: exampl_stat03.90 3.20 0.40 4. LLC .00 3.50 4.xls .20 4.50 4.128 4.00 3.101 4.

90 3. LLC .111 4. the values have been calculated as the means of the two previous charts.xls .106 4.05. 19) F < Fcrit t tcrit(0.51 4. n Standard deviation.05.50 4.10 4.456 2. 36) t < tcrit 20 0.80 0 5 10 15 LWL LAL 20 CL UAL UWL 4.19 0.200 0. and then (with variances not differing in a statistically significant way) the mean were compared using the Student’s t test.© 2009 by Taylor & Francis Group. The remaining values were used to calculate the means and the standard deviation.98 3.013 For the new chart.20 4.40 3. SD Mean n/(n – 1)SD2 F Fcrit(0.30 4.Basic Notions of Statistics 35 Example 1. The variances were compared using the Snedecor’s F test.60 4.22 1.101 4. Mean SD UAL UWL LWL LAL Graph: 4.4 Problem: Draw a new chart based on the data from the previous example. Series 1 Series 2 No results.184 0. Solution: Values 1 and 8 have been removed from the set of data.031 18 0. 17.85 0.88 Excel file: exampl_stat04.011 1.40 4.00 3.

a: a= ∑ y − b∑ x i i =1 i =1 n i n (1. 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.10 Linear Regression Linear correlation is the most frequently used correlation in analytical chemistry.64) . A decisive majority of analytical measurements uses the calibration stage. 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 (1.36 Quality Assurance and Quality Control 1.© 2009 by Taylor & Francis Group. such as: • accuracy — through the determination of systematic errors • linearity • limits of detection Therefore.63) ∑x • Intercept value. a linear regression method is applied.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. To determine the functional dependency that connects the output signal with analyte concentration. The equation of the linear regression is: y = b⋅x + a (1. we present below the course of action for the linear regression method. together with a presentation of the determination method for the calibration chart parameters.

SDxy : SDxy = ∑( y − Y ) i i i =1 n 2 n−2 (1.67) • Residuals. LLC .© 2009 by Taylor & Francis Group.66) ∑ i =1 n xi • Intercept. 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 (1.65) • Values of standard deviations for: • Slope. SDb : SDb = SDxy 2 ∑ i =1 n 1 xi2 − n (1.Basic Notions of Statistics 37 • Regression coefficient.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 . SDa: SDa = SDxy n ∑x i =1 n 2 i n ∑ i =1 n xi2 − ∑ i =1 xi 2 (1.

70 Calculations very often use values with different numbers of significant digits and different number of digits after the decimal point. A value obtained from a calculation(s) should be recorded in an appropriate way. the number of significant digits in a result should be the same as in the value with the fewest significant digits.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.2 Then it should be presented with three significant digits: 73.38 Quality Assurance and Quality Control 1.123 349. Significant digits in the decimal notation of a given number are all the digits without initial zeros.2113 0.2113 0.2 then it should be presented with one digit after the decimal point: 375. if a result is the sum of numbers: 11.0010823 20.546 0. After addition or subtraction.123 349.63 × 104 34. the number should be “read” from left to right until reaching the first digit that is not zero. LLC .8 For multiplication and division.80 0. 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.23 15. strictly dependent on the notation of the values applied in the calculation(s). If a result is a product of the following numbers: 11.1200 507. the significant digits are underlined: 230. In the example below.© 2009 by Taylor & Francis Group. To determine how many significant digits there are in a number. That digit and all the subsequent digits are called significant.23 15. For example.4 × 102 .

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

Analytical data on researched material objects are a specific type of information. Measurement results must be reliable. A new type of market arose. quality of the process. quality of the instruments. but is based on the analyses of appropriate samples. and quality of the work and organization. but also in the economy and technology (related to control over the processes of manufacturing consumer goods). To satisfy the growing demand for analytical data. Quality control — a complex system of actions to obtain measurement (determination results) with the required quality level. The quality of information can be divided into components: quality of results. 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. 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. that — in other words — are characterized by the greatest information capacity possible. samples have to be collected in such a way that the most important criterion — that is.© 2009 by Taylor & Francis Group. This leads to the conclusion 41 .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). This information is not usually obtained through an analysis of the whole object.of Analytical 2 Quality Results 2. Analytical quality — consistency of the obtained results (chemical analysis) with the accepted assumptions.” Access to a variety of information sources facilitates decision making not only in politics. 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. representativeness — is met. LLC .2 Introduction The past decade or so was undoubtedly a period of “information hunger. Therefore. where information is bought and sold.

Results of analytical measurements are a type of a product of the chemical analyst’s work. the notions of quality have a specific meaning. It is the quality of a result.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. its (high) quality must be documented and maintained.© 2009 by Taylor & Francis Group. and assuring the quality of analytical results. together with its control and assurance. The notion of reliability is closely associated with the notion of quality.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. 2. estimation. 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 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 . the expected or the standard one. For the obtained result to be comparable (authoritative. reliable) to the reference value. 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. Both manufactured products and analytical results must have an appropriate quality. that determines and confirms its reliability. In addition. The need for a uniform and defined control system. LLC . The quality of results of analytical measurements must be assured in the first place to draw conclusions about the quality of the examined products.

a schematic presentation of the elements of a quality control/quality assurance system used for obtaining reliable analytical results is shown [7]. 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. RELIABLE ANALYTICAL RESULT UNCERTAINTY TRACEABILITY REQUIREMENTS METHOD VALIDATION REFERENCE MATERIALS INTERLABORATORY COMPARISONS TOOLS FIGURE 2. In Figure 2. LLC .Quality of Analytical Results 43 At present. although increasing attention is paid to the procedures of accreditation [6].1 Position and role of elements of a quality control/quality assurance (QC/QA) system for obtaining a reliable analytical result. . To assure measuring traceability. 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. it is indispensable to use both the certified reference materials and the analytical procedures subject to prior validation. 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.© 2009 by Taylor & Francis Group. introduced by a given laboratory. These are first of all statistical instruments based on metrology. The elements of the quality system are interdependent.1. is in principle voluntary.

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

analytical procedures usually involve two operations associated with calibration: • Periodic reliability test of indications of the instruments used by means of standard mixtures. and the determination of the quality of the control system elements • Helping users to derive inferences on the quality system. Accordingly.3 Necessary parameters for a reliable analytical result. containing strictly defined concentrations of analytes • Testing the reliability of the whole plot of the analytical conduct UNCERTAINTY X ± U SI TRACEABILITY FIGURE 2. A schematic representation of this concept is shown in Figure 2. must be defined in a way that is intelligible for the user. LLC . 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. Each of these elements.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.© 2009 by Taylor & Francis Group.3. 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 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. along with the determination and the control of the elements of the quality control. . 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. These two parameters are the basic requirements for a reliable measurement result. The necessity of presenting the result together with these two basic parameters must be remembered by every “producer” of analytical results. in order to be applicable.

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

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

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

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

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

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

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

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

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

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

+a〉 .xls Example 4. LLC . Cst = 1001 ± 2 mg/dm3 Solution: Due to no additional information. based on data given by the manufacturer. standard uncertainty is calculated accordingly: 2 u(Cst ) = k Excel file: exampl_uncert02. and the calculated standard uncertainty is a 3 where a is the midpoint of the range 〈– a. Data: Standard solution concentration. Cst = 1001 ± 2 mg/dm3 Solution: Because value for k is given. +a〉. we assume a rectangular distribution. +a〉. Uncertainty given by the manufacturer have been calculated for coverage factor k = 2.15 mg/dm3 Excel file: exampl_uncert01. which means that the value is in the range 〈– a. but the occurrence of the mean value from the range is the most probable.xls u(Cst ) = 1 mg/dm3 . When a variable has a triangular distribution.1 Problem: Calculate standard uncertainty for the concentration of magnesium in a standard solution.© 2009 by Taylor & Francis Group. Data: Standard solution concentration.60 Quality Assurance and Quality Control probability) a value in the range 〈– a.2 Problem: Calculate the standard uncertainty for the concentration of magnesium in a standard solution. based on data given by the manufacturer. u(Cst ) = 2 3 u(Cst ) = 1. the calculated standard uncertainty is a 6 Example 4.

it is more convenient to apply relative uncertainties.8 6 u(Vfl ) = 0.3) then the value of the combined uncertainty is described by the following equation: uc( y) = u(2x1 ) + u(2x2 ) + … + u(2xn ) (4.8 cm3 Solution: u(Vfl ) = Excel file: exampl_uncert03. based on data given by the manufacturer.1). Data: Volume. .5) If the value of the analytical result is a quotient/product of the measurement values. LLC .3 Problem: Calculate the standard uncertainty for the determination of volume 500 cm3 using volumetric flask.xls 0.© 2009 by Taylor & Francis Group.32 cm3 4. A relative uncertainty is described by the following relation: ur ( xi ) = u( xi ) xi (4.Uncertainty 61 Example 4. 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. The applied principles of uncertainty propagation in calculating the standard uncertainty of an analytical result.4) Due to the very frequent occurrence of individual measurement values being expressed in different units.2) When the value of an analytical result is the sum or difference of the measurement values y = x1 + x 2 + … + x n (4. Vfl = 500 ± 0. Assume triangular distribution. For a given mathematical model that binds the final results of analysis with measured parameters (Equation 4.

6) then the value of the combined relative uncertainty is described by the following equation: ur ( y) = ur2( x1 ) + ur2( x1 ) + … + ur2( xn ) (4. with a basic diluted solution with a concentration of 1001 ± 2 mg/dm3. Therefore. To establish a uniform distribution for each of the measured parameters.0017 (mol/dm3) cNaOH ± U(k = 2) = 0. Uncertainty calculated according to the aforementioned equation is a combined standard uncertainty of the final determination. 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. Presentations of the final result of the analysis as: result ± expanded uncertainty (after using an appropriate k factor).62 Quality Assurance and Quality Control y= x1x 2… x 3… (4. a standard solution was obtained with the predetermined concentration.1038 mol/dm3 ± 1. 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. After filling the flask to the line and mixing the solution.1038 ± 0. LLC .7) 5.© 2009 by Taylor & Francis Group. To calculate the value of the expanded uncertainty. calculate the following: • The value of the combined and expanded uncertainty (for k = 2) for the obtained standard solution concentration. the standard uncertainty should be multiplied by an appropriate coverage factor k. a correctly presented result of an analysis should be as follows: or Example 4. cNaOH ± U(k = 2) = 0.6 (%) . With the aim of obtaining a standard solution with a concentration of ~0. and present indication results.5 mg/dm3. 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. for which the expanded uncertainty has been calculated Thus. along with units for y and U ) • k factor value.4 Problem: A standard Mg2+ solution was prepared.

02 u(Vf1) 0. LLC .0012 2 c ur(c) U(c) U Result Excel file: exampl_uncert04.5005 mg/dm3 0.50 6. 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. With the aim of obtaining a standard solution with a concentration of ~0. % 6.0115 0.25 ur Cst Vp1 Vf1 Vp2 Vf2 k 0.Uncertainty • The participation percentage of each of the standard uncertainty values in the determined values of the combined uncertainty.2 u(Vp2) 0.03 u(Vf2) 0.25 62.4% 0.501 ± 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 cm3.0122 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. with a basic diluted solution having a concentration of 1001 ± 2 mg/dm3.5 Problem: A standard Mg2+ solution was prepared.0122 mg/dm3 2.5 mg/dm3.012 mg/dm3 Example 4.25 18.xls 0.0012 0. After .0012 0.0035 0.© 2009 by Taylor & Francis Group.75 6.

calculate the following: • The value of the combined and expanded uncertainty (for k = 2) for the obtained standard solution concentration. % 9. and present indication results.0012 0.0012 0.2 u(Vp2) 0.08 18.0012 0. • The participation percentage of each of the standard uncertainty values in the determined values of the combined uncertainty.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. 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. LLC .0012 2 .64 Quality Assurance and Quality Control filling the flask to the line and mixing the solution.18 9.2 u(Vp2) 0.04 u(Vf1) 0.28 9.© 2009 by Taylor & Francis Group. and after being filled to the line a standard solution was obtained with the predetermined concentration. To establish a uniform distribution for each of the measured parameters.09 18.09 ur Cst Vp1 Vf1 Vp2 Vf2 Vp3 Vf3 k 0.0023 0.18 9. 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. After filling the flask to the line and mixing the solution. 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 cm3.04 u(Vf2) 0.03 u(Vf2) 0.0035 0.0023 0.09 27.

for which its manufacturer gave a measurement uncertainty of 0.5 u(mbrutto) Rectangular (R) or triangular (T) mg mg mg mg R Solution: mnetto u(mnetto) k U(mnetto) Result Excel file: exampl_uncert06.5 mg.55 mbrutto 0. Data: Mass (tarra) Mass (brutto) Uncertainty of single measurement Distribution 187. 332.0053 0.82 mg Example 4.7 Problem: The weighed standard sample (Example 4.0053 mg/dm3 1. Calculate the combined standard uncertainty of the obtained standard solution concentration.408 mg 2 0. The mass measurement was carried out using an analytical scale. Calculate the standard uncertainty of the mass measurement.72 mtarra 332.0053 mg/dm3 65 Example 4. Assuming the value of the coverage factor to be 2. Give a correct presentation of the mass measurement result. Assume a rectangular distribution of the parameter.© 2009 by Taylor & Francis Group.1% 0.xls 0.82 mg 144. 187.5 u(mtarra) 0.Uncertainty c ur(c) U(c) U Result Excel file: exampl_uncert05. Assuming .72 mg). calculate the expanded uncertainty of the mass measurement. LLC .xls 144.5005 ± 0.6 Problem: A standard sample was weighed for the preparation of a standard solution.5005 mg/dm3 0.83 mg 0.6) was put into a measurement flask (250 cm3) for which the manufacturer provided an uncertainty value equal to 0.83 ± 0. Assume a rectangular distribution of the parameters. The mass was calculated as the difference of two mass measurements: gross (container with a sample.4 cm3.55 mg) and net (container.

Assume a rectangular distribution of parameters.2 Rectangular (R) or triangular (T) mg mg cm3 cm3 cm3 mg mg cm3 cm3 cm3 R Distribution .© 2009 by Taylor & Francis Group. calculate the expanded uncertainty of the concentration.5793 ± –0. sampling 1 cm3 of the original solution using a pipette for which the manufacturer provided an uncertainty value of 0.72 332.0023 mg/cm3 Example 4. Give a correct presentation of the result. Give a correct presentation of the result.xls 0.00199 2 0.05 cm3.66 Quality Assurance and Quality Control the value of the coverage factor to be 2.8 Problem: The obtained standard solution (Example 4.2 cm3.4 u(Vflask1) u(Vflask2) 0. for which the manufacturer provided an uncertainty value 0.5 0.5 u(mbrutto) 0. Calculate the combined standard uncertainty of the obtained standard solution concentration.5793 mg/cm3 0.5 u(mbrutto) 0.55 mbrutto 250 Vflask u(mtarra) 0.0023 mg/cm3 0. LLC .5 0. Data: Mass (tarra) Mass (brutto) Flask volume Uncertainty of single measurement Distribution mtarra 187.4 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. Data: Mass (tarra) Mass (brutto) Flask volume Flask volume Pipette Uncertainty of single measurement mtarra 187. calculate the expanded uncertainty of the concentration.55 mbrutto 250 Vflask1 Vflask2 10 Vpipette 1 u(mtarra) 0.05 u(Vpipette) 0.7) was dissolved by a 1:10 ratio.72 332. and dissolving in a measurement flask (10 cm3). Assuming the value of the coverage factor to be 2.

013 mg/cm3 67 4. 16]. Therefore. or fishbone diagram — shows the influence parameters (sources of uncertainty) of a whole analytical procedure [15. 6]: • Flow diagram — drawn on the basis of information presented in detail in a standard operating procedure. respectively.2 and 4.Uncertainty Solution: Concentration ur(concentration) k U(concentration) Result Excel file: exampl_uncert08. If that dominating parameter is the repeatability of measurements. • Ishikawa.xls 0.058 ± 0.0579 mg/cm3 0. Weighing of pure substance Dilution STANDARD SOLUTION FIGURE 4. or cause-and-effect.8) where SD is the standard deviation. and n is the number of measurements.1155 2 0. then the expanded uncertainty may be calculated according to the following relation: U=k SD n (4. 4. The most helpful tools used for that are [4.5 Uncertainty and Confidence Interval In some cases. LLC .4 Tools Used for Uncertainty Estimation Correct estimation of uncertainty needs an understanding of whole analytical procedure by analyst. . if one of the parameters has a dominating influence over the uncertainty budget. the value of uncertainty can be estimated as a confidence interval.3.134 mg/cm3 0.© 2009 by Taylor & Francis Group.2 Flow diagram for the procedure of preparation of standard solution. The flow diagram and the Ishikawa diagram for the procedure of preparation of a standard solution are presented in Figures 4. calculation of uncertainty may be limited to the calculation based on the value of that parameter. The basic principle of the uncertainty propagation is underlining the influence of the quantity with the highest value.

9 Problem: The concentration of mercury was determined in water by using the cold vapor atomic absorption spectrometry (CVAAS) technique. the value of confidence interval could be calculated as: ∆x sr = t (α . For a level of significance of α = 0.54 . Provide a correct presentation of the determination result. calculate the expanded uncertainty of the determination result for k = 2.1. Considering the unrepeatability as the main component of the uncertainty budget. The series involved four determinations.© 2009 by Taylor & Francis Group.14 77. On the other hand. the aforementioned equations are thus consistent.9) For a level of significance of α = 0.05 and the number of degrees of freedom f →∞.13 76. the parameter t ≈ 2 (Table A. Data: result series. Appendix). Example 4.68 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. Given this condition. the coverage factor k = 2. f ) SD n (4.3 Ishikawa diagram for the procedure of preparation of standard solution.53 72.05. μg/dm3: 1 2 3 4 Data 71. LLC .

9075 μg/dm3 74. Those intervals are determined using a correlation that is described by the following equation: where ∆yi = Y ± SDxy ⋅ t(α . This step of the analytical procedure has influence on the combined uncertainty of the determination result for the real sample. 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 .6 Calibration Uncertainty A decisive majority of analytical measurements involve a 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. f = n− 2) 1 ( xi − x m )2 + n Qxx (4.68) in Chapter 1.10) ∆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 . drawn based on Equations (1. which is associated with the relative (comparative) character of measurements. it is possible to determine and identify the uncertainty of the determined regression curve through the determination of confidence intervals. xstd ) i • The influence of the manner of preparing the standard samples. Using a calibration curve. At the calibration step.3 ± 2. which is determined using linear regression. Standard uncertainty due to this step should be included in the uncertainty budget.© 2009 by Taylor & Francis Group. usually using a method of consecutive dilutions • Incorrect approximation of measurement points using a regression curve Figure 4.Uncertainty Solution: Mean SD k U Result Excel file: exampl_uncert09. a calibration curve technique is usually used.xls 74.335 μg/dm3 2.9075 μg/dm3 2 2. LLC .9 μg/dm3 69 4. y ) • Uncertainty due to the determination of the reference value for standard samples — u( xsmpl . There are four sources of uncertainty due to the calibration step that can influence the standard uncertainty of a single measurement u( xsmpl ) [8.63)–(1.

y ) may be calculated using the determined regression parameters according to the following relationship: u( xsmpl .0291 r = 0.70 8 7 6 Signal 5 4 3 2 1 0 0 2 4 Quality Assurance and Quality Control y = 0. 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 (4.564x – 0. LLC .© 2009 by Taylor & Francis Group.9989 6 8 Content. SDxy = residual standard deviation t(α. ppm 10 12 14 FIGURE 4. y ) = SDxy b 1 1 ( xsmpl − x m ) + + p n Qxx 2 (4.12) where 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 .11) Standard uncertainty for xsmpl due to the uncertainty of calibration and linear regression method u( xsmpl .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.

564x – 0. standard uncertainty due to the application of standard solutions at the calibration step may be described by the following equation: u( xsmpl . y ) i (4.5 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. Figure 4.0291 r = 0. requires only the value u( xsmpl . 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 ) << u( xsmpl .Uncertainty 8 7 6 Signal 5 4 3 2 1 0 0 2 4 6 8 Content. Because usually only one basic standard is used and then appropriate standard solutions are made (consecutive dilutions). If each standard sample is prepared by consecutive dilutions. .13) Therefore. its value may be estimated by only considering the number of standard samples used at the calibration stage. LLC . ppm 10 12 14 y = 0. the standard uncertainty of a result. y ) . Usually. xstd ) ≈ i u( xstd ) i n (4.9976 71 FIGURE 4. associated with an applied calibration technique. then the uncertainty budget must allow for the standard uncertainties associated with the step of standard sample preparation.© 2009 by Taylor & Francis Group.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.14) Such an uncertainty value does not allow for the uncertainty associated with the manner of standard sample preparation.

08 2.10 Problem: A calibration curve was determined using determinations of analyte concentration in samples of six standard solutions.12 1.78 6.76 5.66 7.67 5. ppm 2 2 2 4 4 4 6 6 6 8 8 8 10 10 10 12 12 12 Data y.33 3.23 3.32 4. 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. making three independent measurements for each of the solutions.72 Quality Assurance and Quality Control Example 4.41 4. LLC . for which three independent measurements were made and the result was calculated using the determined curve Data: results.2 1.32 2.97 6.59 ppm 3 .11 2.54 3.© 2009 by Taylor & Francis Group.51 6. Signal 1.12 4. Calculate: • Regression parameters of the calibration curve • Confidence intervals • Uncertainty value of the determination value for the real sample due to calibration.

The ultimate goal for an analyst is to obtain a result that will most reliably reflect the expected (actual. SDx. LLC .xls 4. r Qxx Uncertainty for result due to calibration Relative uncertainty for result due to calibration t(α = 0.7 Conclusion Each analytical result derives from a conducted measurement. f = n − 2) (from Table A.y Regression coefficient.0291 0. ppm 10 12 14 y = 0.9951 73 18 0.12 Excel file: exampl_uncert10. the basic tool for any analyst.© 2009 by Taylor & Francis Group.05. The certainty of the analytical result depends on the uncertainties occurring at all the steps of an analytical procedure.1433 0.5642 −0.1% 2.9976 210 0.16 ppm 2. b Intercept.1.Uncertainty Solution: n Slope. .0291 R2 = 0. a Residual standard deviation. real) value. Appendix) Graph: 8 7 6 Signal 5 4 3 2 1 0 0 2 4 6 8 Content.5642x – 0.

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

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

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

Laboratory reference materials. but with limits of <100 µmol/mol Defined purity As above. melting point. TABLE 5. e. or surface characteristics) These principal categories are subdivided into subcategories as indicated in the following draft list. but with limits of <50 µmol/mol Biological and clinical properties Physical properties Engineering properties Miscellaneous Chemical nature Single major constituent . SecRMs NONCERTIFIED Primary reference materials. 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. QCMs . being either pure chemical compounds or representative sample matrices.© 2009 by Taylor & Francis Group. animal fats spiked with pesticides for residue analysis). viscosity. or compound) stochiometrically and isotopically certified in amount of substance ratios with total impurities <10 µmol/mol Primary chemicals As above.Pure substances . LRMs . tensile strength. 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.g.1 Classification of Reference Materials Suitable for Chemical Investigations [9] Parameter Property Chemical composition Additional Remarks Reference materials (RMs). LLC ..1 Classification of reference materials [8]. element. either natural or with added analytes (e.g. density Materials characterized for one or more engineering property values (e... hardness.g.Standard solutions FIGURE 5. PRM MATRIX-FREE .Secondary reference materials.Quality control materials.78 Quality Assurance and Quality Control REFERENCE MATERIALS CERTIFIED MATRIX .

Initial material preparation (grinding. labels. in order to carry out a certification process .© 2009 by Taylor & Francis Group.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. 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. LLC . etc. 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. I. sifting.

which are representative parts of a given production batch. according to ISO recommendations. It is also recommended to pack the RM samples into appropriate containers in the argon atmosphere (bottles with fillers. stable. 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. or ampoules). standard deviations. the type of analytical measurements in which it is going to be used. and the confidence intervals) • Determination of values attested to on the basis of hitherto formulated criteria. 10]. are used mainly for the calibration of measuring instruments and checking analytical procedures [9. It may be achieved in two ways: by sending the same material sample or sending material samples with the same parameters (homogeneous.” The requirements at the production stage. but also to sampling [11]. and appropriately packaged samples. 5. and its availability. which can be minimized by decreasing the content of water in the material to the level of 1–3% of relative humidity. LLC .3 Parameters That Characterize RMs 5. RMs should be prepared in such a way that they are homogeneous. That is why CRMs occupy a higher position in the “metrological hierarchy. are more rigorous. RM can perform its function only when each of its users receives a material with exactly the same parameters. calculating means. precision. It is very important to pay special attention not only to the preparation of stable and homogeneous primary materials. Uncertified RMs.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.© 2009 by Taylor & Francis Group. stable since the moment of production until their use) [9]. stable. and printing the attestation certificate A general procedure for preparing RMs is shown schematically in Figure 5. One should take into account microbiological degradation. penicillin vials. including the laboratory RM (cheaper and more available).2 [8].80 Quality Assurance and Quality Control • Statistical analysis of the obtained results (rejection of deviating results. The selection of the RM depends on the needs at a given time. and have constant characteristics over a sufficiently long period. and the uncertainty in determination of given parameters.3. A certification process involves preparation of a great number of homogeneous. stable during storage. which is reflected in their price and thus their availability. The parameters that characterize CRMs [12–18] are as follows: • • • • representativeness homogeneity stability certified value .

breaking down.g. labels.. etc. LLC . .2 A general procedure for certified reference materials preparation — example for solid CRMs [8]. sieving) Preliminary determination of material’s stability Choosing and purchasing of proper containers. drying.Reference Materials Choosing material Obtaining appropriate amount of desired material Preliminary preparation of material (e. 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.© 2009 by Taylor & Francis Group.

the value of a given parameter of the material should be stable over the whole validity period.© 2009 by Taylor & Francis Group.82 Quality Assurance and Quality Control 5. It is carried out at the stage of distributing the RM into the appropriate containers. plays a decisive role in the production of RMs. oxygen. the manner of processing. LLC . This value should be determined by the manufacturer of the RM and taken into account in the uncertainty budget of the certified value. 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. achievement of the required similarity is not always possible.4 Stability A stability study.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. That is why it is necessary to define the minimum amount (mass) of the RM samples for the study. During storage and transportation.3 Homogeneity Homogeneity study is a comparison of the obtained results for the random samples of the RM. and how to achieve a representative sample of the material for further analysis. the user should be informed about the actual state of the material.3. In such cases. the RM is exposed to the influence of various external factors (temperature. .3. humidity. microbiological activity) that may affect its composition [16]. light. However. next to the homogeneity study. 5. 5. Both sources of heterogeneity of reference materials are presented in Figure 5. A user has no influence on between-bottle heterogeneity of the material.3. A material should be homogeneous and stable.3. but in the process of homogenization and stabilization a change may occur in the connection between the analyte and the matrix. 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.

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

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].3. 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. The aim of material certification is to ascribe certain values of individual properties to a group or individual units. Thus the following solutions are applied [23]: • Measurements at a single laboratory.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 . Certification is based on material sample analyses.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. matrix RMs cannot be certified using direct gravimetric measurement. using one or more methods at one or several laboratories.5 Certified Value RM certification is carried out according to strictly determined rules. using the absolute methods. according to the guidelines presented by Guide to the Expression of Uncertainty in Measurement [25]. in which each of the measurement series is carried out with the highest accuracy and traceability. 5. and must be documented by a complete uncertainty budget. In this case. essential for certification [24].© 2009 by Taylor & Francis Group. by two or more analysts • Interlaboratory studies using one or several various methods. should include all the uncertainty sources described in the following equation [26]: 2 2 2 2 uCRM = ucert + ubott + uls + uss (5. In contrast to pure substances and calibration solutions. an additional stage is required: a complete change or the removal of the matrix. LLC . as described in an appropriate ISO norm [19–22]. that is. including the absolute methods It must be remembered that certification studies should be carried out by the laboratories with supreme and proven competence. The final uncertainty value of the CRM. The reliability of the obtained results of analytical measurements is a self-evident condition.

Reference Materials 85 5. for example.comar. can be found in the following databases available at the Internet websites: http://www-naweb. due to the high heterogeneity of matrix compositions and the wide spectrum of analytes present in the examined samples.org/nahu/nmrm/nmrm2003/browse. therefore.de http://www.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. 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. LLC . .iaea. reputation) • Composition of the sample matrix • Price Detailed information concerning the RMs. A good knowledge of analytical procedures and the available materials is. a key to the right choice. the validity period. and help in finding an appropriate RM.bam. concerning the minimum mass of the RM sampled.virm. and the manner of storage.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.© 2009 by Taylor & Francis Group.htm http://www.

RMs are an essential tool for the determination of accuracy and/or precision. It seems practical to provide a graphical comparison of the reference (certified) value with the value obtained during the measurement (determined one).© 2009 by Taylor & Francis Group.2. LLC .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. Possible situations. • The used and nonused RM cannot be placed back into the container. TABLE 5. depending on the information on the two compared values. together with the associated conclusions are presented in Table 5. this book presents the basic formulas and correlations that help in selecting the manner of documenting the values of the determined parameters. Because one of the main problems associated with this process is the interpretation and numerical presentation of the determined parameter.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 .

57 4.76 4. LLC .© 2009 by Taylor & Francis Group. μg/g: 1 2 3 4 5 4.64 ± 0.94 5.28 μg/g.1 Problem: Five independent determinations of total mercury were carried out for the samples of the certified reference material NRCC-DORM-2 — dogfish muscle. Data: result series.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. Using a graphical method. The certified value given by the manufacturer is 4.04 4. test the agreement of the obtained value with the certified value.82 .

Excel file: exampl_RM01. μg/g: 1 2 3 4 5 6 Solution: Mean SD U (k = 2) 0. test the agreement of the obtained value with the assigned value.xls Example 5.© 2009 by Taylor & Francis Group. LLC .019 μg/g 0.372 μg/g 0.83 μg/g 0.16 μg/g Determined Conclusion: An obtained value agreed with the certified one.38 0.34 0.39 0.35 0.40 . The assigned value given by the manufacturer is 0.88 Solution: Mean SD U (k = 2) Graph: 6 5 4 3 2 1 0 CRM Quality Assurance and Quality Control 4.37 0. Data: result series.18 μg/g 0.2 Problem: Six independent determinations of total mercury were carried out for the samples of the reference material GBW 07601 — powdered human hair.36 µg/g Using a graphical method.023 μg/g 0.

© 2009 by Taylor & Francis Group. A comparison of the standard deviation values in the series of measurements for CRM.15 0. and the comparison of the determined values with the certified value. The following options are feasible: 1.3 0.2) .Reference Materials Graph: 0.1 0. 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. LLC .2 0.45 0.35 0.25 0. with the value of expanded uncertainty for CRM.xls An alternative solution is to determine the conformity of the reference value with the determined value using appropriate tests.3) SDdet n < UCRM (5. Excel file: exampl_RM02.05 0 89 RM Determined Conclusion: An assigned value is in the range of obtained value ± uncertainty.4 0.

Using the aforementioned method.4 µg/g. Data: result series.76 4.28 µg/g 4. µg/g: 1 2 3 4 70. test the agreement of the obtained value with the certified value.2 71.3 Problem: Five independent determinations of total mercury were carried out for the samples of the certified reference material NRCC-DORM-2 — dogfish muscle.64 µg/g 4.28 μg/g.94 5. Using the aforementioned method.83 µg/g 0.18 µg/g 5 0.xls Example 5. Data: result series.36 µg/g 4.6 . μg/g: 1 2 3 4 5 Solution: xdet SDdet n SDdet n UCRM xCRM xCRM − UCRM xCRM + UCRM 4. The certified value given by the manufacturer is 4. The certified value given by the manufacturer is 68.04 4.92 µg/g 4.2 ± 1.90 Quality Assurance and Quality Control Example 5.8 70.4 Problem: Four independent determinations of lead were carried out for the samples of the certified reference material NIST-SRM 1633b — coal fly ash.57 4.82 SDdet n < UCRM xCRM − UCRM < xdet < xCRM + UCRM Conclusion: An obtained value agreed with the certified one.4 69.64 ± 0.080 µg/g 0. LLC . Excel file: exampl_RM03. test the agreement of the obtained value with the certified value.© 2009 by Taylor & Francis Group.

4) does not allow for the uncertainty of the certified value.5 µg/g 0. using the uncertainty values for both the values.Reference Materials Solution: xdet SDdet n SDdet n UCRM xCRM xCRM − UCRM xCRM + UCRM 70.68 µg/g 4 0.6 µg/g 91 SDdet n < UCRM xCRM − UCRM < xdet < xCRM + UCRM Conclusion: An obtained value did not agree with the certified one.2 µg/g 66. LLC .342 µg/g 1.5) . Comparison of the certified value with the determined value.xls 2. Formula (5. Excel file: exampl_RM04. The following correlations are examined: x det − xCRM < 2 u(2xdet ) + u(2xCRM ) (5.4) 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 − 1.8 µg/g 69. Application of Student’s t test. 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. The value of parameter t is calculated according to the formula: t= x det − xCRM SDdet n (5.6) t= x det − xCRM u(2xdet ) + u(2xCRM ) n (5.4 µg/g 68.© 2009 by Taylor & Francis Group.

Using the aforementioned method.323 Conclusion: An obtained value agreed with the certified one. The certified value given by the manufacturer is 4.92 Quality Assurance and Quality Control x det − xCRM ≥ 2 u(2xdet ) + u(2xCRM ) (5.76 4. μg/g: 1 2 3 4 5 Solution: xdet SDdet n u( xdet ) u( xCRM ) xdet − xCRM 2 u 2 ( xdet ) 4. test the agreement of the obtained value with the certified value.4 μg/g.83 0. The certified value given by the manufacturer is 68.82 4.64 ± 0.04 4.© 2009 by Taylor & Francis Group. Example 5.2 ± 1.7) Satisfying the first relation implies conformity of the determined value with the certified value.94 5. and satisfying the second relation denotes the lack of conformity between these values.xls Example 5.186 2 ( xCRM ) xdet − xCRM < 2 u(2xdet ) + u(2xCRM ) xdet − xCRM ≥ 2 u(2xdet ) + u(2xCRM ) +u 0.6 Problem: Four independent determinations of lead were carried out for the samples of the certified reference material NIST-SRM 1633b — coal fly ash.28 μg/g.5 Problem: Five independent determinations of total mercury were carried out for the samples of the certified reference material NRCC-DORM-2 — dogfish muscle. Data: result series. LLC .080 0.18 5 0. Excel file: exampl_RM05.57 4.140 0. .

The reasoning is carried out using the following relations: • If Z ≤ 2. which can be calculated as the combined uncertainty of the certified value and the determined value. then the determined value did not agree with the reference value. test the agreement of the obtained value with the certified value.30 1.xls 4.700 2.9) . Excel file: exampl_RM06.68 4 0. The value of the Z score is calculated using the following formula: Z= x det − xCRM s (5.4 69. due to application of CRMs.© 2009 by Taylor & Francis Group.2 71. can be presented as recovery and should be calculated according to the following equations: R= x det ⋅ 100% xCRM (5.56 70.Reference Materials Using the aforementioned method. μg/g: 1 2 3 4 Solution: xdet SDdet n u( xdet ) u( xCRM ) xdet − xCRM 2 u(2xdet ) + u(2xCRM ) 70.8) where s is the value of a deviation unit.342 0. then the determined value agreed with the reference value. Data: result series. • If Z > 2. Trueness value. The application of Z score.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. LLC .5 0.8 70.

xls .140 2 104.57 4.28 μg/g.94 5.83 4.18 5 0. The value of trueness is usually given as: Trueness = R ± U (5.0% 6. LLC .64 0.8% 4. the calculated value of trueness is acceptable.94 Quality Assurance and Quality Control U = k⋅ (u 2 ( xdet ) + u(2xCRM ) x det + xCRM 2 ) (5.11) and is most frequently expressed in %. μg/g: 1 2 3 4 5 Solution: xdet xCRM SDdet n u( xdet ) u( xCRM ) k R U 4.64 ± 0. Data: result series.76 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.080 0.82 Conclusion: A value of 100% is in the range of calculated trueness value. Using the obtained result. Example 5. The certified value given by the manufacturer is 4.© 2009 by Taylor & Francis Group. calculate trueness as a recovery value for k = 2.10) The reasoning should be based on the following: if the range R ± U includes value 100%.04 4. Excel file: exampl_RM07.

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

” Microchem. Qual. .. 249–257. Geneva.M. Van der Veen.P. and Siekmann. A. 2000.. Anal. Anal. 6.” Fresenius J. A.M. “Uncertainty calculations in the certification of reference materials. Chim.. Schimmel. and Tack. 24... H. Assur. ISO Guide 35. Quality system guidelines for the production of reference materials. 28. Rev. “Uncertainty calculations in the certification of reference materials. 370. 395–399. 59.. 360. “Some difficult problems still existing in the preparation and certification of CRMs. 257–263. Fe. H. 6. Qual. Linsinger. A.H. 136–143. and Schimmel. Pauwels.. T. 1999. H. Assur.. 1998. and Pauwels. “Estimation of the CRMs in accordance with GUM: Application to the certification of four enzyme CRMs. 5.P. Qual. ISO Guide 31. F. 359–361... Danko. ISO. 25. J.. Anal. 6.. 361. Pb. T. Assur. 368. Linsinger. R.. “The role of reference materials and reference methods in chemical analysis. “The study of the stability of reference materials by isochronous measurements.L. 5. Caroli.. Linsinger. 30. 2001. Lamberty. J. and Pauwels..J.. Chem.. Chem.P. 2. Van der Veen.H. A. and Iamiceli. Pauwels. and Schimmel. H. T. J. Principles of analysis of variance. J.. 2000. A.” Microchem. A. Uriano. “ICP-AES and ICP-MS quantification of trace elements in the marine macroalga Fucus sample. 17. ISO. Geneva... 2001. L. Geneva.A...” Crit. “Uncertainty calculations in the certification of reference materials. 6. and Pauwels. Robouch. Geneva. Reference materials — Contents of certificates and labels. Characterisation and certification. Linsinger...H. and Gravatt.. J. A. O. Forte. Caroli.. G. ISO.H.. 95–99. Senofonte. A. 16.” Accred.” Fresenius J.J.G. Cu.. “Quantification of the expected shelf-life of certified reference materials.” Anal. and Polkowska-Motrenko.A.1Lamberty. J. J. A.. H.. Van der Veen. 29.. S.. C.. 1977. S.. S. Schimmel. Trends and definitions used in connections with reference materials. 27.” Accred.. a new candidate certified reference material. ISO Guide 34.. 454.. 464– 469. P. “Certified reference materials for research in Antarctica: the case of marine sediment. “Uncertainty calculations in the certification of reference materials. Stability study. Certification of reference materials.M. 26–30. Homogeneity study.. Chem. Schumann.. “Evaluation of uncertainty of reference materials. 2002. G. 1996. Lamberty. 19. Veen. 15.M.M. 361–411... Lamberty. 290–294. 23. Pauwels. 3.N. General and statistical principles.” Fresenius J. 589. 21. J.. A. Chem. R. T.. 4... 20. Sutherland. Lamberty.. 62. A. Chem.H.” Accred. 1993. Qual.. 1998.. 26. and Kramer. Schimmel... 2000. Pauwels.. “Determination of Al. Anal.. G. Assur. 2001.96 Quality Assurance and Quality Control 13. and Pauwels. A. Geneva. J.” Accred. 244–250. G. 1992. 2001. H. 1. 18. Qual. 14..C. and Zn in certified reference materials using the optimized BCR sequential extraction procedure. ISO Guide 30. J. 1998. and Pauwels. 22. Lamberty. A. Anal. 126–130. Caimi. Assur. B.J. Mn. Dybczyński. LLC . Van der Veen. Acta.. J..M.” Accred... Guide to the Expression of Uncertainty in Measurement (GUM). 2000.. Van der Veen.” Fresenius J.. 1989.© 2009 by Taylor & Francis Group.

which can be especially significant for laboratories with accreditation or those applying for accreditation. undertake rectifying action [4].g. 2] Interlaboratory comparisons — organization.6 Interlaboratory Comparisons 6. 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. 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.1 Definitions [1. and in the case of error detection. performance. 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.© 2009 by Taylor & Francis Group. usually with a determined uncertainty.. Proficiency testing — determination of laboratory testing performance by means of interlaboratory comparisons. One of the most crucial means of that monitoring is participation in various interlaboratory studies [3]. A generalized scheme for conducting interlaboratory studies is shown in Figure 6. Certification study — a study that assigns a reference value to a given parameter (e. Moreover. and evaluation of tests on the same or similar test items by two or more laboratories in accordance with predetermined conditions. 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. 6. LLC . Participation in these programs gives a laboratory a chance to compare its results with those obtained by other laboratories and to prove its competence.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. 97 .1 [5].

6.© 2009 by Taylor & Francis Group. LLC .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 6.3 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 .1 A generalized outline for conducting interlaboratory studies [5].

and determinations as well as other details of the study are presented in the research protocol prepared by the organizer of the study. it is possible to compare a few procedures. 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).and interlaboratory precision Systematic error Recovery value Internal parameters of quality assurance Sensitivity Limit of detection Applicability limit In this type of research.© 2009 by Taylor & Francis Group. The obtained sample results are compared with the results obtained by other laboratories or with a known or determined (guaranteed) reference value. all participating laboratories apply the same set of guidelines for each procedure. test samples. This research may be conducted among laboratories that are . and the statistical analysis of the obtained sets of results is conducted separately for each of the procedures. LLC . 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. • By using the same materials or test samples. both on a national and international scale. The obtained results are applied in estimating the characteristic parameters of the procedure: • • • • • • • Intra. • The number of participants. analyte concentration. with regard to the composition of the matrix.Interlaboratory Comparisons 99 Laboratories that wish to confirm their competence should participate in at least one program of interlaboratory research. 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. 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.

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

Moreover. to check if the sample has not changed in an undesirable fashion. 6. 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. the participants try to find the causes of such discrepancies. The achievement of these aims requires a painstaking and reliable organization of this research. interlaboratory comparisons are retrospective studies. 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. all participants gather to discuss the obtained results. with the cooperation of a control center. In this case. Taking into consideration the sample preparation used by participating laboratories. • 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). It generally takes a long time before the participants get to know the obtained results.4 Characteristics and Organization of Interlaboratory Comparisons As one can gather from this current discussion. produces a precise localization of sources and causes of errors and hence an improvement in the quality of analytical results. There are certain limitations associated with performance and participation in proficiency testing. LLC .Interlaboratory Comparisons 101 Sometimes. proficiency testing is unusually time consuming. pilot studies are implemented to select the participants with suitable qualifications to participate in the actual proficiency studies. the so-called key comparisons. With regard to conditions. It gives laboratories a chance to improve their competence. In reality. Moreover. After the initial research.© 2009 by Taylor & Francis Group. a sample may be taken back to the coordinating laboratory before a test by a subsequent participant. 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. • 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). . and improve their performance in the next proficiency test. correct the hitherto existing mistakes. In the case of results distinctly deviating from the assumed range of acceptable results. a thorough analysis of an analytical process. which is why proficiency testing may not affect any decision on quality management. • Those examining the competence of laboratories during the performance of various types of analyses. First of all. each of the aforementioned types may be divided into three further categories: • Samples circulate successively from one laboratory to another.

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

and the selection of suitable tests and solutions depend on the type of research. the accuracy (and/or precision) of . The ultimate aim of all types of studies is to determine.© 2009 by Taylor & Francis Group. based on experimentally obtained numerical data. LLC .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.xls The manner of conducting a statistical analysis of results obtained in interlaboratory comparisons. Respective documents define the precise manner of conduct for a specified type of research.

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

5 15.5 1.5 31.5 15.5 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 5.5 12.5 2.5 4.5 11.5 61. LLC .5 3.xls 3.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 71.© 2009 by Taylor & Francis Group.5 1.5 5.5 4.5 4.5 5.5 5.5 61.5 15.5 66.

66 0.70 11.1 12.1 11.516 0.xls Example 6.423 0.31 0.4 11.0 13.47 12.73 13.26 0.© 2009 by Taylor & Francis Group.8 12.8 11.5 13.6 12.3 Problem: Find outliers in the given sets of measurement results obtained in interlaboratory comparisons.1 10.6 14.43 13. Apply the Grubbs’ test for one outlier to examine the interlaboratory variability.82 0. LLC .0 14.6 13.4 12.070 0. .4 Problem: Find outliers in the given sets of results obtained in interlaboratory comparisons from Example 6.10 12.463 0.106 Quality Assurance and Quality Control Example 6.430 0.33 11.9 11.1 12.1 11. 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 12.01 SD 0. Use the Cochran test to examine the intralaboratory variability.093 0.53 n p C C0.05 C0.1 12.65 0.8 13.68 0.4 13.76 0.67 3 8 0.7 Conclusion: Result obtained by laboratory “lab 6” is correct.5 14.583 0.6 13.3.211 0.670 0.615 SD2 0.8 11. Excel file: exampl_PT03.1 12.443 12.40 13.

881 1.6 12.6 13.1 11.688 min 2. When significant differences are found between the values of random errors (statistically significant differences in the variance values).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 G0.6 14.5 13.274 2. The obtained numerical data are divided into m groups according to their origin (m is the number of laboratories).1 10.47 12.53 3 8 12.01 G0.43 13.6 13.8 13.1 12.0 14.05 Mean 12. 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).4 12.588 0.73 13.4 11.126 12.© 2009 by Taylor & Francis Group.1 12. . one may perform a one-factor (one-dimensional) variance analysis (ANOVA).10 12.8 11.1 12. Excel file: exampl_PT04.1 11. LLC .xls To simultaneously determine the standard deviation as the measures of repeatability and reproducibility.4 13.5 14.8 12.9 11.7 107 Conclusion: Result obtained by laboratory “lab 5” is correct.70 11.40 13.8 11.1 12.0 13.33 11.

f ) increases considerably and the precision of the evaluation decreases. rejecting the outliers. or when for some reason one expects deviation of the obtained measurement results from the normal distribution. A total error. The parameter that is most often used to evaluate the obtained results in interlaboratory comparisons is the Z score parameter. Use Z score. Below 4 degrees of freedom. that is.and intergroup degrees of freedom and calculate the standard deviation within individual groups and among the groups. with the identical value of the variance SD2.1. the value of the parameter t(α.15). find which results are satisfactory. The number of parallel determinations that is greater than 5 occurs only in special cases. Situations in which a single factor completely explains a given phenomenon are rare.© 2009 by Taylor & Francis Group. Draw a graph with Z score values for each laboratory. the standard deviation being the measure of the respective variances. the reference values and the standard deviation sample are calculated according to all the results as the mean value and standard deviations after. Example 6. 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. The lower influence on the size of the certainty range is exerted by the number of parallel analyses conducted at a given laboratory. It shows that the interlaboratory studies should involve at least five laboratories. by the sum of squares describing the variability within groups and the sum of squares describing the variability among groups. consists of a few errors that are summed up according to the law of error propagation. characterizing the results obtained by using an analytical procedure. of course. on the number of laboratories participating in the research. the total sum of the squared deviations from all measurements from the mean.5 Problem: In the series of measurement results given in Example 6. Data: results: Data lab 1 lab 2 lab 3 lab 4 lab 5 . and which are unsatisfactory.8. The essence of variance analysis is the division of the total variability. The manner of calculating this parameter has been described in detail in Chapter 1 (Section 1. Then one should determine the total intra. LLC 123 111 128 138 121 . which are questionable.108 Quality Assurance and Quality Control the data are joined into groups for which the variance values are not statistically significantly different. and then the variance analysis is conducted for each group. The reliability of conclusions depends. to a great extent. 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.

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.© 2009 by Taylor & Francis Group.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 −0.51 −0.55 −0.16 −1.40 2.4 8.40 8.08 0.5 .90 0.97 8.28 −1.58 0.51 −1.61 −0.16 7.08 0.10 −0.43 1.40 −0.28 −0.78 3.69 0.22 7.58 xm SD .

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 .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. which are questionable.1.6 Problem: In the series of measurement results given in Example 6. Use Z score. Example 6.00 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.© 2009 by Taylor & Francis Group. and which are unsatisfactory.00 0.00 8. find for a given reference value which results are satisfactory.00 –2.110 Graph: 10.00 .00 Z-score 4. Draw a graph with Z score values for each of laboratory.00 2. LLC Lab 17 Lab 2 Lab 8 Lab 5 Lab 1 Lab 6 Lab 3 Lab 4 Lab 7 Lab 9 –4.00 6.

42 −1.42 −0.42 8.24 6.01 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. LLC .13 −2.5 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 .13 −3.67 −2.94 −1.01 −3.26 −2.30 −2.67 −3.13 6.06 2.24 −2.24 0.77 −0.24 −1.© 2009 by Taylor & Francis Group.07 5.24 −2.85 −1.

for the given reference value and the combined uncertainty reference value. LLC Lab 17 Lab 2 Lab 8 Lab 5 Lab 1 Lab 6 Lab 3 Lab 4 . Draw a graph with the Z score values for each laboratory.00 Z-score 2.00 0.00 4. and the value of the reference combined uncertainty for a given material.112 Graph: 8.7 Problem: In the series of measurement results given in Example 6.00 6.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 . or unsatisfactory.00 –2.© 2009 by Taylor & Francis Group. Example 6. Use Z score. find which of the results are satisfactory. questionable.1.00 –4.00 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.

64 −2.73 5.82 −1.00 −1.73 −1.55 −2.73 −1.36 4.64 −1.18 −1.73 0.64 −1.64 4.55 4.18 −1.36 −2.09 −0. LLC .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 −1.82 1.64 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 .36 −1.© 2009 by Taylor & Francis Group.27 −0.36 −0.

1.00 4. use Z score. taking into consideration the combined uncertainty reference value.© 2009 by Taylor & Francis Group. LLC Lab 17 . Example 6.xls • Known mean values obtained in participating laboratories and known value of the reference combined uncertainty for a given material.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_PT07.00 5.114 Graph: 6. Draw a graph with the Z score values for each laboratory.00 –2.00 3.00 0.8 14 16 10 11 14 18 23 15 .00 –3.00 Lab 2 Lab 22 Lab 8 Lab 5 Lab 11 Lab 16 Lab 10 Lab 21 Lab 1 –4.00 –1.00 1.8 Problem: In a series of measurement results given in the Example 6.00 Z-score 2. 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.

67 −0.09 −1.92 −0.43 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 .01 −1.70 −1.88 −0. LLC .93 −0.10 −1.12 −0.09 2.51 0.28 −1.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 −1.22 0.50 −0.97 −1.40 −0.11 −1.© 2009 by Taylor & Francis Group.85 3.23 1.80 1.97 −0.

00 –2.00 . the evaluation is not satisfactory where x denotes the relative systematic error (relative deviation).00 Quality Assurance and Quality Control Excel file: exampl_PT08. It is calculated in instances when participants of a given study use various methods to evaluate the obtained results.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 –4.1) .00 –3. assumed as a limit (permissible).xls Another parameter of the individual examination of measurement results is relative error. ε= ( xlab − xref ) 100% xref (6. the evaluation is satisfactory • If ε > x .00 0.00 3.00 –1.116 Graph: 4.00 Z-score 1. 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. It is assumed that: • If ε ≤ x .© 2009 by Taylor & Francis Group.

0% .9 Problem: From the data given in Example 6. calculate the values of the relative errors and make an evaluation for the permissible error value ±20%.© 2009 by Taylor & Francis Group.1 −20.3 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 20.6 −1.1. LLC .7 −8. 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.4 −13. % lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 −12.6 −12.Interlaboratory Comparisons 117 Example 6.1 34.

4 −6.1.7 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.7 −5.© 2009 by Taylor & Francis Group.9 −13. the standardized Z coefficient. Example 6.6 41.16).6 1. En is a parameter that is decidedly less restrictive than.9 −13.8 14 16 10 11 14 18 .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. 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.xls −18. for example.6 34.3 −12. The method of its determination is described in detail in Chapter 1 (Section 1. solely attributable to the high value of the extended uncertainty.9 −12. may be considered an outlier.9 37. 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. Results that are deemed satisfactory may include values significantly deviating from the mean.8. because of the inclusion of the uncertainty value.9 −20. but within the accepted interval.4 −10. LLC .10 Problem: For the data given in Example 6. apply En Score.7 −7.

01 −1. LLC .10 −1.97 −0.11 −1.92 −0.© 2009 by Taylor & Francis Group.xls En −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_PT10.28 −1.70 −1.50 −0.85 3.22 0.12 −0.67 −0.09 −1.40 −0.43 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 .93 −0.88 −0.80 1.97 −1.23 1.51 0.09 2.

then whiskermax is not marked on the diagram. the minimum value in the set of results. after which they are all put into one diagram. as 1. then whiskermin is not marked on the diagram. LLC . Whiskermax. one should divide all the measurement results obtained for a given sample into subsets. whiskers. Then. the maximum value in the set of results. Whiskermin. a diagram (plot) is drawn (separately for a given set of results). separate plots are drawn. 2. On the 0Y axis. In the graphical presentation of results. not smaller than the limit equal q1 – 1. 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.5. for a given series marked by one point on the 0X axis. . 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). for each subset. In drawing such a plot. and which procedure yields more accurate data.120 Quality Assurance and Quality Control 6. 3. whiskers are marked as: a. if the calculated value is equal to q1. in the following manner: 1. On the same plot. each containing results obtained using a specific analytical procedure. the difference between q3 and q1 • Determination of maximum values. if the calculated value is equal to q1.© 2009 by Taylor & Francis Group.1 Comparisons of Results Obtained Using Various Procedures In this type of comparison. Based on data for which the diagrams (plots) are drawn.5 · IQR. it is possible to conclude which of the analytical procedures were used more often. box plots may be used.5 times the IQR Based on calculated values. b. Results out of this range (lower than whiskermin or higher than whiskermax) are marked as outliers. one may examine if the results obtained using various analytical procedures differ among themselves in a statistically significant way.5 · IQR. Due to the type of construction of the graph. not smaller than the limit equal q3 + 1.

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 29.5 121.3 141.11 Problem: For the data given in Example 6.5 × IQR q3 + 1. LLC .5 × IQR min max whiskermin whiskermax 126.Interlaboratory Comparisons 121 Example 6.6 111 198 111 158 123 111 128 138 121 123 188 114 188 122 121 142 125 132 129 121 198 131 158 193 122 111 u 11 9.5 × IQR q1 − 1.6 170.0 19. construct a box plot graph.6 91.1.© 2009 by Taylor & Francis Group.8 14 16 10 11 14 18 23 15 11 13 12 17 19 21 28 14 18 13 14 17 .

0 120.0 40.0 60.0 160. LLC .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 20.© 2009 by Taylor & Francis Group.0 0.0 80.0 100.0 140.0 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 .

Application of this graph shows which of the participating laboratories achieved comparable results and which laboratory obtained deviating results. to determine the presence of systematic errors.0 140.xls 6.0 80.0 100.0 160. • 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).0 40.and interlaboratory variability. .Interlaboratory Comparisons Graph — modified (box plot): 180.0 20. or • Determinations for two different samples In this case.0 0. The graph is constructed as follows: • Measurement results for both the obtained series are marked on the X and Y axes.0 120. • Solid lines are drawn (both vertical and horizontal) that reflect the values of main distribution estimators (arithmetic mean or median).0 60. LLC .5.2 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.0 xLab 123 Excel file: exampl_PT11. It is an easy and very effective method of comparing both intra.© 2009 by Taylor & Francis Group. a graphical method — also called the Youden diagram [8] — may be used.

4 Series 2 12.7 10.8 .7 10. It may indicate a positive or negative bias in the analytical procedure applied in a given laboratory.8 12.2 11. Example 6. LLC .8 11.2 10.8 12.4 11 12.3 11. the results are distributed in a random manner around the mean (median).2 15. 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 11.9 11.8 14. then the majority of points are in the upper right or bottom left quarter of the graph.7 11.9 10.5 10. 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).5 Median 11.8 11 10.9 11.6 11. When the main cause of the deviations from the mean or median are random errors.© 2009 by Taylor & Francis Group.3 11.124 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. produce a Youden graph.7 11.12 Problem: For the two given series of measurement results for two examined samples obtained in the examining laboratories.1 12.7 13.2 11.

8.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. quite common method of graphical presentation of the measurement results obtained by comparing laboratories is the application of Mandel h and k tests. 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.© 2009 by Taylor & Francis Group.xls Another. .17). The manner of conducting Mandel h and k tests is described in Chapter 1 (Section 1. 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. LLC .

03 2.34 6. Draw a graph showing the respective values of the calculated h parameters characterizing the sets of results obtained in individual laboratories.44 4.2 13.1 Level 5 11.23 4.8 12.© 2009 by Taylor & Francis Group.67 9.01 2.12 1.11 4.5 11.11 5. 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.78 1.7 lab 2 lab 3 lab 4 lab 5 lab 6 . LLC .65 2. Example 6.83 7.22 2.3 11 12.126 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.56 Level 4 14. 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. Moreover.7 11 10.48 2.2 14. it shows a smaller repeatability of results obtained by the laboratory compared with the rest of the participants.5 14.2 14 13.73 5.5 10.08 2.2 12 11 11.02 Level 2 7.9 14. When a laboratory tends to obtain only negative values for h.21 7.23 8. for example. Similarly.4 10. calculate the values of Mandel’s h parameter. when a laboratory yields h values in the extreme range.11 Level 3 2.6 10 14.4 15. it achieved an unusually high number of large values for h.22 5.3 13.8 12.34 4.01 8.11 2.45 1.11 11.4 15.77 7.32 4. Data: results: Level 1 lab 1 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 4.54 5.55 2.21 6. one may suppose that there is a source of bias for the results obtained by that laboratory. the situation should be adequately explained.22 4.45 8.7 13.34 7.23 4.8 11.54 7.92 8.22 2.13 Problem: For a given set of results obtained in interlaboratory comparison.77 7.02 8.03 5.15 2. When the graph of the statistical parameter k indicates that a given laboratory deviates from the rest due to numerous high values.98 1.4 11.56 4.56 4.23 5.8 14.67 7.54 7.34 2.8 16. and at the same time different from the sign (plus or minus) of parameter h obtained in other laboratories. one should pay attention to the problem of systematic errors and the small repeatability of results (great variance value).86 2.3 6.44 1.89 1.8 9.76 4.12 2.96 10.73 2.98 4.

33181 −0.76 8.24180 0.3 8.6 11.3 11.35 3.21952 1.21 3.49321 0.21330 0.56151 0.11 9 2.82 127 lab 8 Solution: Level 1 lab 1 lab 2 lab 3 lab 4 lab 5 lab 6 lab 7 lab 8 Graph: 2.06872 −2.63583 0.67 8.71269 0.5 –3 Lab 1 Lab 2 Lab 3 Lab 4 Lab 5 Lab 6 Lab 7 Lab 8 Level 2 −0.22 1.2 11 12 12.11 8.72498 −0.10230 −2.02 7.51103 −0.45130 0.19003 −1.5 2 1.5 1 0.18460 −0.99211 0.32112 0.11 2.51103 0.26898 0.56 1.2 13.38729 1.15017 −2. LLC .08 2.46858 −1.5 –2 –2.25142 Level 4 0.xls . Results obtained by the other laboratories are within the permissible range of changes for all the determined analytes.13474 Level 3 h 0.22323 −0.5 0 –0.85611 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.5 –1 –1.84978 0.88 8.26092 0.8 11.57 13.23756 −0.44 4.66888 1.15886 0.88856 0.40861 −1.55 4.1 13. which indicates the occurrence of a systematic error source for the results obtained by this laboratory.25261 Level 5 0.35552 −0.© 2009 by Taylor & Francis Group.Interlaboratory Comparisons lab 7 1 2 3 1 2 3 4.32 3.45 8. Excel file: exampl_PT13.83572 0.04319 −1.02508 0.77645 1.15118 1.45 1.98 9.

7 11 10.22 1.12845 0.11 9 Level 3 2.3 8.54 5.3 6.68527 1.26287 0.8 11.1 13.77 7.54 7.12845 0.68527 1.71646 1.26287 0.11 2.86 2.11 5.2 12 11 11.89 1.35 3.74083 0.02 7.08 2.55 2.74083 0.9 14.2 14 13.67 8.56 1.8 16. 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 4.34 6.56 4.2 14.44 1.8 11.37413 0.23 5.73 2.22 2.56 4.65 2.57 Level 4 14.98 4.22 4.71646 1.5 10.74083 0.98 9.45 1.3 11 12.74823 1.15 2.8 12.6 11.26287 0.74823 1.21 3.68527 1.83 7.74083 0.11 11.67 9.77 7.67 7. LLC .2 13.45 1.22 2.7 13.55 4.8 14.44274 0.21 6.12 1.5 11.4 15.68527 1.12 2.11 4.8 12.48 2.37413 0.128 Quality Assurance and Quality Control Example 6.44274 0.44274 Level 4 Level 5 .08 2.26287 0.37413 0.88 8.78 1.73 5.68527 1.34 4.8 9.12845 0.76 8.4 11.74823 1.11 8.37413 0.74083 0.71646 1.03 5.71646 1.23 4.02 8.3 13.23 8.56 2.6 10 14.82 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 0.22 5.92 8.11 2.76 4.12845 0.01 8.34 7. Draw a graph showing the respective values of the calculated k parameters characterizing the sets of results obtained in individual laboratories.44274 Level 3 k 0.74823 1.54 7.14 Problem: For a given set of results obtained in an interlaboratory comparison.2 13.32 3.23 4.21 7.74823 1.03 2.02 4.4 10.11 8.37413 0.01 2.© 2009 by Taylor & Francis Group.32 4.26287 0.4 15.34 2.44274 Level 2 0.1 13.44 4.2 Level 5 11.96 10.12845 0.98 1.3 11.44 4.7 11 12 12. calculate the values of Mandel’s k parameter.45 Level 2 7.5 14.71646 1.45 8.

However. and not those that may occur. lab 5. laboratories for many years have participated in various interlaboratory comparisons.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. It is hence obvious that laboratories that do not participate in these comparisons should be deemed unreliable. the obtained values of repeatability exceed the critical value for the 5% level of significance. • Results of interlaboratory studies enable the detection and definition of current problems in a given laboratory. while interpreting the results of the interlaboratory studies.xls 6. Excel file: exampl_PT14.5 1 0.” In the case of individual results (lab 1. and lab 6). and enables issuance of opinions on organizational procedures. 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. In other words. High marks/grades obtained in interlaboratory proficiency studies indicate a high quality of analyses performed by the participating laboratory. .5 2 1. both on national and international scale. LLC . 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.Interlaboratory Comparisons Graph: 2.5 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. Bearing this in mind. one should remember that: • Participation in interlaboratory studies must not serve as a substitute for routine intralaboratory control of the results’ quality. 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.© 2009 by Taylor & Francis Group.

A. Analytical Methods Committee. LLC . Proficiency testing by interlaboratory comparisons: Part 2. D. Assur. 331. 8.” Crit. Use. Konieczka. 1997. “Statistical evaluation of interlaboratory tests. 513–519. Chromatogr.” Accred. Kandel..0. Davies.pdf.” Accred. 11. Tholen. “Proficiency Testing of Analytical Laboratories: Organization and statistical assessment.l‑a‑b.com).. Available at http://www.. “The role of and place of method validation in the quality assurance and quality control (QA/QC) system. 11. M.. 158–167. Thompson.” J. 7. Linsinger. 1999.P.W.” Accred Qual. ISO/IEC Guide 43-1. “Statistical treatment of proficiency testing data.” Fresenius Z. Anal. Anal. “Set-up and evaluation of interlaboratory studies. 1992.R. Selection and use of proficiency testing schemes by laboratory accreditation bodies. 3. ISO/IEC Guide 43-1. 1158. Assur. 12. “Fitness for purpose — the integrating theme of the revised harmonized protocol for proficiency testing in analytical chemistry laboratories. Juniper... 336–341.. 322–327. and Ellison.. P.. 2. 362–366. 3. 117. Qual. P.© 2009 by Taylor & Francis Group. I. 173–190. Proficiency testing by interlaboratory comparisons: Part 1. Chem.J. 1988. Development and operation of proficiency testing schemes.” Accred. and Smeyers-Verbeke... ProficiencyTesting. Chem. 2000. 3.com/webshared/administration/ Hows%20And%20Whys%20of%20of%20.. References 1. 1998. 6.130 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. Y..L. Eurachem Guide on Selection. 2007. 37. and Grasserbauer.. W. 1997. Vander Heyden..R. 10. 97–104. “Quality issues in proficiency testing.HN-Proficiency. 9. 4. 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.L. 4. . 2006. S. “The influence of different evaluation techniques on the results of interlaboratory comparisons. Rev. Assur.. 373–378... Krska. Assur. To sum up. Edition 1. R. M. 1998. J. Qual. and Interpretation of PT Schemes. Qual. T. 5. 2007. the same applies to an analytical method.” Analyst.

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

r Graph: 8 7 6 Signal 5 4 3 2 1 0 0 2 4 6 8 10 12 14 y = 0. Data: results: Data x 1 2 3 4 5 2 2 2 4 4 y 1.00989 0.07702 0.5642x – 0. Apply Student’s t test. LLC .11 2.05.2 Problem: Using the data from Example 7. a Residual standard deviation. b Intercept.© 2009 by Taylor & Francis Group.1.9951 139 18 0.20 1.Method Validation Solution: n Slope.0291 0. examine the significance of the differences in the slope and the intercept of a calibration line and the value 0.xls Example 7.0291 R2 = 0.5642 −0. SDxy Standard deviation of the slope. SDb Standard deviation of the intercept.08 2.1433 0. SDa Regression coefficient.9976 Content Excel file: exampl_valid01. Calculations should be performed for the significance level α = 0.12 1.32 .

1433 0.20 .67 5. b Intercept.32 4.xls Example 7.44 5. r tb ta tcrit 18 0.3 Problem: Using the data from Example 7. draw a graph of sustained response. Data: results: Data x 1 2 2 2 y 1. a Residual standard deviation. SDb Standard deviation of the intercept. SDa Regression coefficient. SDxy Standard deviation of the slope.66 n Slope. No statistically significant difference between the intercept and 0.© 2009 by Taylor & Francis Group.54 3.07702 0.97 6.41 4.78 6.00989 0.120 Conclusions: Statistically significant difference between the slope and 0.33 3.23 3.51 6.378 2. marking the lines of the interval for the values deviating ±5% from the mean.12 1. LLC .76 5.062 0.9976 57.5642 −0. Excel file: exampl_valid02.12 4.0291 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 2.1.

58 0.53 xm + Interval % 0.58 0.57 0.76 5.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 1.57 0.56 0.67 5.44 5.57 0.52 0.60 0.59 0.78 6.11 2.54 0.41 4.56 0.© 2009 by Taylor & Francis Group.56 0.53 0.51 6.33 3.66 141 Solution: y/x 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0.32 4.97 6.56 xm 0.59 .12 4.54 3.56 xm − Interval % 0.58 0.32 2. LLC .55 0.23 3.56 0.08 2.54 0.

08 2.4 Problem: Using the data from Example 7.142 Graph: 0.33 3.41 4.12 1. LLC .51 . and the experimental value to be the value calculated from the calibration curve equation.44 5.65 0.70 0.45 0.76 5.60 0.55 0. calculate the values of the relative errors for individual values x.75 0.54 3.32 2.67 5.xls Example 7.32 4.20 1. Assume an appropriate limit for the relative error and draw conclusions.12 4.50 0.23 3.40 0 2 4 6 Quality Assurance and Quality Control 8 10 12 14 16 Excel file: exampl_valid03. 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.© 2009 by Taylor & Francis Group.1. assuming the reference value to be x.11 2.

83 8. The value of this parameter can serve to determine the influence of noise level on the .65 −0.83 3.08 −3. a ε.21 18 0.Method Validation 16 17 18 12 12 12 6.2.78 6.78 5.08 0.60 −1. % Solution: Number of results.00 143 Relative error — ε.22 4.43 1.01 2.72 −5. % 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Excel file: exampl_valid04.5642 −0.99 1. n Slope. b Intercept.xls 1. LLC .© 2009 by Taylor & Francis Group. 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.66 5.3 Limit of Detection and Limit of Quantitation The next validation parameters that need to be determined are the LOD and the LOQ.37 0.0291 Conclusion OK !!! OK !!! OK OK OK !!! OK !!! OK OK OK OK OK OK OK OK 7.97 6.59 −8.10 −0.92 −1.56 −1. The values of these parameters are closely related to the magnitude of noises in the measurement system.

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

2 Calculation of LOD Based on the Numerical Value of the S/N Ratio When calculating the LOD. is using a measurement for a series of blank samples.© 2009 by Taylor & Francis Group. This quantity is then multiplied by 3 and the obtained signal value is converted into a concentration. that is. where xm = mean value SD = standard deviation In practice. Otherwise. of course.3. LLC . such samples are prepared through spike in the blank samples with quantifiable amounts of the analyte. The method would only have some application when the analyte concentration is measurable for a blank sample. the so-called background level is above the LOD for the applied detector (that is to say. one can determine the LOD value by using the obtained chromatogram for a blank sample. In the case of chromatography. the analyte concentration in a blank sample is at least equal to the LOQ for the applied detector). but one that is also metrologically more correct. obtained when a blank sample is subjected to final determination. it would be most convenient to prepare . and then to directly apply the principle that LOD is three times the noise level for an applied analytical method. 10 independent determinations are performed for samples in which the analyte concentration is close to the expected LOD. 30] — namely. In this instance.2. 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 ± 0.3 Calculation of LOD Based on Determinations for Blank Samples A more labor-consuming method. however. This method can be applied only when it is possible to obtain the baseline of noises. LOD is equal to the mean value magnified by three times the standard deviation in this instance. The manner of conduct is then similar to the previously described one. Of course.2) The modification is the preparation of n samples with analyte concentrations on a level close to the expected LOD. To this end.5 min). it is possible to use the described method with a certain modification [29. one calculates the mean value and the standard deviation. It involves 10 independent measurements for 10 independently prepared blank samples [28]. 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. one uses the determined S/N ratio for the investigated analytical procedure [2]. 7. with the one difference being that LOD is calculated according to the formula: LOD = 0 + 3 ⋅ SD (7.Method Validation 145 7.3. For the thusly obtained 10 results. it is seldom possible to obtain a numerical value for the mean.2. it seems paradoxical to obtain a result for a value that by definition should be a submarginal quantitation.3) LOD = x m + 3 ⋅ SD (7.

then the determined LOD is also the MDL. after which one determines the LOD according to the following dependence: LOD = 3 ⋅ SDo (7. LLC . If determinations are instead performed directly on the prepared standard solution samples. .6) 7. A linear dependence is determined that associates the calculated standard deviations with the respective concentrations: SD = f (c) (7. In this case. 6. LOD = t ⋅ SD (7.2. then IDL is determined in this manner.3 ⋅ SD b (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).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. one determines the absolute term SDo. LOD is calculated using a dependence described by an equation for the number of degrees of freedom f = n − 1. 7].2. one should perform at least six parallel determinations. calculate the standard deviations.7) where b is the slope of calibration curve.5) Next. One then performs an analysis on such prepared samples. receiving a series of n results for which one calculates the mean value and standard deviation. it is indispensable to perform calibration that will influence the LOD [2.146 Quality Assurance and Quality Control standard solutions in which matrix compositions correspond to the matrix composition of real samples. In this case. LOD is calculated using the following dependence: LOD = 3.4) 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. where n is the number of independent samples and the accepted level of significance α.© 2009 by Taylor & Francis Group. For each level of analyte concentration.3.3. and then for each series of measurements obtained in this way. 7.

Hence.3.6 Calculation of LOD Based on a Given LOQ LOQ is the lowest analyte concentration that can be determined with a suitable precision and accuracy.1 Construction of the graph and calculation of the limit of quantitation [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. the coefficient of variation (CV) is calculated and the graph of the f(c) dependence is drawn. . 28].1 presents the construction of the graph and the calculation of the LOQ [28]. and for this value the concentration equal to the LOQ is read on the graph. 10% CV. One performs measurements for standard solutions (matrix standards) on at least five levels of concentrations [2. Figure 7. % LOQ 0 1 2 3 4 5 6 Analyte Content 7 8 9 FIGURE 7. For each level of concentrations. if measurements are conducted based on analyses of blank samples subjected to the whole analytical procedure.g.© 2009 by Taylor & Francis Group.2.. When the LOD is calculated based on parameters of the determined calibration graph (residual standard deviation or standard deviation of the intercept). The required precision for the LOQ is determined (usually = 10%). • As a standard deviation of the intercept of the obtained calibration curve — SDa. 7. the calculated value is the LOD for the measuring instrument. one performs six parallel measurements. e. Of course. • As a residual standard deviation of the calibration curve — SDxy. LLC . 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). described by the dependence (1. the LOD (for the analytical method or the applied detector) will be calculated depending on which parameters were used to calculate the standard deviation. 20 18 16 14 12 10 8 6 4 2 0 Required precision. For each solution. The LOD is calculated as: LOD = LOQ/3.68). the MDL is the determined quantity.

LOQ Table 7. 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. 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. 7.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.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.© 2009 by Taylor & Francis Group.3 compares all the described methods of calculating the LOD.148 Quality Assurance and Quality Control TABLE 7. Disadvantages.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. The .2.3 Methods for Determining Detection Limits: Requirements. LLC . together with their short characterizations [27].3.

while calculating the LOD of an analytical procedure.© 2009 by Taylor & Francis Group. • Analyte concentration should be on a level close to the expected LOD. one should fulfill the following conditions [29]: 10 ⋅ LOD > cmin LOD < cmin (7. with two basic features: • Matrix composition should be as close to the matrix composition of real samples as possible. . If condition (7. Inversely. One should also pay attention to the recovery of the analytical method in measurements conducted for standard solutions. 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. Recovery can be calculated using the dependence [29]: %R = xm ⋅ 100% c (7.8) (7. A recovery being too low results in an undervaluation of the calculated LOD. One should then calculate the limits of detection for newly prepared standard solutions with a lower analyte concentration.8) is not fulfilled. the analyte concentration in the prepared standard samples is too low. To check the calculated LOD. LLC . when the condition is not fulfilled (7. To test the trueness of the calculated LOD.9).9) where cmin is the analyte concentration in a standard solution sample with the lowest concentration. one should remeasure and recalculate using standard solutions in which the analyte occurs in higher concentration levels.Method Validation 149 solutions should be characterized.11) where %R denotes the recovery of an analyte for a given analytical procedure. It can happen that the concentrations in standard samples are considerably higher than the calculated LOD. In this case. one can also estimate the S/N ratio based on the following dependence [29]: S/N= xm SD (7. When it is higher.10) According to the definition of LOD. the numerical value of this ratio should be between 3 and 10. it will signify that the concentration in the prepared standard samples is too high. the determined LOD is greater than the numerical value and one should conduct remeasurement for smaller concentrations of the analyte in standard solution samples.

The described ways of determining the LOD and/or quantitation permit the determination of both the MDL and the measuring instrument detection limit.5 Problem: Using the given analyte concentration determinations for blank samples.155 0.150 Quality Assurance and Quality Control As previously stated. 32]. 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. estimate LOD and LOQ for the validated analytical method. Determining limits of detection and quantitation allows the unequivocal determination and presentation of results in the proximity of these values. the described methods of testing the correctness of the calculated LOD can only be applied when the measurements are performed using prepared standard solutions. TABLE 7.© 2009 by Taylor & Francis Group. it is recommended to apply a less time-consuming method. x x < LOD LOD ≤ x < LOQ x ≥ LOQ Recording of Result Not determined Not quantified Value of concentration Example 7. it is recommended to use a way the assumptions of which are based on chemical metrology. ng/g: Data 1 2 0. Data: results. Using the calculated S/N ratio. examine the correctness of the determined LOD.132 . The determined limits of detection and quantitation also show the quality of measurements conducted using a given analytical method [31.4 Correct Method for Recording a Determination Result Result. For validation of an analytical method. LLC . A correct method for recording a determination result depending on the quantity of an analytical signal is presented in Table 7. the value of the determined LOD is associated with statistical parameters such as: • Level of probability • Number of degrees of freedom For individual measurements.4. It must be stated that the determined LOD should always be given the description and parameters of the method applied in its calculation.

calculated LOD is correct.143 0.235 0.121 0.6 Problem: During the measurements performed on blank samples. Check the trueness of the LOD determination through the comparison with the standard solution concentration.145 0.244 0.54 ng/g 0. Data: results.© 2009 by Taylor & Francis Group.18 ng/g 0.Method Validation 3 4 5 6 7 Solution: xm SD LOD LOQ 0. and based on measurements for these solutions the estimation was made for LOD and LOQ.254 0.250 .135 ng/g 0.258 0. it was noticed that the obtained values of signals cannot be measured. Hence.113 0. ng/g: Data 1 2 3 4 5 6 c 0. the standard solutions were made with concentrations near the expected LOD.258 0.137 151 LOD = xm + 3 ⋅ SD S/N= xm SD S/N 9 Conclusion: S/N ratio is in the range 3–10.253 0. LLC . Excel file: exampl_valid05.xls Example 7.014 ng/g 0.

8 7. estimate the LOD and LOQ of the validated analytical method. calculated LOD is correct.5 6. Data: results.3 mg/dm3 8.6 0.8 7.© 2009 by Taylor & Francis Group.2 9.152 Solution: SD LOD LOQ Quality Assurance and Quality Control 0. mg/dm3: Data 1 2 3 4 5 6 7 α Solution: xm SD t LOD LOQ 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.xls Example 7.8 mg/dm3 8. check correctness of the determined LOD.027 ng/g 0.447 mg/dm3 2.13 mg/dm3 2.7 Problem: Using the data given in determinations of the analyte concentrations for blind samples.0091 ng/g 0. using Student’s t test.05 . LLC . Excel file: exampl_valid06.4 9.6 9. Using the data calculated S/N ratio.41 mg/dm3 1.

11 0.15 0.15 Signals 198 177 132 156 205 193 135 298 237 222 257 243 313 235 0.© 2009 by Taylor & Francis Group. check correctness of the LOD determination through a comparison with the standard solution with the lowest concentration.2 0.23 .1 30. calculated LOD is correct. ppm 0. In addition.11 1 2 3 4 5 6 7 Solution: Concentration 0.8 Problem: Using the given analyte concentration determinations for standard solution samples. Present LOD in units of the analyte concentration in standard solutions applied for LOD estimation.9 257.013 0. Data: results: Concentration. Draw an appropriate graph.4 101 144 124 174 102 111 121 0.xls Example 7. Excel file: exampl_valid07.3 170. LLC . estimate the LOD and LOQ using a graphical method.Method Validation LOD = t ⋅ SD S/N= S/N xm SD 7 153 Conclusion: S/N ratio is in the range 3–10.23 SDo LOD LOQ Signal 125.040 signal ppm ppm SD (Signal) 26.1 34.9 19.

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. LLC .5x + 19.0 30.15 Signal 101 144 124 174 102 111 121 198 177 .11 0.11 0. ppm 1 2 3 4 5 6 7 8 9 0.1 0.0 0. comparing the calculated value with the value of the analyte concentration in the standard solution with the lowest concentration.11 0. Present the value of LOD in units of standard solution concentration.15 0.8.0 35. calculated LOD is correct.0 20.11 0.11 0.11 0.05 Quality Assurance and Quality Control y = 67. Excel file: exampl_valid08.© 2009 by Taylor & Francis Group.0 15. Data: results: Concentration.154 Graph: 40.0 10.2 0.0 SD (signal) 25. estimate the LOD and LOQ via the method using parameters of the calibration curve.0 0 0.2 0. applied in LOD estimation.11 0.9 Problem: Using the data from Example 7.xls Example 7. Also check the correctness of LOD determination.15 Concentration 0.0 5.

23 0.05 y = 1102x + 4. ppm 0.© 2009 by Taylor & Francis Group.6 0.23 0.15 Concentration.23 0.089 ppm 0.Method Validation 10 11 12 13 14 15 16 17 18 19 20 21 Solution: Number of results.y) LOD (SDa) LOD (mean) LOD = 0.23 0.2 0. r LOD (SDx.077 ppm 3.1 0.066 ppm 0. a Residual standard deviation.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. LLC .15 0. b Intercept. n Slope.15 0.25 .3 ⋅ SD b 21 1102 4.8901 0.1 0. SDa Regression coefficient.15 0. SDb Standard deviation.23 0.23 0. SDxy Standard deviation.63 29.15 0.6 129 22.15 0.

calculated LOD is correct.5 2. Data: results: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Concentration. by a method using the parameters of the calibration curve.xls Example 7.2 1.3 3.2 1. Excel file: exampl_valid09.5 2.3 3.5 2.© 2009 by Taylor & Francis Group.2 1.3 3.5 2. Also check the correctness of LOD determination comparing the calculated value with the analyte concentration in a standard solution with the lowest concentration.3 3. ppm 1.2 2.2 1.10 Problem: Using the analyte concentration determinations for standard solution samples. LLC . estimate the LOD and LOQ. Present the values of LOD in units of standard solution concentrations applied for LOD determination.5 2.156 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.2 1.3 3.5 3.3 Signal 1460 1725 1150 1025 1825 1310 1950 1630 2200 1650 2000 1980 2900 3200 3245 2850 3500 3890 .

SDxy Standard deviation. SDa Regression coefficient. LLC .3 ⋅ SD b 18 831 254 447 122 303 0.5 1 y = 831x + 254 1. SDb Standard deviation.2 ppm 1. .© 2009 by Taylor & Francis Group.5 10 ⋅ LOD > cmin LOD < cmin Conclusion: Because the concentration of a solution with the lowest concentration is lower than the calculated LOD.8627 157 Graph: 4500 4000 3500 3000 Signal 2500 2000 1500 1000 500 0 0 0.5 2 2.5 ppm 3.8 ppm 1. n Slope.5 Concentration. b Intercept. 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). r LOD (SDxy) LOD (SDa) LOD (mean) LOD = 1. ppm 3 3. a Residual standard deviation.Method Validation Solution: Number of results.

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. SDxy Standard deviation. b Intercept.4 ppm 1.7 4.5 3. SDa Regression coefficient.3 ppm 1.5 2.3 3.5 2.7 4.9132 . a Residual standard deviation.7 4. n Slope.5 2.4 ppm 2.3 3.5 2.3 3.158 Data (2): results: Quality Assurance and Quality Control Concentration. LLC .© 2009 by Taylor & Francis Group.5 2. r LOD (SDxy) LOD (SDa) LOD (mean) 1. ppm 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Solution (2): Number of results.7 4.3 4.3 3.3 3. SDb Standard deviation.7 4.

LLC .0 198 177 232 200 205 193 235 20. Excel file: exampl_valid10. estimate the LOQ and then the LOD using an LOQ determination method based on the assumed value of determination precision.© 2009 by Taylor & Francis Group. Assume the maximum value of the coefficient of variation to be CV = 5%. Data: results: Concentration.0 444 450 470 400 445 450 470 30.5 Concentration.0 1000 990 995 1010 1005 1015 995 Signals . Draw an appropriate graph.5 1 1. calculated LOD is correct.0 1 2 3 4 5 6 7 104 144 124 124 102 111 121 10.xls Example 7. Present LOD in units of the standard solution concentration applied for LOD determinations.0 635 650 660 620 610 625 615 40.5 3 3.11 Problem: Using the values of the analyte concentration determinations for standard solution samples. ppm 5.5 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.0 800 810 805 825 820 840 830 50.5 2 2.Method Validation Graph (2): 5000 4500 4000 3500 Signal 3000 2500 2000 1500 1000 500 0 0 0. ppm 4 4.

2 5.4 Range Determination of linearity and the LOQ enables the determination of a measuring range for an analytical method. Calculate the regression parameters of the calibration curve.0% 2. .2 2.2.0 10.160 Solution: Quality Assurance and Quality Control Concentration.3 ppm 20 22 30 40 Concentration 50 60 Excel file: exampl_valid11.9 1.12 Problem: Determine the calibration curve based on analyte concentration determinations in eight standard solutions samples (five independent measurements for each solution). LLC .0% 5% 4.8 0. A measuring range is a range of values (analyte concentrations) in which the error of a measuring instrument is below the assumed value. % 12.0% 12. Example 7.0 30. ppm 5.9 22 ppm 7.0% 0 10 CV.0% 10.0% CV (signal) 8.0% 6.0 50.xls 7. In practice.0 20. Prepare an appropriate graph. 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.2 10.0% 0.0 LOQ LOD Graph: 14.0 40.© 2009 by Taylor & Francis Group.

65 0.65 0.12 1.44 2.4 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 .75 3.4 10.75 3.Method Validation 161 Using the determinations for standard solution samples for three lowest concentration levels.8 7.8 10.12 1. LLC .65 0.44 3. Present the LOD in units of standard solution concentration applied in LOD estimation. estimate the LOD and LOQ using a technique based on using parameters of the calibration curve. Present the measuring range of the analytical method. Data: results: Concentration.44 2.75 5.12 2.65 0.12 1.8 7.25 7.75 3.25 5.44 2.25 5. Also check the correctness of LOD determination.65 1.25 5.12 1.44 2. comparing the calculated value with the analyte concentration in the standard solution with the lowest concentration. 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 0.8 7.25 5.© 2009 by Taylor & Francis Group.8 7.75 3.

089 ppb 0. SDxy Standard deviation.115 ppb 0.27 ppb 0. SDxy Standard deviation.3 ⋅ SD b LOD = . r Solution (LOD): Number of results.4 10. n Slope.3 ppb 3.9991 0.162 33 34 35 36 37 38 39 40 Solution (calibration): 10.27–13.© 2009 by Taylor & Francis Group. n Slope. b Intercept. b Intercept.4 13.1 0.3 200 7 52.3 13. SDb Standard deviation.2 44.3 13.4 10. SDa Regression coefficient.3 13.8 15 24. SDb Standard deviation. a Residual standard deviation. r LOD (SDxy) LOD (SDa) LOD (mean) LOQ Range 40 1256 59. LLC .9993 15 1280 −52.3 Quality Assurance and Quality Control 13300 13000 12950 16600 16745 16600 16200 16500 Number of results.3 0.063 ppb 0. a Residual standard deviation. SDa Regression coefficient.3 13.

**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

α = 0. there is a statistically significant difference in variance value.© 2009 by Taylor & Francis Group. Excel file: exampl_valid14. check (at the significance level of α = 0. Data: result series: Series 1 1 2 3 4 5 6 7 Solution: Series 1 Number of results.95 Series 2 6 0. the series differ in precision.983 1. Apply Snedecor’s F test. Equinumerous series — apply Hartley’s Fmax test.160 10 12 13 14 18 15 17 Series 2 11 11 13 11 13 12 Conclusion: Because F > Fcrit . LLC . . check (at the significance level of α = 0.05) if the values of the standard deviation for the given series of results are statistically significantly different.15 Problem: For the given series of measurement results.14 Problem: For the given series of measurement results.05) 7 2.111 7.xls Example 7.85 4.795 9.Method Validation 167 Conclusion: Because χ2 > χ2 crit . SD n/(n − 1) · SD2 F Fcrit (f1 = 6.xls Example 7.05) if the standard deviation values for both series are statistically significantly different. f2 = 5. Excel file: exampl_valid13. n Standard deviation. there is a statistically significant difference in variance values for the compared series.

76 4 15 2.356 2 15 1. SD Fmax Fmaxo (k = 5.05) if the values of the standard deviation for a given series of results are statistically significantly different. f = 14. Excel file: exampl_valid15. n Standard deviation. Not equinumerous — apply the Bartlett test.506 3 15 2. α = 0.05) 15 1. check (at significance level of α = 0. .384 5 15 1.09 4. there is no statistically significant difference in variance values for the compared series.699 Conclusion: Because Fmax < Fmaxo.© 2009 by Taylor & Francis Group.017 3.16 Problem: For the given series of measurement results.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. LLC .xls Example 7.

22 11.733 26.077 0.769 26.125 0.250 1. n Standard deviation − SD 1/(n − 1) (n − 1) · log(SD2) (n − 1) · SD2 c SDo2 Q χ2 crit (f = k − 1 = 5.270 8.672 4. LLC .529 1.04 3.07 Conclusion: Because Q > χcrit 2.095 7.000 50.091 3.071 0.701 4.xls .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.356 1.245 9.075 2. there is a statistically significant difference in variance values for the compared series.803 2.137 0.000 76.257 25.100 0.083 0.727 56.845 51. α = 0.© 2009 by Taylor & Francis Group.05) 2 3 4 5 6 15 11 14 13 9 12 1.635 2. Excel file: exampl_valid16.

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 0.18 Problem: To determine the values of repeatability.xls Example 7.8090 0. Using the obtained measurement results.7 8.8 9.9 9.201 8.3 9.8 9.1 8. six independent series of measurements were performed for six standard solution samples.7 8.170 Quality Assurance and Quality Control Example 7. In each series. calculate value repeatability for the analytical method.440 0.1 8.9 9.2 9.816 1. Excel file: exampl_valid17.6 10.9 8.1 9.4 9.3 8.1 9.2 9.17 Problem: For the given series of measurement results–dependent variables.9 9. Apply the Morgan test.05) if the values of the standard deviation for the given series of results are statistically significantly different. .6 8.1 Series 2 9.1 10. LLC .© 2009 by Taylor & Francis Group.6 Conclusion: Because t < tcrit. there is no statistically significant difference in variance values for the compared series.2 9. check (at the significance level of α = 0.580 0.576 2.2 9.7 9.1 9. five repetitions were made.

34 7.34 Series 4 10.8 17.19 Problem: To determine the values of repeatability and the intermediate precision of the analytical method.3 17.119 1.34 3 5 0. LLC .142 5.3 10.76 11.Method Validation Data: result series: Series 1 1 2 3 4 5 2. Series 1 Number of results. If variances are homogeneous.05) 5 0. one should reject the deviating value (series) and perform the calculations again.05 5 5 0. repeatability should be calculated as a mean value CV for the given series.65 4 5 0.14 7.11% Example 7. .2 10.65 2. It is possible to calculate repeatability as a mean value CV for the given series.0 17.12 5. SD Coefficient of variation.67 2.54 2. six repetitions were performed. α = 0.28 6 5 0.60 2 5 0. n Standard deviation.09 7.121 2.9 14.34 5.5 171 Solution: Because the levels of analyte concentrations in the investigated standard solutions samples are different.16 5.© 2009 by Taylor & Francis Group.8 13.50 Conclusion: Because Fmax < Fmaxo.15 7. six independent series of measurements for the samples were performed for one standard solution. there is no statistically significant difference in variance values for the compared series. the calculations should use the values of CV and not SD.52 29.9 11.1 Series 5 14.305 1.43 2. f = 4. Because series are equinumerous.4 Series 6 17. In each series.24 5. If variances are not homogeneous.2 17. CV repeatability Excel file: exampl_valid18.2 10. CV (%) Fmax Fmaxo (k = 6.327 2.2 14. one should apply the Hartley Fmax test.34 Series 2 5. calculate the values of repeatability and intermediate precision for the analytical method.xls 3. Using the obtained measurement results.3 14.532 5. The first step is to check the homogeneity of variances for individual series of results.02 Series 3 7.

repeatability should be calculated as a mean value CV for the given series.483 3 4 5 6 1. If variances are homogeneous.05) 6 1.172 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.507 7.70 Conclusion: Because Fmax > Fmaxo.581 26. one should apply the Hartley Fmax test.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 .© 2009 by Taylor & Francis Group. there is a statistically significant difference in variance values for the compared series. f = 5. α = 0.941 6 6 2.422 6 6 3. one should reject the deviating value (series) and perform the calculations again. If variances are not homogeneous. Series 1 Number of results.52 18. Because series are equinumerous. n Standard deviation. Results from series 4 should be rejected due to lack of homogeneity of variances and calculations should be performed again.472 2 6 2. SD Fmax Fmaxo (k = 6. LLC . Excel file: exampl_valid19a.

507 – 5. 37–39]: • Sample analysis of suitable certified reference materials • Comparison of the obtained result with a result obtained using a reference (original.483 3 4 5 6 1. It is influenced mostly by the bias of the analytical method. Repeatability was calculated as a mean of SD values for individual series.37 2.422 6 0 3.75 7. final) method [40–42] • Standard addition method .© 2009 by Taylor & Francis Group.05) 6 1. Trueness is the closeness of agreement between the average of an infinite number of replicate measured quantity values and a reference quantity value [26].30 173 Conclusion: Because Fmax < Fmaxo. the more accurate the result of a single measurement is. 9]. α = 0.472 2 6 2. SD repeatability SD intermediate precision Excel file: exampl_valid19b. n Standard deviation. Analysis of these definitions shows that the hitherto existing notion of “accuracy” was replaced by the term “trueness. SD Fmax Fmaxo (k = 5.” It is trueness that describes the conformity of results obtained using a given analytical method to real (expected) results.68 16. other parameters such as linearity and sensitivity also influence the accuracy of an analytical method. calculated using all the 30 results. and accuracy are presented schematically in Figure 7.Method Validation Series 1 Number of results.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].2. Of course. f = 5. LLC . The truer and more precise the results obtained using a given method. precision.” and the previously applied notion of “accuracy of a single measurement” is now simply “accuracy.xls 2. Intermediate precision is SD.3 [4. Accuracy is a combination of trueness and precision. there is no statistically significant difference in variance values for the compared series. Relationships between trueness.941 6 6 2. Trueness and accuracy can be determined using different approaches [33.

174 Quality Assurance and Quality Control INCREASING TRUENESS µ µ INCREASING ACCURACY µ µ INCREASING PRECISION FIGURE 7.3 Relationships between trueness. The value of a single measurement result may differ (and actually always differs) from the expected (real) value. LLC . 7. Depending on the type of errors. their influence on measurements varies.12) ε xi = d xi µx (7. 9]. The difference is due to the occurrence of different errors [44].2. one can distinguish: • Absolute error dx. and accuracy [4. 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 7.13) .1 Measurement Errors The notion of accuracy is closely connected with the notion of errors [43].7. described by the equation: (7.© 2009 by Taylor & Francis Group. which can be described by the dependence: d xi = x i − µ x • Relative error εx. precision. With regard to the manner of presenting a determination result.4 [9].

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). . LLC . and therefore to eliminate. With regard to the source of errors. as described by the following equation [45]: d xi = xi − µ x = ∆x sys + ∆xi + δxi (7. • It is a random variable — however. 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]. • It is the easiest to detect.© 2009 by Taylor & Francis Group. • It appears only in some measurements. one with unknown distribution and an unknown expected value.Method Validation ∆xsys δxj 175 x2 x1 x3 x4 x5 x6 µx xm xj ∆x1 ∆xm FIGURE 7. there is a high probability of detecting a result(s) with a gross error. Gross error is characterized by the following properties: • It is the result of a single influence of a cause acting temporarily.

μ2x are the expected values for two standard samples. Table 7. for example. often described as “outliers” [9].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. often described as “outliers” [9]. TABLE 7.5 schematically presents the selection criteria for a suitable manner of action in detecting and rejecting results with gross errors. x2m are the mean values determined for standard samples. Methods of gross error determination are described in Chapter 1.15) where μ1x. 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 presents specific methods of bias determination [9]. using the developed method. LLC . . the trueness of the obtained final determination (most often the mean value of the measurement series) is influenced by biases and/or random errors.176 Quality Assurance and Quality Control • It assumes both positive and negative values (unlike bias).5 Selection criteria for a suitable manner of action in detecting and rejecting results with gross errors.© 2009 by Taylor & Francis Group. a mistake in instrument reading or a mistake in calculations. After eliminating results with gross errors. The determination of biases is one way to determine the trueness of an analytical method. • The cause of its occurrence can be. Figure 7. Constant bias asys is determined according to the formula: asys = µ1x x1m − µ 2 x x 2 m µ1 x − µ 2 x (7. and x1m. Each of them is applied in specific conditions. There are many known ways of detecting results with gross errors.

The value of variable bias is determined according to Equation (7. The relationship between results obtained by the reference method (0Y axis) and results obtained by the developed method (0X axis) is determined. The correction multiplier value is determined according to the formula: B= x 2 m (ref ) − x1m (ref ) x 2 m − x1m (7.63) and (1. and x1m.19) (continued) . x2m(ref) are the mean values determined for the first and second standard when using the reference method. Regression parameters of the regression line Y = b · X + a are determined according to Equations (1.17) 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.16) Variable where Ct is the expected value increase of analyte concentration due to standard addition. x mCst are the mean values determined for sample and sample with standard addition.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). The value of variable bias is determined according to the equation: bsys = 1− B B (7. and xm. x2m are the mean values determined for the first and second standard with using the developed method. The correction multiplier value is determined according to the formula: B= Cst x mCst − x m (7.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.© 2009 by Taylor & Francis Group. LLC . The values of constant bias and variable bias are determined according to the formulas: (a) constant bias a = − asys b (7.17).

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.© 2009 by Taylor & Francis Group. LLC .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. the random error is negligibly small with relation to the bias when n → ∞.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. than s → 0. In this case. then with a large number of conducted measurements.22) b = B (7.21) asys = − a b (7.24) 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 . the following dependence is true [45]: d xmet = E ( x met ) − µ x = ∆x sys (7.

According to the general definition. can a result have a random error. whose value is not relative to analyte concentration levels — asys • A variable bias.© 2009 by Taylor & Francis Group. reference material is characterized by a constant and strictly defined analyte concentration and with a known concentration determination uncertainty [26].10) or Aspin and Welch test (Section 1. the bias of an analytical method is determined. one may use for “poor” (small) result series the “approximate test” of Cochran’s C and Cox test (Section 1. Its value influences the precision of the obtained results. whose value depends (most often linearly) on analyte concentration levels — bsysµx Bias is described by the dependence: ∆x sys = asys + bsys µ x (7. However. In this case. 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. one should perform independent determinations for a blind sample and correct the .9) for the significance of differences between two results. one can compare both results visually. 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.5). Trueness or accuracy can be determined using different techniques [37–39].Method Validation 179 In this manner. Section 1.8.11). Of course.8. however. Another manner (most often applied) to determine the trueness or accuracy is the analysis of a reference material sample (or better still. 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).8. Of course. 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. One may differentiate between two types of bias: • A constant bias.8.26) Only after rejecting results with a gross error and determining biases (regarding their values and correcting the determination result). one should prepare a standard solution by adding a strictly specific quantity of analyte into the investigated sample and subject it to determination. In each case.25) Assuming that the value of a random error is negligibly small compared to the bias value. In case of its inaccessibility. but it is more metrologically correct to use Student’s t test (Section 1. it is not always possible to use reference material samples precisely satisfying given needs. LLC . samples of the certified reference material) using the investigated analytical method.

29) 2 u(2xi ) + u(2xref ) (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. LLC . the ratio should be 1) and values of uncertainty for such a determined quantity.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. one should apply Student’s t test (Section 1. The inference is as follows: if the interval of a determined ratio ± the uncertainty of its determination (R ± U ) includes 1.© 2009 by Taylor & Francis Group.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 . To test if the obtained measurement value does not differ in a statistically significant manner from the certified value (expected value).180 Quality Assurance and Quality Control result for the sample with the known analyte concentration by the obtained measurement result. One determines the ratio of the obtained means (if the values did not differ between themselves. one should infer that the compared mean values do not differ in a statistically significant manner.8.27) and then the uncertainty U.9). one should calculate the value of the R relation according to the formula: R= x1m x2m (7. Using obtained values. using a dependence described by the equation: U=k x1m + x 2 m 2 ( SD 2 1 2 + SD2 ) (7.

It is also recommended to perform a series of measurements for the reference material using a so-called primary method. 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. Data: result series. the corrected mean obtained for the investigated method is compared with the one obtained by the primary method. The mean obtained for blank samples is then deduced from the mean obtained for the reference material. Apply the confidence interval method. after the initial outlier rejection.05. characterized by a null value of bias. check if there is a result with a gross error. Assume the value α = 0. 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. and so the corrected value is compared against the certified value. however.Method Validation 181 Inference.2 9.3 . mg/dm3: Data 1 2 3 4 5 8.8 9. EURACHEM [28] recommends 10 parallel determinations for a blank sample and the same number of determinations for reference material samples. 7].© 2009 by Taylor & Francis Group.5 6. 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). 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. LLC . 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.8 7. in this instance. Example 7. In this case. According to recommendations by ICH [6.20 Problem: In the given series of measurement results. • When.

8 9.2 2.10 ÷ 10. Assume the value α = 0. mg/dm3: Data 1 2 3 4 5 6 8. apply the confidence interval method.8 9.59 mg/dm3 n SD n− 2 g = xm ± tcrit g 8.20.447 8.182 6 7 8 α Solution: xmin xmin + 1 xmax xmax − 1 tcrit Initially.67 mg/dm3 (7.1 8.xls Example 7.8 0. Excel file: exampl_valid20.8 7. without the initial outlier rejection.3 7.05 6.2 .5 6.77 ± 1.44) mg/dm3 Conclusion: The value xmin lies outside the determined confidence interval — hence it has a gross error. Data: result series.21 Problem: Using the data given in Example 7.3 8.5 9. xm SD Quality Assurance and Quality Control 8.© 2009 by Taylor & Francis Group. LLC .77 mg/dm3 0. the result xmin was rejected.05.2 9.2 9.

xls 8.© 2009 by Taylor & Francis Group.5 1.1 8.05. mg/dm3: Data 1 2 3 4 5 6 7 8 α 8.3 8. apply the Dixon Q test. Assume the value α = 0.8 7.05 .59 mg/dm3 Example 7.8 9.93 mg/dm3 (6.5 6. LLC .2 9.03 mg/dm3 g = xm ± w α ⋅ SD g 8.87 9.2 9.22 Problem: Using the data given in Example 7.3 9.46 mg/dm3 1.8 0.1 8.Method Validation 7 8 α Solution: xmin xmax wα xm SD 6.39) mg/dm3 Conclusion: The value xmin lies outside the determined confidence interval — hence it has a gross error. Data: result series.77 mg/dm3 0. It should be rejected and the values of xm and SD should be calculated for the new series of data.8 0.20.05 183 8. xm SD Excel file: exampl_valid21.53 ÷ 10.46 ± 1.

23 Problem: In a given series of measurement results.094 0.2 13.7 14.05.59 mg/dm3 Example 7.1 13.8 .4 13. LLC .© 2009 by Taylor & Francis Group.2 13. xm SD Excel file: exampl_valid22.77 mg/dm3 0. the value xmin has a gross error. Data: result series.xls 8.4 13. Assume the value α = 0. check if there are any results with a gross error.469 0. R Q1 Qn Qcrit R = xn − x1 Q1 = Qn = x2 − x1 R xn − xn−1 R 8 3.9 14.468 Conclusion: Because Q1 > Qcrit.2 13. It should be rejected and the values of xm and SD should be calculated for the new series of data.20 0.4 13.7 13.184 Solution: Quality Assurance and Quality Control Number of results Range.2 11.3 13. Apply the confidence interval method. ppm: Data 1 2 3 4 5 6 7 8 9 10 11 12 13 13.

0 13.2 13.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 13.2 14.7 13.65 14.92) ppm Data 1 2 3 4 5 6 7 13.2 13.9 14.2 13.2 15.8 0.© 2009 by Taylor & Francis Group.1 14.9 13.4 13.2 14.7 13.2 13.73 ppm 0.4 13.6 13.72 ppm 1.73 ± 1.8 13.7 13.05 185 g = xm ± kα ⋅ SD g 13.1 13.4 13.4 .7 14.2 13.53 ÷ 14. LLC .19 ppm (12.8 15.3 13.1 14.

After their rejection the values of xm and SD were calculated again.7 13. Results of measurements were obtained using a method for which the standard deviation method had been determined.7 14.1 14.4 13.2 13.68 ppm 0.9 13. xm SD Excel file: exampl_valid23.2 Outlier Outlier 13.2 14.3 13. and 17 lie outside the determined confidence interval — hence they have a gross error.xls 13. .© 2009 by Taylor & Francis Group.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 13.4 13.7 13.2 13.2 Outlier 13.2 14.8 13.8 Conclusion: Results 10. Apply the critical range method.2 13.2 13.6 13.1 14.05.24 Problem: Check if there are results with a gross error in a given series of measurement results.0 13. LLC .36 ppm Example 7. Assume the value α = 0.8 14.7 14. 16.

8 ppb 113 125 120 127 115 118 117 134 124 0.Method Validation Data: result series. Data (2): result series. 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 4.39 21. new calculations for the new series should be done.05 4. a result xmax is considered to be an outlier.0 ppb 19. ppb: Data 1 2 3 4 5 6 7 8 9 113 125 120 127 115 118 117 – 124 .5 187 Rcrit = z ⋅ SDg Conclusion: Because R > Rcrit. LLC .© 2009 by Taylor & Francis Group.

3 ppb Conclusion: Because R < Rcrit.2 55.3 51. there are no more outliers in the series.9 52.05. Results were obtained using a method for which a standard deviation had been determined before.29 14.1 51.188 Solution (2): xmin xmin + 1 xmax xmax – 1 z R Rcrit Quality Assurance and Quality Control 113 115 127 125 4. Assume the value α = 0.8 56. the values of xm and SD could be calculated. Data: result series.7 53.2 54.8 53.25 Problem: Check if there is a result with a gross error in a given series of measurement results.1 56.65 ppb Example 7.1 57. ng/g: Data 1 2 3 4 5 6 7 8 9 10 11 12 13 14 55.5 57.xls 121.© 2009 by Taylor & Francis Group. Apply the confidence interval method.0 56.7 . LLC .0 ppb 19.00 ppb 6.1 54. xm SD Excel file: exampl_valid24.

9 ± 3.9 189 Result xmin was initially rejected. check them for the occurrence of outliers. xm SD Excel file: exampl_valid25.xls 54. LLC . performing three parallel determinations per each sample.9 57.1) ng/g Conclusion: An initially rejected result xmin lies in the determined confidence interval.7 51. .7 ng/g 1.05.05 1. xm 54. and the values of xm i SD were calculated again.0 1.Method Validation 15 16 17 α SDg Solution: xmin xmin + 1 xmax xmax – 1 ka 51. Using the data obtained measurement results.© 2009 by Taylor & Francis Group.8 ng/g Example 7.26 Problem: Determinations were made for 25 samples.2 54.6 ÷ 58.2 ng/g (51. Apply the critical range method.65 54.1 57. It has been included in the series.5 0. The confidence interval value was calculated for the new series. Assume the value α = 0.3 55.9 ng/g n n −1 g = xm ± kα ⋅ SDg g 54.

82 3.45 3.08 0.14 3.12 3.45 3.07 3.99 3.52 3.33 3.01 3.33 3.23 3.65 3.11 3.41 3.23 3.07 3.67 3.11 3.45 3.72 3.44 3.49 1. 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 3.51 3.22 3.49 3.33 3.09 0.12 3.23 3.20 3.33 0.98 3.01 3.32 3.49 3.41 3.13 3.41 0.28 3.62 3.33 3.34 0.71 3.23 3.05 0.12 3.56 3.11 3.74 3.22 3.62 Conclusion OK OK OK OK OK OK OK OK .190 Data: result series.© 2009 by Taylor & Francis Group.96 Result 3 3.58 Quality Assurance and Quality Control Result 2 3.13 3.41 3.24 0.34 0.04 3.62 3.04 3.01 3.04 3.60 3. LLC .41 3.08 3.48 3.82 3.65 3.98 3.01 3.65 3.21 3.62 α Rm zα Solution: Sample 1 2 3 4 5 6 7 8 Ri 0.35 3.67 3.37 3.11 3.56 3.45 3.88 3.20 0.

59 0.41 5.19 0.27 Problem: Analyte concentrations were determined in two standard solution samples. Ri > Rcrit results should be rejected as an outliers. Using the obtained result series.98 0.04 5.61 0.74 0.33 5. A second standard solution was obtained by double dilution of the first standard solution.12 .11 10.28 10.07 10.60 0.23 Series 2 5.66 0. determine the value of the constant bias asys.15 0.01 10. with seven parallel determinations performed per sample. ppm: Results Series 1 1 2 3 4 5 6 7 10.96 191 Ri = xni − x1i Rcrit = zα ⋅ Rm Conclusion: For series 9 and 21.xls Example 7.45 5.47 0. Data: result series.28 0.Method Validation 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 0.13 5.11 5.49 0.© 2009 by Taylor & Francis Group.87 0.54 Rcrit Outlier OK OK OK OK OK OK OK OK OK OK OK Outlier OK OK OK 0. Excel file: exampl_valid26.97 0.22 10. LLC .23 10.29 0.67 0.

9 Cst Solution: xm xmCst 33.xls Quality Assurance and Quality Control 10. six parallel measurements were made.192 x1st x2st Solution: x1m x 2m k k= asys = asys Excel file: exampl_valid27. correct the values obtained for the real sample.4 33. Data: result series.5 58.2 56.2 57.8 58.8 34.16 5.© 2009 by Taylor & Francis Group. LLC .23 2 x1st x2st kx1m − x2m k −1 0.85 25. Using the calculated value of the correction multiplier.72 57.5 .2 33.9 33. determine the value of the variable bias bsys. For each of the samples.290 ppm Example 7. ppm: Results Series 1 1 2 3 4 5 6 33.1 33. Using the data obtained result series.9 58.0 5.0 10.0 Series 2 57.28 Problem: Analyte concentrations were determined in a real sample and in a real sample with the addition of the standard.

Using the obtained result series.© 2009 by Taylor & Francis Group.xls 1. using both methods.62 767.035 34. eight parallel measurements were made. correct the values obtained using the validated method.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 .Method Validation Cst xmCst − xm 1− B B 193 B= bsys = xm( corr ) = B ⋅ x B bsys xm(corr) Excel file: exampl_valid28. using an investigated method and the reference method.63 978. LLC .29 Problem: Analyte concentrations were determined in two real samples. Using the calculated value of the correction multiplier. Data: result series.93 Example 7.036 −0. 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. determine the value of the variable bias bsys.88 954.

8 56. ppb: Validated Method x 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 46.3 109 59. Apply the linear regression method.975 0.2 39.3 90.83 Excel file: exampl_valid29.3 38. determine the variable bias bsys and the constant bias asys.2 37.9 103 56.2 12.30 Problem: Analyte concentrations were determined in 15 real samples using the validated method and the reference method. and the mean values were presented. For each of the samples three parallel measurements were made using each of the methods.08 953.3 106 .1 27. LLC .2 35.026 748. Using the obtained data.2 44.9 88.3 57. Data: result series.2 111 Reference Method xref 45.© 2009 by Taylor & Francis Group.2 19.2 26.xls Example 7.8 77.5 101 79.3 47.7 86.9 97.194 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) 0.4 21.6 10.8 89.

However.972x – 0.2. The greater the influence of slight changes in parameters of the measurement process on final determination results. 47].972 0. the greater the attention one should pay to maintaining these parameters at a stable level.© 2009 by Taylor & Francis Group.589 0. LLC . Robustness influences the manner of conducting measurements using a given analytical method. and can be estimated based on reproducibility [46.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.xls 7.0292 Excel file: exampl_valid30.607 0. It is a parameter concerning changes in internal conditions [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.589 195 a b 1 −1 b –0. . 47]. ruggedness (flexibility) is a parameter describing the usefulness of a given analytical method in different conditions.

Conclusion: Such a measurement method guarantees high selectivity for indicating mercury for two reasons: 1. the influence of fluctuations in temperature. Problem 1: Determine the selectivity of the CV-AAS method. conditions of chromatographic isolations) [46. 7. together with a description of its determination. changes in purity and types of reagents. 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.2.9 Uncertainty Uncertainty is not considered a basic validation parameter. is presented in Chapter 4. although the influence of fluctuations from some measurement conditions (in a method subjected to validation) may be conducted in one laboratory (e. 47]. one can determine the usefulness of a given analytical method for a given determination. with a wavelength of 253. and slightly fluctuating the parameters of the applied analytical method. determine the linearity of the method. Problem 2: Based on measurement results for the series of standard solutions.© 2009 by Taylor & Francis Group. sent by hollow mercury cathode lamp. but it should be presented in the final method validation report. in which an absorption measurement is conducted.g. The exact characterization of this parameter. and then (after an eventual reduction to atomic mercury). The validation process method was conducted. determining the appropriate validation parameters.31 General problem: An analytical procedure was developed.. LLC . Example 7. mercury is released from the analyzed sample. pH fluctuations.7 nm. 2. Based on the estimated uncertainty value. The amalgamation reaction is a selectivity reaction for mercury.196 Quality Assurance and Quality Control Similarly to the reproducibility of an analytical method. 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. The absorption measurement is realized using a characteristic wavelength for mercury. Solution: In the case of the cold vapor technique. it is trapped on the gold bed as an amalgam. the amalgam is heated to 600°C and the released atomic mercury is directed through the air stream to the absorption cell. . After this step.

b Intercept.© 2009 by Taylor & Francis Group.90 2.1 66.05. Excel file: exampl_valid31_1. SDb Standard deviation.8 140.1 Solution: Before constructing the calibration curve. a calibration curve was constructed and their regression parameters were determined. with a significance level of α = 0.1 136.730 −2.20 80 5 129.79% 40 5 62.620 2.9 40 67.40 25.2 170.Method Validation Data: results: Unit Content of Hg Signal 20 33.5 98.880 2. SDa Regression coefficient.3 0.5 34.61% 60 5 92.7 0.1 35.79% Conclusion: There are no statistically significant differences in variation values.58 0. LLC . SDxy Standard deviation. SD CV.4 63.8 33. Hartley’s Fmax test was applied. mean Standard deviation.1 100. n Slope.2 95. r 25 1.838 1.2 138 100 167.2 32.290 1.93% 10.6 Hg.9986 .6 80 142. the homogeneity of variation for the results of the series being analyzed should be checked.03 1. % Fmax Fmaxo 20 5 31. Content of Hg No results.781 1. a Residual standard deviation.12 1.8 68.6 175.3 99. n Signal.3 171.77% 100 5 158.xls Due to no statistically significant differences in variation for the compared series.019 1.11 2. For this.03 2.2 169. ng 197 60 99.3 66.1 137.

In addition.73x – 2. LLC . SDxy Standard deviation.65 1. with the fulfillment of the equal distribution of the standard in the range of the calibration line.9987 . a Residual standard deviation.198 Graph: 200 180 160 140 Signal 120 100 80 60 40 20 0 0 20 40 Quality Assurance and Quality Control y = 1.62 1.974 0.99 ng 2. and the range. SDb Standard deviation.3SD b 2.45 ng 15 1. determine the LOD values. requires a high linearity procedure. ng 100 120 Excel file: exampl_valid31_2. r.xls Conclusion: A high value of the regression coefficient. Solution: n Slope. r The LOD value was determined using the equation: LOD = LOD (SDxy) LOD (SDa) LOD (mean) 3. SDa Regression coefficient. check the correctness of the LOD determination.© 2009 by Taylor & Francis Group. the LOQ value.91 ng 1. Problem 3: Based on the series of measurement results for the standard solutions with the three lowest mercury content levels (20.4 0.9986 60 80 Hg Content.11 r = 0. and 60 ng).023 0. b Intercept. 40.

ng 50 60 70 y = 1.4 ng Whereas the range was presented as: 7.xls Problem 4: Based on the series of results for the three real samples (lyophilized muscle tissue of great cormorant).65 199 The correctness of LOD determination was made according to the equations: 10 ⋅ LOD > cmin where cmin = 20 ng. LOD < cmin Conclusion: The determined LOD value is correct. .62x + 1. calculate the repeatability. LLC .Method Validation Graph: 120 100 80 Signal 60 40 20 0 0 10 20 30 40 Hg Content. Based on the relationship: LOQ = 3 · LOD The LOQ value was calculated to be: LOQ = 7.4 ÷ 100 ng Excel file: exampl_valid31_2.© 2009 by Taylor & Francis Group.

ng 78.8 21.82 78.© 2009 by Taylor & Francis Group.22 Hg Concentration.00 Mean Hg Concentration.90 90. the Dixon Q test was applied (with a significance level α = 0.26 3.01 79.69 74.27 2.41 3.21 2.21 3.62 70.44 2.82 72.25 2.2 37.97 3.79 3.69 3.4 20.15 3.4 32.5 33.8 28.2 27. mg 1 2 3 4 5 6 7 21.54 3.32 2. one should check whether there are no outliers in the measurement results series.3 24. mg 1 2 3 4 5 6 7 25.09 3. ppm 2. For this.1 Hg Content. LLC . ppm 3.33 3.200 Quality Assurance and Quality Control Data: Measurement results for individual samples: Sample 1 Sample Mass.50 Mean Sample 3 Sample Mass.25 Solution: Before performing the calculation.3 35.84 71.87 3.14 71.0 Hg Content.1 20.11 3.3 22.61 3.20 91.7 31.77 Hg Concentration.68 Mean Sample 2 Sample Mass. ng 64.9 24.8 30.5 Hg Content.9 19.92 3.11 2. ppm 3.17 2.93 72.77 80.91 81.05). .11 78.93 84.37 85. ng 83.7 22.7 20.9 25.94 92.42 66. mg 1 2 3 4 5 6 7 30. in order to indicate precision.48 81.

063 0.507 Sample 3 7 0.162 4. Excel files: exampl_valid31_3a.Method Validation Sample 1 No.678 2.05 was chosen).118 3. determine the trueness value (as a recovery value). Sample 1 7 0. there are no outliers.116 201 Conclusion: In the series of measurement results.xls Determinations were conducted for three different real samples.24% Excel file: exampl_valid31_4.32 0.67% 1. before calculating the repeatability value (as the mean of the coefficient variation for the results of the three series results).30% No results. of results. however.68 8. LLC .38 Sample 3 7 0.250 0.xls exampl_valid31_3b. therefore.xls Problem 5: Based on the results determined for certified reference material samples (BCR-463 — Tuna fish: total Hg and methylmercury). The calculated repeatability value. the homogeneity of the variation should be checked for the series of results to be analyzed.753 2. n Standard deviation.xls exampl_valid31_3c. Data: results are given as (ng/mg): Data 1 2 3 4 5 2.516 2. SD CV.33 0.918 .107 4.© 2009 by Taylor & Francis Group.76% Sample 2 7 0.970 2.163 0.43 0. can be calculated as a mean value from the coefficient of variation counted for three series: CVrepeatability = 4.364 Sample 2 7 0. n Range. The Hartley Fmax test was applied with this aim (a significance level of α = 0.182 0. % Fmax Fmaxo Conclusion: There are no statistically significant differences in variation values. R Q1 Qn Qcrit 7 0.

as well as the uncertainty value from the indication of trueness.767 0. the uncertainty value related to the unrepeatability of the measurement results.2% Determined Conclusion: Results obtained with the use of the developed method are correct.184 0.85 Quality Assurance and Quality Control U 0.5 1 0.1 ± 8. Excel file: exampl_valid31_5.202 Value CRM Solution: Mean SD U R U (k = 2) 2.1% 8. . obtained with the use of the elaborated method.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. the following were recognized: the uncertainty value resulting from the calibration curve. LLC .5 0 CRM Trueness = 97.5 2 1.164 97.© 2009 by Taylor & Francis Group.xls Problem 6: Estimate an uncertainty value for the determination results of the total mercury content in real samples.16 k 2 2. Solution: As the main components of the uncertainty budget.5 3 2.

10 0.39 4.77 0. % usmpl. ng Hg.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.63 4. % Graph: 200 180 160 140 Signal 120 100 80 60 40 20 0 0 20 40 60 Content. Sample 1 No results.14 2.51 0.69 3.9 Sample 3 7 72.14 0.© 2009 by Taylor & Francis Group. % urep. ppm Usmpl (k = 2). ng 80 100 120 y = 1.25 1. Calculations were conducted for minimal weighted masses for each of the analyzed real samples.17 0. ppm Usmpl (k = 2). concentration.21 9. % utrue.43 0. n Minimum Hg content.73x – 2.11 r = 0.61 0.96 1. % usmpl.11 1.96 1. conducting measurements for the series of standard solutions.29 8.22 0.9986 Sample 2 7 71.10 4.2 .34 9.80 4. an approximation of measurement points of the calibration line using line regression) was conducted on the basis of the calibration parameters. LLC .82 3. ppm ucal.0 7 61.

precision class. 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. LLC . and reference materials used. etc.77 ± 0.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: 2.34 ppm Excel files: exampl_valid31_6a.22 ± 0. standards.xls exampl_valid31_6c.4 Conclusions Validation of an analytical method should be finished with a final report containing [2. producer. 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 . together with precise specification (purity.). 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.29 ppm sample 3: 3. quality. 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. in case of laboratory synthesis.xls exampl_valid31_6b.© 2009 by Taylor & Francis Group.25 ± 0.21 ppm sample 2: 3.xls exampl_valid31_7. for example. and.

The analytical procedure is intended for determining whole mercury content in muscle tissue samples from great cormorants. Germany) • Buffer solution. .16 μg/g.00 ± 0. However.. During the analytical procedure. in this case. concentration 100. 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.32 Problem: Based on the validation parameters indicated for the analytical procedure in Example 7. Poland) • Nitric acid — suprapure (Merck. Ltd. Kyoto. Marine birds spend a significant portion of their lives in coastal or marine environments and are exposed to a wide range of chemicals. LLC . The released mercury is detected using the cold atomic absorption method at a wavelength of 253. Belgium) • Deionized water Preparation of Standard Solutions There are various methods available for preparing standard solutions. making them susceptible to bioaccumulation of pollutants. Inc. USA) • l-Cysteine.© 2009 by Taylor & Francis Group.3% HCl (Inorganic Ventures. the following reagents are used: • Mercury standard — MSHG. 100-ppm. IRMM (Geel.. 98% (Nacalai Tesque. wide geographical ranges. Poland) • CRM: BCR-463: Total and methyl mercury in tuna fish. The analytical procedure pertains to the indication of total mercury content (after converting the total mercury content into an atomic form). Great cormorants (Phalacrocorax carbo) were used as bioindicators for mercury contamination. Inc. because most occupy higher trophic levels. Solution: Seabirds are useful bioindicators of coastal and marine pollution.05 (POCh. The mercury collection agent is then heated up to 600°C to release atomic mercury. pH 7. Japan) • Additive B (Wako Pure Chemical Industries. Nippon Instrument Corporation obtained good results using l-cysteine.31. due to their specific feeding habits.Method Validation 205 • Recapitulation and conclusions • Confirmation and signature of the person responsible for the test and confirmation of the validation. dark place. standard solutions should be kept in a cool..22 μg/ml in 3. mercury is further atomized. 2. Therefore.48 ± 0. Japan) • Additive M (POCh. Example 7. and long life span. Mercury content is determined in lyophilized muscle tissue of great cormorants. create a validation report.7 nm in the detector’s absorption cell.85 ± 0. solution stability degrades with age or due to long storage in a warm place. A sample is thermally decomposed. Measurements of the content of total mercury will be performed using the cold vapor AAS technique.

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

LLC . . As a method of removing any substances that could interfere with the measurement. preheating the gold-coated diatomite particle support collection agent allows for the measurement to be done without the influence of any organic components. if not done so. which would adversely affect measurements. The PC reads the resulting measurements in the order that the various analyses.32-2. Homogenized samples should be directly weighed (10–50 ± 0.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. To remove any interfering substances that are generated when thermally decomposing a sample.32-2 Schematic diagram of MA-2000.1 mg) into precleaned combustion boats and automatically inserted into the Mercury/MA-2000 system (NIC. Japan). it is recommended that two types of additives be used: additive B (activated alumina) and additive M (sodium carbonate and calcium hydroxide). Before use. including statistical calculations. Analytical Procedure Carefully separated bird tissues should be immediately deeply frozen. A block diagram of the apparatus is presented in Figure 7.© 2009 by Taylor & Francis Group. Homogenized samples should be stored in a refrigerator with a temperature of 0–6°C. to be measured. which would be physically absorbed to a certain extent. gas washing is performed. In addition. and homogenized. freeze-dried (lyophilized). the additives should be subjected to a heat treatment in a heat treatment furnace at 750°C for at least 3 hours. can be performed.

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).32-3. Taking this into account. cover with additive B. dose at least five different volumes of the standard solution with a concentration of 1-ppm from the 20.to 100μL section. repeat at least three times. as well as the level of its homogeneity.© 2009 by Taylor & Francis Group. 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.208 Sample Quality Assurance and Quality Control Additive M Put the dispensed sample onto additive M. the maximum sample mass is restricted by the maximum . Using an automatic pipette. once more cover with additive M. Additive B en. The sample boats that will be used should also be subjected to the same heat treatment. this value should not be less than 20 mg. Sample M B M FIGURE 7. Additive Finally. For each mercury mass. The method for using the additives is presented schematically in Figure 7. LLC . However.32-3 Method for using the additives.

Compare these values with the determined values that are contained in the report. For each of the solutions. The next steps of the analytical procedure are schematically presented in Figure 7. the calibration curve corresponds to the range of Hg values in lyophilized tissue 0. three independent measurements were conducted. the values for the following parameters were indicated.32-4. LLC . • Temperature program – MODE2. Draw the calibration curve and indicate the value of the regression parameters. which can be introduced into the ceramic boat and consequently into the furnace. Linearity A series of standard solutions was prepared with a mercury content of 20–100 ng. regression parameters were indicated and . substance mass. The absorption radiation measurement is realized for mercury’s characteristic wavelength. 2. This value should not exceed 200 mg.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. and based on the obtained results. or the values corresponding to the section of values that most often appear in muscle tissue of great cormorants. The amalgamation reaction is a selective reaction for mercury.1–5.© 2009 by Taylor & Francis Group. ASA ANALYSIS FINAL DETERMINATION Calculation of Hg content on the basis of calibration curve parameters FIGURE 7.32-4 A schematic presentation of the analytical procedure for the determination of total mercury content in muscle tissue of great cormorant samples. During the analytical validation procedure. Selectivity Applying the measurement technique ensures high selectivity for indicating mercury for two reasons: 1. Taking this into account.

210 Quality Assurance and Quality Control the calibration curve was determined.730 −2. n Slope.7 0.11 r = 0. r 25 1.32-1. r after fulfilling conditions for a “uniform” concentration distribution in terms of the calibration curve.11 2. A high regression coefficient.019 1.32-5 Calibration curve for linearity determination. SDa Regression coefficient. LLC . TABLE 7. parameters that determined LOD values. ng 100 120 y = 1.9986 FIGURE 7. SDb Standard deviation of the intercept. and the calibration curve is presented in Figure 7. and 60 ng). The obtained values are presented in Table 7. 40. b Intercept.73x – 2.32-5.3SD b A calibration plot is presented in Figure 7. and the relationship: LOD = 3.32-6.© 2009 by Taylor & Francis Group.3 0. SDxy Standard deviation of the slope.32-1 Calculated Regression Parameters for Linearity Determination Number of results. commands a high linear procedure. a Residual standard deviation. A calibration curve was outlined based on the obtained measurement results. 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 (20.9986 200 180 160 140 Signal 120 100 80 60 40 20 0 0 20 40 60 80 Hg Content. .

© 2009 by Taylor & Francis Group. equaling 7. it is equal to: 7.4 ng (assuming the mass of the 20 mg sample corresponds to a concentration of 0. assuming the mass of the sample that underwent indication is an even 20 mg.2%. to a maximum standard solution concentration used for calibration. 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. .65 211 FIGURE 7.32-7. A series of five independent determinations are conducted. Range The measurement range is a concentration range from the LOQ section. The determined repeatability value is equal to: CVrepeatability — 4.37). corresponds to the mercury concentration in tissue samples of an even 0. This value is determined as an average CV value for three series.45 ng. the LOQ value was determined to be LOQ = 3 · LOD.32-6 Calibration curve for LOD determination. which. However.Method Validation 120 100 80 Signal 60 40 20 0 0 10 20 30 40 Hg Content. ng 50 60 70 y = 1. corresponds to a mercury concentration of: 0.62x + 1.4 ÷ 100 ng which.12.24%.37 ÷ 5.1 ± 8. The LOD value was deemed to be 2. Therefore.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. The determined trueness value is graphically presented in Figure 7. LLC . assuming the sample mass that underwent indication of an even 20 mg.

During the revalidation process. the uncertainty value related to the unrepeatability of measurement results. in relation to the repeatability of measurement results utrue = the standard uncertainty of results related to the determination of trueness. Uncertainty The main components of the uncertainty budget were the uncertainty value resulting from the determination of the calibration curve.© 2009 by Taylor & Francis Group. Determination of standard uncertainty related to the calibration step (preparation of a series of standard solutions. LLC .5 3 2. Calculations are conducted for minimal masses for each of the analyzed real samples. 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 7.32-7 Comparison of the determined value with a certified Hg content value — trueness determination. .5 0 CRM Quality Assurance and Quality Control Determined FIGURE 7. as well as repeatability.5 2 1. 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.212 3.32-1). related to calibration urep = the standard uncertainty for determination results for real samples. indicated based on CRM determinations.5 1 0. realization of measurements for the 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. attention should be paid to the stability of the calibration curve.1% (as an average of the three samples). The calculated uncertainty value for k = 2 equals 9. The consecutive parameter is trueness. whose value should not exceed CV = 5%. as well as the uncertainty value indicating trueness.

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

Results obtained while using this method are characterized by low uncertainty (about 10%)... De Loose. 4. International Conference on Harmonization (ICH) of Technical Requirements for the Registration of Pharmaceuticals for Human Use: Validation of Analytical Procedures: Metrology.K. A. Wisconsin Department of Natural Resources. Traverniers... Marine Environmental Support Office. “A closer look at analytical signals. Vogelgesang. 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. assuming the minimal mass is an even 20 mg.” Accred. 2007. Nippon Instruments Corporation. repeatability (CV = 4.. 37.5 ng of total mercury) in the sample. Draft 1. Rev. ICH-Q2A. 6. 126. 2000. Johnston. 3. 5. 1998.labcompliance. NIC-600-2009-04. Chem. Validation and Qualification in Analytical Laboratories. J. The procedure is characterized by high selectivity. 2004. 23. 376–382. 256–259. www. “The role of and place of method validation in the quality assurance and quality control (QA/QC) system. Assur. Specifying and evaluating analytical chemistry quality requirements for ecological risk assessment. P. and Valente. 1996. 380. “Measurement of near zero concentration: recording and reporting results that fall close to or below the detection limit. R. Jacek Namies ´nik. FAO/IAEA Training and Reference Centre for Food and Pesticide Control.” Instruction Manual. Gdan ´ sk.24%).. 2. Analytical method validation and quality assurance.or.” Anal. 20 June 2008. identification.. LLC .at/programmes/nafa/d5/trcfpc/d5-trcfpc. Geneva. No. Huber. K.. International Conference on Harmonization (ICH) of Technical Requirements for the Registration of Pharmaceuticals for Human Use: Text on Validation of Analytical Procedures.com. References 1.. Practical Approach to Validation of Methods for Analysis of Residues. corresponds to a concentration of 0. Ambrus. J.1 ± 8. http://www. . trueness (recovery = 97. 242–255. 1994.© 2009 by Taylor & Francis Group.214 Quality Assurance and Quality Control Analytical Methods Committee. R.” Crit. ICH-Q2B. 1998. Chem. and therefore. San Diego. Konieczka.2%). 535–552. and determination: a statistical approach for practitioners. 2001.” Trends Anal. 1999.. “Trends in quality in the analytical laboratory: II... Technical Memorandum 99-01.html. 7. Geneva. “Mercury/MA-2000. Analytical Detection Limit Guidance.. and Hädrich.iaea. 173–190.. Mercury Analysis System. and Van Bockstaele. “Limits of detection. I. Danzer. M. Piotr Konieczka. The validation report was checked and confirmed by Prof.M.” Analyst. 6. high precision. and allows for the discovery of trace amounts of mercury in analyzed samples. L. 2004. Qual.12. E. Chem. The validation process was conducted by Dr. 1996. Bioanal. The estimated LOD value (LOD = 2. Anal.

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

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

306 2.861 2.776 2.841 4.021 2.604 4.648 2.878 2.576 217 .012 2.120 2.179 2.706 4.093 2.977 2.639 2.05 12.009 2.626 2.947 2.© 2009 by Taylor & Francis Group.032 3.160 2.250 3.Appendix Table A.845 2.169 3.131 2.086 2.106 3.262 2.678 2.1 Critical Values.707 3.000 1.365 2. LLC .048 2.149 2.042 2.660 2.447 2.074 2.182 2.056 2.960 α = 0.101 2.01 63.064 2.030 2.990 1.355 3.706 2.925 5.779 2.690 2.984 1.571 2.303 3.763 2.567 9.797 2.228 2.994 1.921 2.716 2.201 2.110 2.819 2.055 3.750 2.898 2. 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 = 0.499 3.014 2.

814 1.423 2.903 1.949 1.956 1.348 2.923 1.936 1.487 2.935 1.910 1.492 2.368 2.948 1.955 1.945 1.916 1.218 Appendix Table A.537 2.208 2.524 2.399 2.479 2.920 1.960 α = 0.931 1.294 2.142 2.© 2009 by Taylor & Francis Group.498 2.870 1.941 1.412 2.645 1.05 1.518 2.926 1.944 1.385 2.509 2.470 2.414 1.547 2.943 1.933 1.460 2.940 1.256 2.918 2.576 .950 1.324 2.440 2.954 1.715 1.848 1.553 2.01 1.937 1.529 2.953 1.051 2.885 1.951 1.432 2.928 1.447 2.454 2. LLC .409 1.757 1.895 1.542 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.

679 0.72 4.43 2.66 α = 0.4 6.63 4.91 4.6 7.560 0.590 0.39 10 49.76 1.86 2.32 5.17 8 45.90 4.80 2.349 α = 0.57 1.36 3.83 2.23 4.91 4.60 4.73 4.© 2009 by Taylor & Francis Group.94 4.642 0.39 4.96 1.63 5 37.46 4.03 7 43.47 11 50.23 2.20 5.31 4.11 4.16 4.555 0.4 6.69 3.84 3.01 3.0 7.507 0.80 5.10 0.20 5.31 5.0 4.55 4.83 4.14 1.65 4.92 4.Appendix 219 Table A. LLC .82 4.96 3.40 3.0 3.04 3.88 3.03 4.29 9 47.889 0.62 4.4 Critical Values of Parameter zα n 2 3 4 5 α = 0.65 4.36 4.48 4.06 1.64 4.637 0.74 3.1 5.05 2.3 Critical Values of z Parameter for Significance Level α = 0.399 0.64 3.22 4.98 3.71 1.44 4.765 0.4 6.527 .30 4.780 0.78 4.01 4.988 0.482 0.412 α = 0.60 4.15 3.72 4.468 0.33 5.08 3.72 5.77 3 27.81 4.52 4.67 3.50 α = 0.95 2.370 0.83 5.56 4.437 0.1 6.5 Critical Values (Qcrit) of Dixon’s Q Test f 3 4 5 6 7 8 9 10 α = 0.98 Table A.8 5.1 6.05 n f 1 5 10 15 20 30 40 60 120 ∞ 2 18.67 4.00 4.01 0.58 5.49 3.46 5.12 4.24 4.10 2.99 5.46 1.01 2.23 4.434 0.45 4.60 3.79 3.40 5.62 Table A.74 4.941 0.77 4.86 6 40.886 0.05 0.10 4.10 4.08 4.60 5.557 0.44 3.89 2.17 5.31 4 32.92 3.33 4.37 4.698 0.58 3.30 4.55 12 53.

377 0.458 0.472 0.670 0.627 0.417 0.489 0.740 0.611 0.467 0.450 0.412 0.672 0.580 0.530 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 = 0.462 0.477 0.517 0.628 0.© 2009 by Taylor & Francis Group.438 .526 0.535 0.483 0.429 0.423 0.586 0.579 0.443 0.647 0.635 0.608 0.697 0.926 0.502 0.495 0.402 0.374 0.514 0.407 0.478 0.555 0.994 0. LLC .610 0.501 0.680 0.436 0.442 0.564 0.970 0.388 0.717 0.01 0.454 0.489 0.567 0.459 0.546 0.397 0.594 0.544 0.220 Appendix Table A.565 0.371 α = 0.605 0.384 0.710 0.451 0.510 0.821 0.05 0.393 0.446 0.569 0.529 0.479 0.468 0.381 0.829 0.

48 20.99 7.41 32.81 18.41 37.98 44.59 28.80 36.65 α = 0.64 9.49 11.19 37.34 13.84 5.29 41.67 23.68 25.03 22.87 30.31 .57 38. LLC .© 2009 by Taylor & Francis Group.69 29.30 27.81 9.Appendix 221 Table A.36 23.7 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 = 0.59 14.41 34.00 26.64 42.00 33.14 30.31 19.28 15.09 21.07 12.72 26.07 15.93 40.92 18.68 21.22 27.21 24.51 16.05 3.92 35.58 32.14 31.17 36.21 11.67 33.09 16.01 6.

01 4.23 5.67 5 19.88 10.56 3.06 7.49 6.76 9.84 27.64 3.46 6.59 6.53 9.78 27.15 4.94 27.47 3.22 5.69 5.86 4.48 5.00 99.79 3.10 7.78 2.39 4.09 14.34 5.21 3.01 3.94 4.71 6.71 6.90 4.44 6.63 6.97 4.34 8.76 27.33 5.68 6.54 4.37 5.62 3.96 4.41 8.13 5.82 4.28 29.79 7.80 4.58 6.27 4.16 99.99 3.78 10.63 10 19.98 5.66 4.54 3.19 3.36 8.07 7 19.30 9.78 4.14 10.40 8.13 5.86 6.80 3.74 3.42 3.97 4.09 5.47 3.21 8.25 9.95 4.46 .63 3.98 4.50 6.12 7.01 (Bottom Row) f1 f2 2 3 4 5 6 7 8 9 10 11 2 19.00 5.45 4.27 5.36 99.06 2.46 3.24 6.39 8.33 8.81 6.14 5.87 7.79 13.82 10.26 15.20 5.18 5.84 3.18 2. Snedecor’s F Test for Significance Level α = 0.22 4 19.26 2.19 3.05 (Top Row) and α = 0.45 4.34 6.74 9 19.01 9.10 7.07 5.71 3.20 3 19.39 99.84 7.88 8 19.88 27.23 5.8 Critical Values.96 14.55 4.97 7.48 6.37 3.32 6 19.81 27.38 8.04 14.35 3.15 4.62 3.60 6.10 3.59 16.98 7.26 3.52 5.07 7.92 4.29 5.39 3.38 99.55 30.© 2009 by Taylor & Francis Group.74 10.17 9.54 11 19.21 4.36 5.63 6.94 18.33 99.46 8.15 8.30 99.73 6.91 6.55 3.06 3.67 4.37 99.28 8.39 15.03 7.16 15.03 3.67 6.65 4.06 4.12 28.25 99.19 11.57 3.95 10.85 2.31 5. LLC .45 4.87 3.91 3.74 9.95 2.59 3.98 3.02 4.85 3.01 28.93 14.00 3.222 Appendix Table A.99 3.40 99.05 10.82 3.70 9.10 5.39 5.69 6.41 12.26 8.35 8.02 4.05 4.00 14.

8 9.2 19.3 12.40 4.0 28.95 3.59 4.3 11.70 8.19 4.2 10.29 5.03 7.5 8. LLC .18 6.5 10.9 33.46 2.44 7.Appendix 223 TABLE A.5 22.29 2.33 1.1 9.34 4.67 4.38 6.5 18.45 8. Hartley’s Fmax Test for Significance Level α = 0.5 13.4 8.6 13.00 5.8 9.1 12.5 11.6 14.94 3.26 1.00 10 550 104 44.00 4 142 39.54 2.00 5 202 50.10 3.15 5.61 1.36 2.17 1.9 Critical Values.07 1.22 1.91 2.1 9.01 5.37 3.3 9.85 1.95 7.8 15.7 9.11 6.00 11 626 114 48.00 9 475 93.12 2.4 9.7 25.0 15.92 4.37 3.9 41.02 2.00 7 333 72.6 20.© 2009 by Taylor & Francis Group.2 16.03 3.7 10.00 .60 7.42 4.82 4.5 37.5 18.8 8.72 2.12 7.24 3.95 2.05 k f 2 3 4 5 6 7 8 9 10 15 20 30 60 ∞ 2 39.1 24.66 5.34 4.96 1.87 5.00 8 403 83.7 17.49 3.29 2.54 2.80 6.30 1.9 16.85 3.5 27.0 29.11 1.7 13.91 8.31 5.94 6.00 3 87.41 7.40 1.43 4.21 2.2 20.76 2.77 4.8 15.04 1.86 2.78 2.67 1.6 26.3 12.01 3.0 11.99 4.7 10.00 6 266 62.78 8.7 10.68 3.7 15.

70 1.69 1.70 1.94 1.67 1.71 1.65 1.76 1.71 1.70 1.64 1.85 1.71 1.73 1.70 1.2 1.81 1.73 1.70 1.78 1.64 0.72 1.75 1.70 1.8 1.85 1.94 1.94 1.73 1.73 1.75 1.70 1.90 1.70 1.72 1.76 1.86 1.79 1.73 1.82 1.70 1.72 1.72 1.70 1.90 1.79 1.76 1.94 1.85 1.69 1.85 1.82 1.72 1.70 1.71 1.© 2009 by Taylor & Francis Group.72 1.94 1.80 1.65 1.71 1.75 1.72 1.64 1.66 1.82 1.73 1.80 1.69 1.82 1.82 1.69 1.86 1.85 1.72 1.86 1.73 1.64 1.94 1.65 1.70 1.72 1.73 1.76 1.73 1.78 1.66 1.72 1.64 0.78 1.94 1.76 1.69 1.76 1.70 1.224 Appendix TABLE A.71 1.72 1.90 1.68 1.78 1.85 1.64 1.66 1.81 1.69 1.64 1.65 1. LLC .64 1.72 1.73 1.86 1.78 1.76 1.67 1.72 1.72 1.72 1.76 1.73 1.71 1.73 1.70 1.66 1.76 1.71 1.86 1.66 1.75 1.69 1.80 1.80 1.65 1.65 1.69 1.64 1.82 1.86 1.73 1.69 1.79 1.76 1.64 0.70 1.65 1.81 1.85 1.64 0.81 1.69 1.75 1.73 1.78 1.69 1.78 1.69 1.76 1.72 1.70 1.71 1.68 1.86 1.71 1.71 1.75 1.71 1.78 1.80 1.79 1.67 1.72 1.65 1.73 1.64 1.69 1.81 1.73 1.73 1.71 1.0 1.64 0.80 1.6 1.69 1.68 1.90 1.67 1.82 1.66 1.81 1.82 1.90 1.82 1.74 1.80 1.94 1.70 1.80 1.90 1.73 1.70 1.65 1.71 1.76 1.71 1.70 1.71 1.75 1.76 1.75 1.94 1.71 1.76 1.70 1.76 1.65 1.73 1.70 1.76 1.81 1.81 1.66 1.90 1.71 1.71 1.73 1.86 1.64 0.76 1.90 1.66 1.76 1.69 1.86 1.70 1.71 1.75 1.71 1.71 1.78 1.72 1.73 1.81 1.81 1.79 1.7 1.70 1.69 1.71 1.1 1.85 1.75 1.73 1.76 1.64 1.76 1.76 1.71 1.72 1.80 1.82 1.64 0.73 1.70 1.71 1.75 1.82 1.73 1.74 1.71 1.72 1.72 1.66 1.86 1.65 1.75 1.78 1.64 8 10 15 20 ∞ .68 1.70 1.9 1.68 1.65 1.73 1.75 1.76 1.69 1.69 1.66 1.85 1.72 1.73 1.70 1.71 1.85 1.71 1.71 1.73 1.66 1.72 1.73 1.65 1.70 1.80 1.79 1.80 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 ∞ 0.73 1.64 1.70 1.67 1.72 1.0 1.71 1.73 1.81 1.76 1.78 1.76 1.90 1.71 1.68 1.80 1.90 1.85 1.67 1.76 1.66 1.3 1.72 1.76 1.76 1.76 1.79 1.76 1.74 1.94 1.86 1.65 1.74 1.70 1.86 1.73 1.5 1.72 1.72 1.72 1.72 1.76 1.72 1.73 1.65 1.74 1.79 1.73 1.81 1.4 1.76 1.72 1.65 1.94 1.65 1.70 1.72 1.65 1.71 1.73 1.94 1.71 1.68 1.75 1.68 1.79 1.70 1.65 1.72 1.73 1.90 1.73 1.70 1.70 1.79 1.69 1.67 1.72 1.79 1.72 1.73 1.68 1.65 1.73 1.71 1.64 0.74 1.64 1.67 1.90 1.10 Critical Values vo of Aspin-Welch Test for Significance Level α = 0.76 1.76 1.65 1.79 1.76 1.65 1.73 1.78 1.74 1.72 1.64 0.85 1.82 1.

138 0.906 0.906 0.261 0.252 0.293 0.834 0.230 0.131 0.127 0.492 0.121 0.177 0.147 0.684 0.224 0.265 0.215 0.248 0. LLC .532 0.124 0.418 0.218 0.255 0.147 0.307 0.181 0.124 0.131 0.168 0.308 0.149 0.209 0.168 0.391 0.450 0.165 0.781 0.516 0.391 0.325 0.121 0.197 0.624 0.465 0.246 0.696 0.335 0.294 α= 0.198 0.273 0.838 0.114 α= 0.508 0.161 0.425 0.633 0.189 0.407 0.288 0.221 0.167 0.798 0.403 0. .393 0.155 0.995 0.959 0.234 0.205 0.707 0.175 0.164 0.471 0.427 0.287 0.794 0.228 0.101 0.684 0.212 0.164 0.928 0.246 0.128 α= 0.116 0.599 0.181 0.331 0.262 0.243 0.204 0.220 0.352 0.01 0.144 0.664 0.238 0.126 n=5 α= 0.450 0.145 0.310 0.259 0.135 0.684 0.392 0.208 0.251 0.235 0.281 0.05 0.388 0.270 0.746 0.238 0.598 0.306 0.138 0.237 n=3 α= 0.307 0.140 0.788 0.209 0.326 0.354 0.278 0.05 – 0.347 0.281 0.174 0.357 0.230 0.864 0.363 0.142 0.968 0.241 0.235 0.137 0.288 0.200 0.871 0.262 0.480 0.349 0.568 0.270 0.332 0.616 0.564 0.291 0.377 0.366 0.203 0.05 0.445 0.256 0.475 0.097 p = number of laboratories.676 0.942 0.413 0.160 0.242 0.356 0.157 0.118 0.463 0.293 0.137 0.202 0.305 0.291 0.131 0.452 0.520 0.111 0.372 0.339 0.146 0.186 0.134 0.680 0.05 0.127 0.297 0.403 0.329 0.151 0.937 0.590 0.387 0.242 0.319 0.322 0.01 0.274 0.300 0.722 0.979 0.280 0.343 0.883 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= 0.271 0.158 n=4 α= 0.159 0.276 0.372 0.160 0.196 0.119 0.506 0.134 0.05 0.575 0.877 0.173 0.151 α= 0.113 0.316 0.727 0.434 0.343 0.230 0.168 0.240 0.134 0.127 0.188 0.389 0.192 α= 0.215 0.626 0.334 0.402 0.318 0.191 0.151 0.172 0.155 0.249 0.373 0.243 0.301 0.262 0.532 0.629 0.418 0.478 0.099 0.883 0.425 0.159 0.150 0.466 0.168 0.204 0.615 0.480 0.515 0.348 0.232 0.191 0.197 0.768 0.184 0.108 0.358 0.300 0.213 0.417 0.155 0.124 0.325 0.198 0.144 0.445 0.332 0.176 0.286 0.116 0.304 0.754 0.360 0.447 0.318 0.397 0.993 0.161 0.179 0.155 0.173 0.514 0.250 0.276 0.588 0.150 0.140 0.544 0.267 0.781 0.255 0.261 0.496 0.197 0.© 2009 by Taylor & Francis Group.208 0.111 0.179 0.246 0.185 0.355 0.164 0.144 0.431 0.184 0.117 0.190 0.303 0.481 0.01 – 0.721 0.129 0.01 0.382 0.573 0.365 0.131 0.220 0.192 0.212 0.330 0.189 0.521 0.164 0.841 0.288 0.219 0.141 0.166 0.308 0.536 0.01 0.504 0.343 0.262 0.178 0.541 0.140 0.108 n=6 α= 0.159 0.229 0.204 0. n = number of results for one level.179 0.134 0.120 0.975 0.200 0.170 0.Appendix 225 Table A.638 0.423 0.967 0.223 0.172 0.103 0.106 0.392 0.939 0.209 0.480 0.131 0.222 0.193 0.369 0.274 0.437 0.312 0.229 0.602 0.255 0.154 0.196 0.561 0.220 0.653 0.114 0.793 0.154 0.438 0.332 0.570 0.718 0.182 0.316 0.553 0.185 0.371 0.172 0.214 0.

031 3.2016 0.4638 0.6316 0.4376 0.003 3.218 3.4510 0.620 2.226 Appendix Table A.636 2.5672 0.5941 0.0090 0.0563 0.3822 0.286 3.5192 0.651 2.3927 0.139 2.3398 0.0349 0.4857 0.764 1.253 3.381 Lower α = 0.968 3.4556 0.973 2.001 3.3112 0.3367 0.12 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 = 0.733 2.369 3.5091 0.020 2.356 3.2767 0.781 2.© 2009 by Taylor & Francis Group.155 1.330 3.0018 0.215 2.2530 0.908 2.135 3.991 3.755 2.758 2.5381 0.2990 0.979 2.0116 0.5574 0.4025 0.859 2.343 3.3761 0.1492 0.4214 0.5470 0.5766 0.4391 0.301 3.496 1.6247 0.126 2.5789 0.6101 0.5360 0.060 3.1150 0.01 – 0.316 3.274 2.887 2.4759 0.5554 0.199 3.178 3.5288 0.4994 0.6175 0.1738 0.852 2.564 2.5469 0.3585 0.3200 0.462 2.4985 0.0002 0.2280 0.014 3.087 3.549 2.1864 0.894 2.876 2.709 2.270 3.5714 0.681 2.155 1.025 3.924 2.938 2.3603 0.4085 0.387 2.1448 0.112 3.893 2.507 2.6445 p = number of laboratories .952 2.05 – 0.822 2.4234 0.0851 0.0708 0.157 3.6023 0.335 2.5862 Lower α = 0.036 Two Greatest and Two Smallest Upper α = 0.699 2.412 2.482 2.806 2.715 1.2537 0.1101 0.5123 0.965 2.2213 0.5856 0.6382 0.4711 0.932 2.05 1.2836 0.01 1.585 2.481 1.0308 0.5245 0.4875 0.841 2.5636 0.0000 0.290 2. LLC .802 2.236 3.

60 1.60 1.57 1.54 1.10 4 1.89 1.47 2.77 1.91 2. LLC .50 1.79 1.32 2.68 1.55 1.06 2.46 1.65 1.79 1.25 2.65 1.46 1.42 2.63 1.77 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 1.56 1.78 1.70 1.44 2.59 1.53 1.57 1.86 1.55 1.45 2 1.57 1.53 1.51 1.05 2.91 1.64 1. .89 1.74 1.53 1.71 7 1.47 2.65 1.65 1.07 2.59 1.60 1.52 1.53 1.51 1.70 1.69 1.79 1.53 1.69 1.05 2.79 1.44 2.00 2.70 1.38 2.51 1.30 2.99 2.62 1.09 2.58 1.59 1.68 1.60 1.60 1.37 2.08 2.03 2.48 2.39 1.55 1.13a Parameters h and k of Mandel’s Test for Significance Level α = 0.78 1.56 1.56 1.29 2.85 1.02 2.40 2.64 1.79 1.71 1.65 1.62 1.54 1.87 1.65 1.50 1.62 1.27 2.78 1.67 1.53 1.64 1.71 1.52 1.71 1.47 2.56 1.90 1.49 1.42 2.09 2.79 1.14 2.84 1.41 2.55 1.42 2.© 2009 by Taylor & Francis Group.36 2.87 1.Appendix 227 TABLE A.52 1.66 1.76 1.91 1.60 1.97 1.64 1.20 2.53 1.88 1.63 1.43 1.90 1.55 1.49 2.64 1.52 1.01 2.05 2.53 1.65 8 1.74 1.13 2.06 2.33 2.52 1.65 1.71 1.90 1.85 1.63 1.45 2.58 1.49 3 1.48 2.53 1.59 1.71 1.56 1.43 1.53 1.44 2.82 1.60 1.25 2.67 1.94 1.59 1.70 1.60 1.06 2.71 1.04 2.36 2.98 2.61 9 1.64 1.75 1.89 1.34 2.48 1.60 1.90 1.71 1.51 1.53 1.79 6 1.08 2.22 2.59 1.58 1.46 2.88 1.53 1.91 5 1.53 1.70 1.49 1.71 1.56 1.66 1.57 1.55 1.52 1.53 p = number of laboratories.47 1.77 1.70 1.76 1.63 1.45 1.60 1.49 2.56 1.55 1.72 1.65 1.79 1.54 1.07 2.07 2. n = number of results for one level.53 1.90 1.52 1.08 2.90 1.50 1.35 2.73 1.90 1.69 1.57 1.78 1.56 1.15 1.85 1.60 1.65 1.56 1.68 1.57 10 1.46 2.65 1.60 1.39 2.87 1.58 1.48 1.76 1.41 2.08 2.48 1.51 1.56 1.69 1.53 1.58 1.32 2.77 1.45 2.09 2.44 2.43 2.73 1.81 1.56 1.63 1.49 1.88 1.18 2.77 1.55 1.78 1.44 2.61 1.39 2.09 2.71 1.39 2.59 1.41 1.

47 1.36 1.32 1.48 1.58 1.43 1.88 1.44 1.41 1.38 1.36 p = number of laboratories.52 1.41 1.72 1.81 1.38 1.38 1.57 1.38 10 1.53 1.41 9 1.34 1.71 1.59 1.90 1.53 1.40 1.82 1.47 1.53 1.52 1.42 1.87 1.71 1.60 1.47 1.37 1.47 1.49 1.36 1.41 1.38 1.93 1.71 1.76 1.67 1.51 1.44 1.94 1.53 1.38 1.43 1.59 1.94 1.44 1.36 1.85 1.32 1.38 1.90 1.46 1.40 1.94 1.40 1.45 1.41 1.36 1.47 1.36 1.38 1.90 1.38 1.66 1.35 1.57 1.47 1.55 1.85 1.59 1.60 1.69 1.59 1.60 1.38 1.43 1.42 1.47 1.47 1.37 1.89 1.38 1.52 1.62 1.36 1.86 1.59 1.40 1.38 1.71 1.44 1.37 1.40 1.37 1.44 1.39 1.68 1.94 1.50 1.71 1.69 1.39 1.36 1.58 1.51 1.53 6 1.78 1.52 1.91 1.42 1. LLC .86 1.90 1.43 1.84 1.56 1.52 1.91 2 1.37 1.71 1.41 1.36 1.60 1.35 1.38 1.48 1.36 1.40 1.45 1.43 1.36 1.38 1.45 1.44 1.46 1.44 1.51 1.34 1.48 1.72 4 1.41 1.29 1.41 1.44 1.43 1.87 1.33 1.52 1.41 1.36 1.35 1.48 1.44 1.89 1.42 1.43 1.80 1.41 1.91 1.94 1.70 1.93 1.36 1.53 1.60 1.93 1.91 1.38 1.70 1.40 1.69 1.31 1.52 1.39 1.94 1.70 1.53 1.36 1.71 1.60 1.36 1.40 1.43 1.35 1.65 1.52 1.40 1.33 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 1.34 1.46 1.37 1.47 1.© 2009 by Taylor & Francis Group.40 1.48 7 1.47 1.36 1.34 1.68 1.37 1.71 1.93 1.92 1.35 1.44 8 1. n = number of results for one level.13b Parameters h and k of Mandel’s Test for Significance Level α = 0.50 1.66 1.30 1.42 1.42 1.35 1.88 1.92 1.40 1.89 1. .53 1.58 1.75 1.93 1.94 1.46 1.35 1.90 1.92 1.59 1.60 1.71 1.36 1.83 1.53 1.48 1.94 1.40 1.38 1.50 1.70 1.38 1.41 1.57 1.94 3 1.36 1.43 1.60 1.53 1.15 1.35 1.71 1.88 1.60 5 1.71 1.44 1.48 1.59 1.48 1.36 1.59 1.54 1.39 1.36 1.46 1.50 1.71 1.38 1.44 1.41 1.60 1.94 1.52 1.91 1.64 1.94 1.90 1.37 1.228 Appendix Table A.

64 0.14 1.80 0.44 Table A.36 1.25 0.22 1.45 0.14 Critical Values (λ α ).89 0.20 α = 0.77 0.25 .22 0.71 0.50 0.54 0. LLC .80 0.97 0.62 0.33 0.02 0.70 0.60 0.90 0.71 0.77 0.10 0.15 0.02 0.49 0.83 0.07 1.67 0.83 0.99 λα 1.52 1.25 0.30 0. Kolmogorov-Smirnov Test α 0.01 0.58 0.20 0.47 0.35 0.05 0.63 0.59 0.57 0.39 0.56 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 = 0.63 1.60 0.28 0.75 0.Appendix 229 Table A.44 0.© 2009 by Taylor & Francis Group.74 0.27 0.40 0.53 0.42 0.05 0.01 0.30 0.38 0.50 0.71 0.66 0.87 0.35 0.

- Desde-La-Cruz.pdf
- 19 Nuevos Evangelizadores para la Nueva Evangelizacion - Jose H Prado Flores.pdf
- 9 Salvar Lo Que Estaba Perdido.pdf
- 15 El Silencio Del Musico (Spanish Edition) - Martin Valverde
- 18 Una Primavera Espiritual para la Música Católica - Marco Lopez.pdf
- 17 Una Nueva Cancion_ El Poder de la Música - Marco Lopez.pdf
- 14 El Ministerio de Musica - P Diego Jaramillo
- La Cara Oculta de La ONU - Michel Chooyans
- Santidad Ahora - P Jose Kentenich
- Eurachem Incertidumbre Español
- Bautismo en El Espiritu Santo
- Dormir permite limpieza del cerebro.docx
- Alimentos-antioxidantes.pdf
- Decodificando el significado de las enfermedades.docx
- Naturaleza y Fin de La RCCE
- ECCLA_2008
- 6 La Alabanza - Ivan Herrera
- SVE - Brasil
- La Interpretación de La Biblia en La Iglesia
- Maria Corredentora - MIRAVALLE
- Virgen Maria -Mensajera Del Amor de Dios
- Por Qué Pierden La Fe Los Jóvenes
- EL BAÑO DE LUZ
- Cómo Fomentar Una Vida Digital Saludable
- Uso o Pierdo Mi Tiempo Libre

- Chapter12_solutions.pdf
- 0645.Trace Environmental Quantitative Analysis. Principles, Techniques and Applications, Second Edition by Paul R. Loconto
- Acid Rain II
- Valid Analytical Methods
- ISEpostlab
- HPLC a Practical Guide
- 1-Introduction to Analytical Chemistry
- 5-Notes on Statistics and Data Quality for Analytical Chemists
- s15 Investigation of Pitting Potencial of Carbon Steel Using Experimental Design Methods
- hplc
- 42.Full
- Principles and Practices of Method Validation. RSC-AOAC-IUPAC
- Erm Cc017repe
- QUÍMICA FÍSICA
- PSEAW139-2013-21888-85312 Criterios de Inspeção de Mangueiras Sistema Hidráulico
- Chromatography
- Guidelines for Validation of Analytical Procedures
- Laboratory
- Managing the Exploitation Life of the Mining Machinery for an Unlimited Duration of Time
- Separation Chemistry
- T. S. Duffy- Strength of Materials Under Static Loading in the Diamond Anvil Cell
- Handbook of Water Analysis
- 345. Surface Tension and Viscosity of Industrial Alloys From Parabolic Flight Experiments - Results of the ThermoLab Project
- Statistics and Chemo Metrics for Analytical Chemistry 6th Ed. - J. Miller, J. Miller (Pearson, 2010) BBS
- Non-Boiling Two Phase Flow Heat Transfer
- Analytical Chemistry Gary D. Christian
- laconductivité
- Applied Mathematical Modelling Volume 36 Issue 9 2012 [Doi 10.1016%2Fj.apm.2011.11.073] S. Narendar; S.S. Gupta; S. Gopalakrishnan -- Wave Propagation in Single-walled Carbon Nanotube Under Longi
- An Optimised Method to Determine the Degree of Acetylation of Chitin and Chitosan by FTIR Spectroscopy
- 1497650X

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd