ISO TR 22971 pdf download – ISO TR 22971 pdf Accuracy (trueness and precision) ofmeasurement methods and results —Practical guidance for the use of lSo 5725-2:1994 in designing, implementing and statistically analysinginterlaboratory repeatability and reproducibility results

ISO TR 22971 pdf download – ISO TR 22971 pdf Accuracy (trueness and precision) ofmeasurement methods and results —Practical guidance for the use of lSo 5725-2:1994 in designing, implementing and statistically analysinginterlaboratory repeatability and reproducibility
results.
3.1.2.3 Mandels plots of It and A tests statistics
3.1.2.3.1 Manders h plot
For a particuLar level of interest, the mean values obtained For all the laboratories are used to calculate a single overall mean value. This value Is men used to calculate Mandels h statistic for all the laboratones for this level, This statistic is defined in ISO 5725-2:1994, EquatIon (6). This statistic is the ratio of the d4ference between the mean for a particular set of data and the mean of aN sets of data, and the standard deviation of the means from all the sets of data. This quotient value is then plotted d compared with computed or tabulated ratio values obtained for 95 % and 99 % confidence levels. The same procedure is then used to calculate Mandels I. statistic for all the laboratones for all the other levels of interest (see Figure 5). It should be noted mat both positive and negative values can be plotted
3.1.2.3.3 Graphical InspectIon
From the plots, Individual results can be Identified for each laboratory that might be considered different from the expected distribution of results, For example, the h plot for particular levels of interest for each laboratory might approach or exceed the computed Mandel’s Ii statistic value at the 95 % or 99% confidence level if the Grubbe’ test shows outliers to be present. In addition. the k plot for particular levels of mterest for each laboratory might approach or exceed the computed Mandel’s k statistic value at the 95 % or 99 % confidence level If Cochran’s test shows outliers to be present
3.2 Tests for outliers
3.2.1 General points
3.2.1.1 Level of confidence
The treatment of outliers is dealt with ii ISO 5725-2:1994, Clause 7, particularly 7.1 to 7.3. An outlier can be considered as a result which Is suffIciently different from all other results to warrant further investigation. Depending on the type of distribution rto which the results fit, a result that appears to be an outlier could, in reality, be a valid result. ISO 5725-2:1994, 7.3.2.1 and 7.3.3.2, recommends confidence levels of 95% for outliers termed ‘stragglers, and 99 % for outliers termed “statistical outliers’. For individual circumstances, the selection 0195 % and 99 % confidence levels means that one result in 20, and one result in 100. respectively. might be erroneously misinterpreted, Hence, this one result could occur by chance and the degree of confidence stated in ISO 5725-2 might not be appropriate for individual needs. This might represent a degree of acceptability that is not sufficient for certain purposes. This would mean that individual circumstances would merit individual consideration as to whether ISO 5725-2, in terms of the confidence levels used, should be applied.
3.2.1.2 Basic assumptions
In the tests used to determine the presence or absence of outliers, It is assumed that the results are distributed in a Gaussian manner (commonly referred to as a normal distribution; ISO 5725-2:1994, 1.4) or at least a single unimodal distribution (ISO 5725-2:1994, 7.3.1.7). Hence, before undertaking any test, especially one that involves a large number of results, a check to confirm this assumption should be made. It is also assumed (ISO 5725-2:1994, 1.3 and 5.1.1) that the number of results within each set of data (from each laboratory) Is the same and that the number of results for each level of Interest, or number of different samples. is the same. Thus, the results are balanced. If the results are not ‘balanced’, then it is recommended (ISO 5725-2:1994, 7.2.2) that results from appropriate sets of data be randomly discarded until a Thalancect’ situation is created. Although a “balanced” situation Is preferred, it is recognized (even within the examples Illustrated In ISO 5725-2) that “unbalanced’ situations can be accommodated, It Is further assumed (ISO 5725-1:1994, 4.4, arid ISO 5725-2:1994, 7.3,3.3) that results are obtained under repeatability conditions Hence, it can be assumed that the samples for a specific level of interest are homogeneous, identical in all respects, and analysed within a short period of time using the same reagents and calibration solutions. In theory, these criteria have to be satisfied before any tests can be used to establish the presence or absence of outliers.
3.2.1,3 Dectaratlon of outliers
When carrying out the outlier tests. it should be understood that outliers should not be discarded or rejected purely from a statistical point of view, For each sample. the reason why the result is different from all the others should be investigated and identified. Outtier tests (based on the assumplions used) indicate whether there sufficient statistical cause for an outlier; It will not indicate why It has occurred. It Is only after thorough Investigations have been undertaken to identify likely causes that data should be declared outliers and discarded.