Χ ± ∪
Which alternative top down MU method will work best for you?
There are several alternative top down methods available for the estimation of measurement uncertainty (MU) in chemical and microbiological laboratories. One or two will work best under your laboratory conditions.
All accredited laboratories would have implemented a robust laboratory quality management system in accordance to the ISO/IEC 17025 standards. In this scenario, you would have carried out certain method validation and verification on your in-house test methods and standard/official methods, respectively. You would also have consistently been running laboratory control samples (LCS’s) as and when you conduct a batch of sample analysis to monitor the accuracy of your test results. Similarly, you should also have been participating some proficiency testing (PT) programs at regular intervals as required by the national accreditation body.
With this in mind, I recommend the following top down approaches which will be suitable for your MU evaluation:
- For established standard test methods (like ISO, AOCS, BS, EN, ASTM, etc.) that you have been running routinely with readily available QC data, the use of data repeatability, reproducibility and trueness estimates will be fine for estimating the MU, such as following the ISO 21748:2010;
- If you have been using stable laboratory control samples (LCS’s) to monitor your test accuracy regularly, you can consider plotting the LCS data on a Shewhart control chart against time and apply the variance of the data moving average as the estimation of its standard uncertainty (see ASTM D6299-08). Certain pre-requisites however, do apply here, such as statistically confirming your QC data are completely random and independent by the Anderson-Darling (AD) or Shapiro-Wilk tests for normality, and visually checking the data trend of the Shewhart chart to follow a set of chart rules laid down by the standards;
- If your laboratory is not able to take part in any PT program because there is no such program to check your test parameters, you can use your within-lab reproducibility (or intermediate precision) data and the result bias estimates which are available in your method validation process. The relevant reference of this approach is ISO 11352:2012;
- For certain types of chemical tests involving a series of certified reference materials as calibration standards (e.g. the determination of total sulphur in petroleum product by X-ray fluorescence spectrometric method – ASTM D4294 – which is a direct read-out from the energy-dispersive X-ray fluorescence sulphur meter with 4-point or 5-point calibration using different sulphur reference standards), the laboratory concerned can plot a linear calibration curve involving the actual test results of these reference standards against the assigned reference values to estimate the uncertainty. See reference ISO 11095:1996(2012).
- The GUM component-by-component method is not suitable for the MU estimation of microbiological count experiment because the distribution of such data is not strictly normal whilst GUM has made this assumption. In fact, the Poisson probability distribution is better in this situation. The MU estimation for microbiological counting therefore is based on the holistic performance of the test method and we need to make logarithmic transformation of the within-lab reproducibility data before evaluating its variances. Various established documents are available for reference, such as ISO 10936:2006, BS 8496:2007, A2LA G108, NMKL Procedure No. 8, etc.
In my opinion, the ISO 11095 method in (d) is the most difficult one for a laboratory analyst who has acquired only elementary statistical skills, because of its application of more complicated statistical tools in estimating the constant and proportional variances. The other approaches are quite straight forward and will be easily appreciated by the analysts with average statistical knowledge.