This publication provides technical information regarding the use, application, and metrology related to liquid handling instrumentation.

I welcome DefiningAccuracy, Precision new words, or old words used & Trueness

in new ways, Language Figure 1. Accurate results are achieved by improving influences both precision and trueness. provided the thinking Improving Accuracy result is more and careful use of words Decreasing Uncertainty when evaluating data allows precision, added us to understand how best to improve laboratory accuracy.

color or greater Accuracy is a particularly important concept because it expressiveness. is foundational to the quality of laboratory measurements. William Safire Accuracy invokes an image of something that is correct, reliable and trustworthy. As a conceptual goal, accuracy Improving Trueness Improving is universally sought after. Errors Systematic Decreasing

However, what exactly do we mean when we speak of accuracy? Try this experiment: The next time Improving Precision Decreasing Random Errors someone uses the word “accuracy”, engage in a conversation and try to understand as clearly as possible Removing Defining 6 key what they mean by the term. ambiguity terms These are some possibilities: Historically, there have been Six terms related to accuracy • Highly consistent results (e.g., two different ways that the are: Precision, Trueness, a low standard deviation, or word “accuracy” is used in the Accuracy, Random Error, small CV). context of liquid handling. Systematic Error and Uncertainty. We can use this word to There is a logical relationship • An average which is very close describe a single liquid delivery, between these six terms, as to the true value. and also how the average value shown in Figure 1 and Table 1. • Knowing that each measurement of a group of dispenses can correctly represents what is be evaluated for accuracy. This Accuracy present in the sample. “double usage” of the term is still is a combination of trueness common today, but recent work and precision. Good accuracy in developing definitions for requires good trueness and ISO IWA 151 (based on the good precision. Accuracy is vocabulary of metrology2) has measured and reported as provided more precise expla- an uncertainty. Thanks to the nations which help to clarify growing popularity of laboratory our thinking, and allow us to requirements standards such be more exact in what we are as ISO 170253, much has been talking about. written about uncertainty and the detailed procedures to how to calculate or estimate uncer- standard deviation (RSD) or tainty. Uncertainty is frequently coefficient of variation (CV). Table 1. Relationships between error concepts and the way presented in a very mathemat- Precision is necessary for achieving they are quantified. ical way, with lots of equations. good accuracy, but it is not suffi- CONCEPT QUANTITATIVE MEASUREMENT The mathematical expressions cient. Good accuracy requires in authoritative documents such both precision and trueness. Accuracy Uncertainty as the guide to the expression of uncertainty in measurement4, Trueness Precision Random Error can create the impression that is labeled on the vertical axis of uncertainty is a mysterious and Figure 1, and like precision is a Trueness Systematic Error difficult concept. It is helpful to concept. The measure of trueness remember that uncertainty is is systematic error. The idea is that simply a quantitative expression a measurement is true when it is that tells us the accuracy of a aimed squarely at the center of the measurement. target. A pipette or automated liquid handler which is true, will Precision deliver on average a result which is a concept meaning that results is close to the center. As shown are tightly clustered. Precision can in the upper left target in Figure 1, be achieved regardless of where it is possible to be true, while also the cluster falls on the target. having poor precision. Trueness and The degree of precision can be precision are independent of one Q: When can a liquid handling device be considered reliable? measured by quantifying the another. Each can be increased or A: liquid handling device is reliable if it maintains accuracy overall effect of all random errors decreased without changing the over a period of time. The reliability can only be established using descriptive statistics such other. by checking the method, using appropriate primary standards as standard deviation, relative and control specimens3.

Q: What is an outlier? A: Outliers are extreme values which are outside of the For the mathematically inclined: expected range of values. They can occur by chance, but also may indicate an experimental error. Outliers should be investigated To determine systematic error (%SE) in an attempt to find a root cause. There are a many different tests to identify outliers5. N N N 111 VVVV−− T VV= i T Q: How many data points should I collect when testing for VV= ∑ VV= N i=1 i i %%SESE == ××100%100% ∑∑ precision and trueness? N i=1 V N i=1 VT T A: The number of data points will vary depending upon your laboratory’s needs, and whether you are testing for trueness, N precision or both. Please see Artel Lab Report, Issue 1, isN the1 average of all measured volumes; 1 How Many Data Points6. VV= ∑ i VV= Ni i=1 ∑is the number of replicate deliveries; N N i=1 References: 1 1. ISO IWA 15:2015, Specification and method for the determination VV= is a single measured volume; of performance of automated liquid handling systems. www.iwa15.org ∑VV−i T %SEN =i=1 ×100% is the target volume, the volume intended to be delivered. 2. JCGM 200:2013 International vocabulary of metrology – Basic and VT general concepts and associated terms (VIM 3rd edition). www.bipm.org/en/publications/guides/

To determine random error (%CV) 3. General requirements for the competence of testing and calibration laboratories. ISO/IEC 17025:2005. N N N 4. JCGM 100:2008 Evaluation of measurement data – Guide to the 1 2 2 VV= ∑ i expression of uncertainty in measurement www.bipm.org/en/ N i=1 ()VV()VVi −− ∑∑ i publications/guides/ 100%100% i=1i=1 %%CVCV == 5. NIST Engineering statistics Handbook, What are outliers in the VNVN−−11 data? www.itl.nist.gov/div898/handbook/prc/section1/prc16.htm 6. Artel Lab Report 1, How Many Data Points: www.artel-usa.com/ resource-library/how-many-data-points/ 19A3229D

25 Bradley Drive, Westbrook, Maine 04092 USA 888-406-3463 | [email protected] | © 2019 Artel®, Inc.