Determination of Accuracy and Precision

The levels of accuracy and precision determine the quality of a measurement. The data are as good as random numbers if these parameters are not specified. Accuracy is determined by analyzing samples of known concentration (evaluation samples) and comparing the measured values to the known. Standard reference materials are available from regulatory agencies and commercial vendors. A standard of known concentration may also be made up in the laboratory to serve as an evaluation sample.

Efective use of evaluation samples depends on matching the standards with the real-world samples, especially in terms of their matrix. Take the example of extraction of pesticides from fish liver. In a real sample, the pesticide is embedded in the liver cells (intracellular matter). If the calibration standards are made by spiking livers, it is possible that the pesticides will be absorbed on the outside of the cells (extracellular). The extraction of extracellular pesticides is easier than real-world intracellular extractions.

Consequently, the extraction e‰ciency of the spiked sample may be significantly higher. Using this as the calibration standard may result in a negative bias. So matrix e¤ects and matrix matching are important for obtaining high accuracy. Extraction procedures that are powerful enough not to have any matrix dependency are desirable.

 Precision is measured by making replicate measurements. As mentioned before, it is known to be a function of concentration and should be determined at the concentration level of interest. The intrasample variance can be determined by splitting a sample into several subsamples and carrying out the sample preparation/analysis under identical conditions to obtain a measure of RSD. For example, several aliquots of homogenized fish liver can be processed through the same extraction and analytical procedure, and the RSD computed. The intersample variance can be measured by analyzing several samples from the same source. For example, di¤erent fish from the same pond can be analyzed to estimate the intersample RSD.

The precision of the overall process is often determined by the extraction step rather than the analytical step. It is easier to get high-precision analytical results; it is much more di‰cult to get reproducible extractions. For example, it is possible to run replicate chromatographic runs (GC or HPLC) with an RSD between 1 and 3%. However, several EPA-approved methods accept extraction e‰ciencies anywhere between 70 and 120%. This range alone represents variability as high as 75%. Consequently, in complex analytical methods that involve several preparative steps, the major contributor to variability is the sample preparation.