Calibration, issues in

Abstract Calibration establishes and standardizes common units of measurement for commerce and scientific study. Through calibration the measured item or sample is related to items or samples with known properties, the standards. Statistical methods are typically used to identify and quantify the form of the relationship between sample measurements and one or more standards. They are also used to evaluate the accuracy and precision of the current as well as alternative measurement methods. Statistical tools are also used to determine the optimal frequency of recalibration as well as the frequency with which calibration should be checked. Calibration models can take the form of linear relationships or a nonlinear relationship. Two‐variable calibration models are often monotonic in their form. While many calibration models often involve only two variables, they can be multivariate in nature. Keeping bias out of calibration models requires not only selecting the correct model form but also ensuring that the model is fit, using an appropriate optimization criteria. Calibration models typically provide very high explanatory powers as compared to other typical scientific modeling applications such as plant or process optimization studies. Calibration models also typically yield extremely good levels of statistical significance. Current common statistical toolsets used in calibration studies are often not well structured for handling the actual, and sometimes complex, error structures that occur in routine application of calibration.