Use of mean square prediction error analysis and reproducibility measures to study near infrared calibration equation performance

Monitoring of calibration equation performance is essential if high quality of predicted analytical data is to be sustained. In this paper we outline and illustrate the use of some statistical methods which are well suited for postprediction data scrutiny. Mean square prediction error is partitioned into three components, viz, mean bias, systematic bias and random error. Reproducibility measures such as concordance correlation (r(c)), intraclass correlation (r(2)) and correlation between difference and sum (r((X- Y)(X + Y))) are also discussed. Other topics discussed include the maximisation of R-2, type II regression (both variables with error model) and new graphical displays.