Multivariate calibration. I. Concepts and distinctions

Abstract This is the first of two introductory papers on multivariate calibration. We show how precise quantitative chemical analysis is made possible even in 'dirty' sample types like biological tissue, by compensating for systematic interferences in the measured data. This drastically reduces the required sample preparation work, making high-speed, nonspecific instrument measurements possible. Multivariate calibration also allows various types of automatic error detection, improving reliability in chemical analysis. The present paper treats on a conceptual basis the following topics: Univariate vs. multivariate calibration, direct vs. indirect calibration and controlled vs. natural calibration. Some aspects of multivariate quantitative modelling is illustrated, and multiwavelength near infrared spectrometry is given as a practical example. The compromise between necessary complexity vs. danger of statistical overfitting is discussed.