Examination of Some Misconceptions about Near-Infrared Analysis

Abstract

There are some recurrent misconceptions about near-infrared (near-IR or NIR) analysis and similar chemometric techniques that have resurfaced as this technology has become more popular and widely used. These misconceptions are that (1) near-IR can never perform better than the primary reference technique from which it was calibrated, (2) near-IR models that are linear in absorbance cannot account for nonlinear effects, and (3) bias and skew adjustments are purely mathematical manipulations without physical assumptions. This paper examines how these misconceptions arise and, using real and synthetic spectra and simulated noise, presents counter examples to them. This paper also shows how one can approximate a change of scale using bias and skew adjustments to the model and lists some conditions on a model's coefficient weights that make a model insensitive to small changes in spectrometer response.