Diagnosing indirect relationships in multivariate calibration models

Problems concerning covariance among independent variables are well understood and dealt with by inverse regression methods like partial least squares regression. However, between dependent has only received minor attention. Biological samples often complex mixtures of multiple covarying compounds. During multivariate calibration, analyte predictions may be mediated through relationships interfering compounds, which implies that the calibration model is not providing a direct link measurements interest. This compromises robustness validity model—important aspects when applying to future data sets. study discusses issues modeling strong structures exist interest We propose projection-based method diagnose whether indirect dominate model. The proposed tested on two-constituent Beer's law system consisting 20 aqueous amounts fructose (analyte interest) riboflavin (interfering compound). Transmission obtained all in visual near-infrared wavelength ranges. Riboflavin absorption region, whereas exclusively absorbs region. Hence, concentrations, from range only, fully riboflavin, riboflavin.