Spectroscopic bilinear least-squares methods exploiting the second-order advantage. Theoretical and experimental study concerning accuracy, sensitivity and prediction error

The singular value decomposition (SVD) calibration method followed by least-squares (LS) prediction is discussed. Four different algorithms were applied to second-order instrumental data, all capable of achieving the second-order advantage. Both numerical simulations and experimental data were employed to assess the accuracy and precision of these algorithms, concerning the determination of two therapeutic drugs in spiked human urine and serum samples by fluorescence excitation–emission matrices. Prediction uncertainties were experimentally estimated by sample replication, and also numerically by Monte Carlo noise-addition. Both experimental and Monte Carlo standard errors were found to be in agreement with values provided by error propagation theory, involving the so-called HCD net analyte signal approach, after Ho, Christian and Davidson.